Upload folder using huggingface_hub

README.md

@@ -1,5 +1,5 @@
 ---
-title: ImageGen - SDXL
+title: ImageGen
 emoji: 🖼
 colorFrom: purple
 colorTo: red
@@ -7,4 +7,5 @@ sdk: gradio
 sdk_version: "5.50.0"
 app_file: app.py
 short_description: Multi-task image generator with dynamic, chainable workflows
+pinned: true
 ---

app.py

@@ -1,7 +1,6 @@
 import spaces
 import os
 import sys
-import requests
 import site
 
 APP_DIR = os.path.dirname(os.path.abspath(__file__))
@@ -45,9 +44,11 @@ def dummy_gpu_for_startup():
     print("--- [GPU Startup] Startup check passed. ---")
     return "Startup check passed."
 
+
 def main():
     from utils.app_utils import print_welcome_message
     from scripts import build_sage_attention
+    from comfy_integration import setup as setup_comfyui
 
     print_welcome_message()
     
@@ -58,7 +59,9 @@ def main():
     except Exception as e:
         print(f"--- [Setup] ❌ SageAttention installation failed: {e}. Continuing with default attention. ---")
 
-
+    print("--- [Setup] Starting ComfyUI initialization ---")
+    setup_comfyui.initialize_comfyui()
+    
     print("--- [Setup] Reloading site-packages to detect newly installed packages... ---")
     try:
         site.main()
@@ -66,48 +69,16 @@ def main():
     except Exception as e:
         print(f"--- [Setup] ⚠️  Warning: Could not fully reload site-packages: {e} ---")
 
-    from comfy_integration import setup as setup_comfyui
-    from utils.app_utils import (
-        build_preprocessor_model_map,
-        build_preprocessor_parameter_map,
-        load_ipadapter_presets
-    )
-    from core import shared_state
-    from core.settings import ALL_MODEL_MAP, ALL_FILE_DOWNLOAD_MAP
-
-    def check_all_model_urls_on_startup():
-        print("--- [Setup] Checking all model URL validity (one-time check) ---")
-        for display_name, model_info in ALL_MODEL_MAP.items():
-            repo_id, filename, _, _ = model_info
-            if not repo_id: continue
-
-            download_info = ALL_FILE_DOWNLOAD_MAP.get(filename, {})
-            repo_file_path = download_info.get('repository_file_path', filename)
-            url = f"https://huggingface.co/{repo_id}/resolve/main/{repo_file_path}"
-            
-            try:
-                response = requests.head(url, timeout=5, allow_redirects=True)
-                if response.status_code >= 400:
-                    print(f"❌ Invalid URL for '{display_name}': {url} (Status: {response.status_code})")
-                    shared_state.INVALID_MODEL_URLS[display_name] = True
-            except requests.RequestException as e:
-                print(f"❌ URL check failed for '{display_name}': {e}")
-                shared_state.INVALID_MODEL_URLS[display_name] = True
-        print("--- [Setup] ✅ Finished checking model URLs. ---")
+    print("--- Initiating GPU Startup Check & SageAttention Patch ---")
+    try:
+        dummy_gpu_for_startup()
+    except Exception as e:
+        print(f"--- [GPU Startup] ⚠️ Warning: Startup check failed: {e} ---")
 
-    print("--- Starting Application Setup ---")
-    
-    setup_comfyui.initialize_comfyui()
+    from utils.app_utils import load_ipadapter_presets
 
-    check_all_model_urls_on_startup()
+    print("--- Starting Application Setup ---")
     
-    print("--- Building ControlNet preprocessor maps ---")
-    from core.generation_logic import build_reverse_map
-    build_reverse_map()
-    build_preprocessor_model_map()
-    build_preprocessor_parameter_map()
-    print("--- ✅ ControlNet preprocessor setup complete. ---")
-
     print("--- Loading IPAdapter presets ---")
     load_ipadapter_presets()
     print("--- ✅ IPAdapter setup complete. ---")

chain_injectors/diffsynth_controlnet_injector.py

@@ -0,0 +1,75 @@
+def inject(assembler, chain_definition, chain_items):
+    if not chain_items:
+        return
+
+    model_sampler_name = chain_definition.get('model_sampler_node')
+    ksampler_name = chain_definition.get('ksampler_node', 'ksampler')
+    
+    target_node_id = None
+    target_input_name = 'model'
+    
+    if model_sampler_name and model_sampler_name in assembler.node_map:
+        model_sampler_id = assembler.node_map[model_sampler_name]
+        if target_input_name in assembler.workflow[model_sampler_id]['inputs']:
+            target_node_id = model_sampler_id
+            print(f"ControlNet Model Patch injector targeting ModelSamplingAuraFlow node '{model_sampler_name}'.")
+
+    if not target_node_id:
+        if ksampler_name in assembler.node_map:
+            ksampler_id = assembler.node_map[ksampler_name]
+            if target_input_name in assembler.workflow[ksampler_id]['inputs']:
+                target_node_id = ksampler_id
+                print(f"ControlNet Model Patch injector targeting KSampler node '{ksampler_name}'.")
+        else:
+            print(f"Warning: Neither ModelSamplingAuraFlow node '{model_sampler_name}' nor KSampler node '{ksampler_name}' found for ControlNet patch chain. Skipping.")
+            return
+            
+    if not target_node_id:
+        print(f"Warning: Could not find a valid 'model' input on target nodes. Skipping ControlNet patch chain.")
+        return
+
+    current_model_connection = assembler.workflow[target_node_id]['inputs'][target_input_name]
+    
+    vae_source_str = chain_definition.get('vae_source')
+    vae_connection = None
+    if vae_source_str:
+        try:
+            vae_node_name, vae_idx_str = vae_source_str.split(':')
+            if vae_node_name in assembler.node_map:
+                vae_connection = [assembler.node_map[vae_node_name], int(vae_idx_str)]
+            else:
+                print(f"Warning: VAE source node '{vae_node_name}' not found for ControlNet patch chain. VAE will not be connected.")
+        except ValueError:
+            print(f"Warning: Invalid 'vae_source' format '{vae_source_str}' for ControlNet patch chain. Expected 'node_name:index'. VAE will not be connected.")
+    else:
+        print(f"Warning: 'vae_source' not defined for ControlNet patch chain definition. VAE may not be connected.")
+
+    for item_data in chain_items:
+        patch_loader_id = assembler._get_unique_id()
+        patch_loader_node = assembler._get_node_template("ModelPatchLoader")
+        patch_loader_node['inputs']['name'] = item_data['control_net_name']
+        assembler.workflow[patch_loader_id] = patch_loader_node
+
+        image_loader_id = assembler._get_unique_id()
+        image_loader_node = assembler._get_node_template("LoadImage")
+        image_loader_node['inputs']['image'] = item_data['image']
+        assembler.workflow[image_loader_id] = image_loader_node
+
+        apply_cn_id = assembler._get_unique_id()
+        apply_cn_node = assembler._get_node_template(chain_definition['template'])
+        
+        apply_cn_node['inputs']['strength'] = item_data.get('strength', 1.0)
+        apply_cn_node['inputs']['model'] = current_model_connection
+        apply_cn_node['inputs']['model_patch'] = [patch_loader_id, 0]
+        apply_cn_node['inputs']['image'] = [image_loader_id, 0]
+        
+        if 'vae' in apply_cn_node['inputs'] and vae_connection:
+            apply_cn_node['inputs']['vae'] = vae_connection
+        
+        assembler.workflow[apply_cn_id] = apply_cn_node
+
+        current_model_connection = [apply_cn_id, 0]
+
+    assembler.workflow[target_node_id]['inputs'][target_input_name] = current_model_connection
+    
+    print(f"ControlNet Model Patch injector applied. Target 'model' input re-routed through {len(chain_items)} patch(es).")

chain_injectors/flux1_ipadapter_injector.py

@@ -0,0 +1,46 @@
+def inject(assembler, chain_definition, chain_items):
+    if not chain_items:
+        return
+
+    ksampler_name = chain_definition.get('ksampler_node', 'ksampler')
+    if ksampler_name not in assembler.node_map:
+        print(f"Warning: KSampler node '{ksampler_name}' not found for Flux1 IPAdapter chain. Skipping.")
+        return
+        
+    ksampler_id = assembler.node_map[ksampler_name]
+
+    if 'model' not in assembler.workflow[ksampler_id]['inputs']:
+        print(f"Warning: KSampler node '{ksampler_name}' is missing 'model' input. Skipping Flux1 IPAdapter chain.")
+        return
+        
+    current_model_connection = assembler.workflow[ksampler_id]['inputs']['model']
+    
+    for item_data in chain_items:
+        image_loader_id = assembler._get_unique_id()
+        image_loader_node = assembler._get_node_template("LoadImage")
+        image_loader_node['inputs']['image'] = item_data['image']
+        assembler.workflow[image_loader_id] = image_loader_node
+
+        ipadapter_loader_id = assembler._get_unique_id()
+        ipadapter_loader_node = assembler._get_node_template("IPAdapterFluxLoader")
+        ipadapter_loader_node['inputs']['ipadapter'] = "ip-adapter.bin"
+        ipadapter_loader_node['inputs']['clip_vision'] = "google/siglip-so400m-patch14-384"
+        ipadapter_loader_node['inputs']['provider'] = "cuda"
+        assembler.workflow[ipadapter_loader_id] = ipadapter_loader_node
+        
+        apply_ipa_id = assembler._get_unique_id()
+        apply_ipa_node = assembler._get_node_template("ApplyIPAdapterFlux")
+        
+        apply_ipa_node['inputs']['weight'] = item_data['weight']
+        apply_ipa_node['inputs']['start_percent'] = item_data.get('start_percent', 0.0)
+        apply_ipa_node['inputs']['end_percent'] = item_data.get('end_percent', 0.6)
+        
+        apply_ipa_node['inputs']['model'] = current_model_connection
+        apply_ipa_node['inputs']['ipadapter_flux'] = [ipadapter_loader_id, 0]
+        apply_ipa_node['inputs']['image'] = [image_loader_id, 0]
+        
+        assembler.workflow[apply_ipa_id] = apply_ipa_node
+        current_model_connection = [apply_ipa_id, 0]
+
+    assembler.workflow[ksampler_id]['inputs']['model'] = current_model_connection
+    print(f"Flux1 IPAdapter injector applied. KSampler model input re-routed through {len(chain_items)} IPAdapter(s).")

chain_injectors/newbie_lora_injector.py

@@ -0,0 +1,63 @@
+from copy import deepcopy
+
+def inject(assembler, chain_definition, chain_items):
+    if not chain_items:
+        return
+
+    output_map = chain_definition.get('output_map', {})
+    current_connections = {}
+    for key, type_name in output_map.items():
+        if ':' in str(key):
+            node_name, idx_str = key.split(':')
+            if node_name not in assembler.node_map:
+                print(f"Warning: [NewBie LoRA Injector] Node '{node_name}' in chain's output_map not found. Skipping.")
+                continue
+            node_id = assembler.node_map[node_name]
+            start_output_idx = int(idx_str)
+            current_connections[type_name] = [node_id, start_output_idx]
+        else:
+            print(f"Warning: [NewBie LoRA Injector] output_map key '{key}' is not in 'node:index' format. Skipping this connection.")
+
+    template_name = chain_definition.get('template')
+    if not template_name:
+        print(f"Warning: [NewBie LoRA Injector] No 'template' defined for chain. Skipping.")
+        return
+
+    for item_data in chain_items:
+        template = assembler._get_node_template(template_name)
+        node_data = deepcopy(template)
+        
+        node_data['inputs']['lora_name'] = item_data.get('lora_name')
+        node_data['inputs']['strength'] = item_data.get('strength_model', 1.0)
+        node_data['inputs']['enabled'] = True
+        
+        if 'model' in current_connections:
+            node_data['inputs']['model'] = current_connections['model']
+        if 'clip' in current_connections:
+            node_data['inputs']['clip'] = current_connections['clip']
+
+        new_node_id = assembler._get_unique_id()
+        assembler.workflow[new_node_id] = node_data
+        
+        current_connections['model'] = [new_node_id, 0]
+        current_connections['clip'] = [new_node_id, 1]
+
+    end_input_map = chain_definition.get('end_input_map', {})
+    for type_name, targets in end_input_map.items():
+        if type_name in current_connections:
+            if not isinstance(targets, list):
+                targets = [targets]
+            
+            for target_str in targets:
+                try:
+                    end_node_name, end_input_name = target_str.split(':')
+                    if end_node_name in assembler.node_map:
+                        end_node_id = assembler.node_map[end_node_name]
+                        assembler.workflow[end_node_id]['inputs'][end_input_name] = current_connections[type_name]
+                    else:
+                        print(f"Warning: [NewBie LoRA Injector] End node '{end_node_name}' for dynamic chain not found. Skipping connection.")
+                except ValueError:
+                    print(f"Warning: [NewBie LoRA Injector] Invalid target format '{target_str}' in end_input_map. Skipping.")
+
+    if chain_items:
+        print(f"NewBie LoRA injector applied. Re-routed model and clip through {len(chain_items)} LoRA(s).")

chain_injectors/reference_latent_injector.py

@@ -0,0 +1,157 @@
+def inject(assembler, chain_definition, chain_items):
+    if not chain_items:
+        return
+
+    guider_node_name = chain_definition.get('guider_node')
+    guider_target_inputs = chain_definition.get('guider_target_inputs', [])
+    start_connections_map = chain_definition.get('start_connections', {})
+    vae_node_name = chain_definition.get('vae_node', 'vae_loader')
+
+    if guider_node_name and guider_node_name in assembler.node_map and guider_target_inputs:
+        guider_id = assembler.node_map[guider_node_name]
+        if vae_node_name not in assembler.node_map:
+            print(f"Warning: VAE node '{vae_node_name}' not found for Guider chain. Skipping.")
+            return
+        vae_node_id = assembler.node_map[vae_node_name]
+        
+        print(f"ReferenceLatent injector targeting DualCFGGuider node '{guider_node_name}'.")
+
+        current_connections = {}
+        for target_input in guider_target_inputs:
+            conn_str = start_connections_map.get(target_input)
+            if not conn_str:
+                print(f"Warning: No start connection defined for '{target_input}' in Guider chain. Skipping this input.")
+                continue
+            try:
+                node_name, idx_str = conn_str.split(':')
+                node_id = assembler.node_map[node_name]
+                current_connections[target_input] = [node_id, int(idx_str)]
+            except (ValueError, KeyError):
+                print(f"Warning: Invalid start connection '{conn_str}' for '{target_input}'. Skipping.")
+
+        encoded_latents = []
+        for i, img_filename in enumerate(chain_items):
+            load_id = assembler._get_unique_id()
+            load_node = assembler._get_node_template("LoadImage")
+            load_node['inputs']['image'] = img_filename
+            assembler.workflow[load_id] = load_node
+            
+            scale_id = assembler._get_unique_id()
+            scale_node = assembler._get_node_template("ImageScaleToTotalPixels")
+            scale_node['inputs']['megapixels'] = 1.0
+            scale_node['inputs']['upscale_method'] = "lanczos"
+            scale_node['inputs']['image'] = [load_id, 0]
+            assembler.workflow[scale_id] = scale_node
+            
+            vae_encode_id = assembler._get_unique_id()
+            vae_encode_node = assembler._get_node_template("VAEEncode")
+            vae_encode_node['inputs']['pixels'] = [scale_id, 0]
+            vae_encode_node['inputs']['vae'] = [vae_node_id, 0]
+            assembler.workflow[vae_encode_id] = vae_encode_node
+            encoded_latents.append([vae_encode_id, 0])
+        
+        for target_input_name, start_connection in current_connections.items():
+            current_chain_head = start_connection
+            for i, latent_conn in enumerate(encoded_latents):
+                ref_latent_id = assembler._get_unique_id()
+                ref_latent_node = assembler._get_node_template("ReferenceLatent")
+                ref_latent_node['inputs']['conditioning'] = current_chain_head
+                ref_latent_node['inputs']['latent'] = latent_conn
+                ref_latent_node['_meta']['title'] = f"{target_input_name} RefLatent {i+1}"
+                assembler.workflow[ref_latent_id] = ref_latent_node
+                current_chain_head = [ref_latent_id, 0]
+            
+            assembler.workflow[guider_id]['inputs'][target_input_name] = current_chain_head
+            print(f"  - Input '{target_input_name}' of node '{guider_node_name}' re-routed through {len(chain_items)} reference images.")
+
+        return
+
+    flux_guidance_name = chain_definition.get('flux_guidance_node')
+    ksampler_name = chain_definition.get('ksampler_node', 'ksampler')
+
+    if ksampler_name not in assembler.node_map:
+        print(f"Warning: KSampler node '{ksampler_name}' not found for ReferenceLatent chain. Skipping.")
+        return
+    if vae_node_name not in assembler.node_map:
+        print(f"Warning: VAE loader node '{vae_node_name}' not found for ReferenceLatent chain. Skipping.")
+        return
+
+    ksampler_id = assembler.node_map[ksampler_name]
+    vae_node_id = assembler.node_map[vae_node_name]
+
+    pos_target_node_id = None
+    pos_target_input_name = None
+    if flux_guidance_name and flux_guidance_name in assembler.node_map:
+        flux_guidance_id = assembler.node_map[flux_guidance_name]
+        if 'conditioning' in assembler.workflow[flux_guidance_id]['inputs']:
+            pos_target_node_id = flux_guidance_id
+            pos_target_input_name = 'conditioning'
+            print(f"ReferenceLatent injector targeting FluxGuidance node '{flux_guidance_name}' for positive chain.")
+    
+    if not pos_target_node_id:
+        if 'positive' in assembler.workflow[ksampler_id]['inputs']:
+            pos_target_node_id = ksampler_id
+            pos_target_input_name = 'positive'
+            print(f"ReferenceLatent injector targeting KSampler node '{ksampler_name}' for positive chain.")
+        else:
+            print(f"Warning: Could not find a valid positive injection point for ReferenceLatent chain. Skipping.")
+            return
+
+    current_pos_conditioning = assembler.workflow[pos_target_node_id]['inputs'][pos_target_input_name]
+
+    neg_target_node_id = ksampler_id
+    neg_target_input_name = 'negative'
+    if 'negative' not in assembler.workflow[neg_target_node_id]['inputs']:
+        print(f"Warning: KSampler node '{ksampler_name}' has no 'negative' input. Skipping negative ReferenceLatent chain.")
+        neg_target_node_id = None
+    
+    current_neg_conditioning = None
+    if neg_target_node_id:
+        current_neg_conditioning = assembler.workflow[neg_target_node_id]['inputs'][neg_target_input_name]
+
+    for i, img_filename in enumerate(chain_items):
+        load_id = assembler._get_unique_id()
+        load_node = assembler._get_node_template("LoadImage")
+        load_node['inputs']['image'] = img_filename
+        load_node['_meta']['title'] = f"Load Reference Image {i+1}"
+        assembler.workflow[load_id] = load_node
+        
+        scale_id = assembler._get_unique_id()
+        scale_node = assembler._get_node_template("ImageScaleToTotalPixels")
+        scale_node['inputs']['megapixels'] = 1.0
+        scale_node['inputs']['upscale_method'] = "lanczos"
+        scale_node['inputs']['image'] = [load_id, 0]
+        scale_node['_meta']['title'] = f"Scale Reference {i+1}"
+        assembler.workflow[scale_id] = scale_node
+        
+        vae_encode_id = assembler._get_unique_id()
+        vae_encode_node = assembler._get_node_template("VAEEncode")
+        vae_encode_node['inputs']['pixels'] = [scale_id, 0]
+        vae_encode_node['inputs']['vae'] = [vae_node_id, 0]
+        vae_encode_node['_meta']['title'] = f"VAE Encode Reference {i+1}"
+        assembler.workflow[vae_encode_id] = vae_encode_node
+        
+        latent_conn = [vae_encode_id, 0]
+
+        pos_ref_latent_id = assembler._get_unique_id()
+        pos_ref_latent_node = assembler._get_node_template("ReferenceLatent")
+        pos_ref_latent_node['inputs']['conditioning'] = current_pos_conditioning
+        pos_ref_latent_node['inputs']['latent'] = latent_conn
+        pos_ref_latent_node['_meta']['title'] = f"Positive ReferenceLatent {i+1}"
+        assembler.workflow[pos_ref_latent_id] = pos_ref_latent_node
+        current_pos_conditioning = [pos_ref_latent_id, 0]
+
+        if neg_target_node_id:
+            neg_ref_latent_id = assembler._get_unique_id()
+            neg_ref_latent_node = assembler._get_node_template("ReferenceLatent")
+            neg_ref_latent_node['inputs']['conditioning'] = current_neg_conditioning
+            neg_ref_latent_node['inputs']['latent'] = latent_conn
+            neg_ref_latent_node['_meta']['title'] = f"Negative ReferenceLatent {i+1}"
+            assembler.workflow[neg_ref_latent_id] = neg_ref_latent_node
+            current_neg_conditioning = [neg_ref_latent_id, 0]
+
+    assembler.workflow[pos_target_node_id]['inputs'][pos_target_input_name] = current_pos_conditioning
+    if neg_target_node_id:
+        assembler.workflow[neg_target_node_id]['inputs'][neg_target_input_name] = current_neg_conditioning
+    
+    print(f"ReferenceLatent injector applied. Re-routed inputs through {len(chain_items)} reference images.")

chain_injectors/sd3_ipadapter_injector.py

@@ -0,0 +1,66 @@
+def inject(assembler, chain_definition, chain_items):
+    if not chain_items:
+        return
+
+    ksampler_name = chain_definition.get('ksampler_node', 'ksampler')
+    if ksampler_name not in assembler.node_map:
+        print(f"Warning: KSampler node '{ksampler_name}' not found for SD3 IPAdapter chain. Skipping.")
+        return
+        
+    ksampler_id = assembler.node_map[ksampler_name]
+
+    if 'model' not in assembler.workflow[ksampler_id]['inputs']:
+        print(f"Warning: KSampler node '{ksampler_name}' is missing 'model' input. Skipping SD3 IPAdapter chain.")
+        return
+        
+    current_model_connection = assembler.workflow[ksampler_id]['inputs']['model']
+    
+    clip_vision_loader_id = assembler._get_unique_id()
+    clip_vision_loader_node = assembler._get_node_template("CLIPVisionLoader")
+    clip_vision_loader_node['inputs']['clip_name'] = "sigclip_vision_patch14_384.safetensors"
+    assembler.workflow[clip_vision_loader_id] = clip_vision_loader_node
+
+    ipadapter_loader_id = assembler._get_unique_id()
+    ipadapter_loader_node = assembler._get_node_template("IPAdapterSD3Loader")
+    ipadapter_loader_node['inputs']['ipadapter'] = "ip-adapter_sd35l_instantx.bin"
+    ipadapter_loader_node['inputs']['provider'] = "cuda"
+    assembler.workflow[ipadapter_loader_id] = ipadapter_loader_node
+
+    for item_data in chain_items:
+        image_loader_id = assembler._get_unique_id()
+        image_loader_node = assembler._get_node_template("LoadImage")
+        image_loader_node['inputs']['image'] = item_data['image']
+        assembler.workflow[image_loader_id] = image_loader_node
+
+        image_scaler_id = assembler._get_unique_id()
+        image_scaler_node = assembler._get_node_template("ImageScaleToTotalPixels")
+        image_scaler_node['inputs']['image'] = [image_loader_id, 0]
+        image_scaler_node['inputs']['upscale_method'] = 'nearest-exact'
+        image_scaler_node['inputs']['megapixels'] = 1.0
+        assembler.workflow[image_scaler_id] = image_scaler_node
+
+        clip_vision_encode_id = assembler._get_unique_id()
+        clip_vision_encode_node = assembler._get_node_template("CLIPVisionEncode")
+        clip_vision_encode_node['inputs']['crop'] = "center"
+        clip_vision_encode_node['inputs']['clip_vision'] = [clip_vision_loader_id, 0]
+        clip_vision_encode_node['inputs']['image'] = [image_scaler_id, 0]
+        assembler.workflow[clip_vision_encode_id] = clip_vision_encode_node
+        
+        apply_ipa_id = assembler._get_unique_id()
+        apply_ipa_node = assembler._get_node_template("ApplyIPAdapterSD3")
+        
+        apply_ipa_node['inputs']['weight'] = item_data.get('weight', 1.0)
+        apply_ipa_node['inputs']['start_percent'] = item_data.get('start_percent', 0.0)
+        apply_ipa_node['inputs']['end_percent'] = item_data.get('end_percent', 1.0)
+        
+        apply_ipa_node['inputs']['model'] = current_model_connection
+        apply_ipa_node['inputs']['ipadapter'] = [ipadapter_loader_id, 0]
+        apply_ipa_node['inputs']['image_embed'] = [clip_vision_encode_id, 0]
+        
+        assembler.workflow[apply_ipa_id] = apply_ipa_node
+        
+        current_model_connection = [apply_ipa_id, 0]
+
+    assembler.workflow[ksampler_id]['inputs']['model'] = current_model_connection
+    
+    print(f"SD3 IPAdapter injector applied. KSampler model input re-routed through {len(chain_items)} IPAdapter(s).")

chain_injectors/style_injector.py

@@ -0,0 +1,71 @@
+def inject(assembler, chain_definition, chain_items):
+    if not chain_items:
+        return
+
+    flux_guidance_name = chain_definition.get('flux_guidance_node')
+    ksampler_name = chain_definition.get('ksampler_node', 'ksampler')
+
+    target_node_id = None
+    target_input_name = None
+    
+    if flux_guidance_name and flux_guidance_name in assembler.node_map:
+        flux_guidance_id = assembler.node_map[flux_guidance_name]
+        if 'conditioning' in assembler.workflow[flux_guidance_id]['inputs']:
+            target_node_id = flux_guidance_id
+            target_input_name = 'conditioning'
+
+    if not target_node_id:
+        if ksampler_name in assembler.node_map:
+            ksampler_id = assembler.node_map[ksampler_name]
+            if 'positive' in assembler.workflow[ksampler_id]['inputs']:
+                target_node_id = ksampler_id
+                target_input_name = 'positive'
+        else:
+            return
+            
+    if not target_node_id:
+        return
+
+    current_conditioning = assembler.workflow[target_node_id]['inputs'][target_input_name]
+
+    style_model_loader_id = assembler._get_unique_id()
+    style_model_loader_node = assembler._get_node_template("StyleModelLoader")
+    style_model_loader_node['inputs']['style_model_name'] = "flux1-redux-dev.safetensors"
+    assembler.workflow[style_model_loader_id] = style_model_loader_node
+
+    clip_vision_loader_id = assembler._get_unique_id()
+    clip_vision_loader_node = assembler._get_node_template("CLIPVisionLoader")
+    clip_vision_loader_node['inputs']['clip_name'] = "sigclip_vision_patch14_384.safetensors"
+    assembler.workflow[clip_vision_loader_id] = clip_vision_loader_node
+
+    for item_data in chain_items:
+        image = item_data.get('image')
+        strength = item_data.get('strength', 1.0)
+        if not image or strength is None:
+            continue
+
+        load_image_id = assembler._get_unique_id()
+        clip_vision_encode_id = assembler._get_unique_id()
+        style_apply_id = assembler._get_unique_id()
+
+        load_image_node = assembler._get_node_template("LoadImage")
+        clip_vision_encode_node = assembler._get_node_template("CLIPVisionEncode")
+        style_apply_node = assembler._get_node_template("StyleModelApply")
+        
+        load_image_node['inputs']['image'] = image
+        clip_vision_encode_node['inputs']['crop'] = "center"
+        clip_vision_encode_node['inputs']['clip_vision'] = [clip_vision_loader_id, 0]
+        clip_vision_encode_node['inputs']['image'] = [load_image_id, 0]
+
+        style_apply_node['inputs']['strength'] = strength
+        style_apply_node['inputs']['strength_type'] = "multiply"
+        style_apply_node['inputs']['conditioning'] = current_conditioning
+        style_apply_node['inputs']['style_model'] = [style_model_loader_id, 0]
+        style_apply_node['inputs']['clip_vision_output'] = [clip_vision_encode_id, 0]
+        
+        assembler.workflow[load_image_id] = load_image_node
+        assembler.workflow[clip_vision_encode_id] = clip_vision_encode_node
+        assembler.workflow[style_apply_id] = style_apply_node
+        current_conditioning = [style_apply_id, 0]
+
+    assembler.workflow[target_node_id]['inputs'][target_input_name] = current_conditioning

comfy_integration/nodes.py

@@ -23,6 +23,11 @@ CLIPTextEncodeSDXL = NODE_CLASS_MAPPINGS['CLIPTextEncodeSDXL']
 LoraLoader = NODE_CLASS_MAPPINGS['LoraLoader']
 CLIPSetLastLayer = NODE_CLASS_MAPPINGS['CLIPSetLastLayer']
 
+if 'EmptyHunyuanImageLatent' in NODE_CLASS_MAPPINGS:
+    EmptyHunyuanImageLatent = NODE_CLASS_MAPPINGS['EmptyHunyuanImageLatent']
+else:
+    print("⚠️ Warning: 'EmptyHunyuanImageLatent' not found in NODE_CLASS_MAPPINGS. HunyuanImage txt2img may fail if this node is required.")
+
 try:
     KSamplerNode = NODE_CLASS_MAPPINGS['KSampler']
     SAMPLER_CHOICES = KSamplerNode.INPUT_TYPES()["required"]["sampler_name"][0]

comfy_integration/setup.py

@@ -41,13 +41,8 @@ def initialize_comfyui():
 
 
     print("--- Cloning third-party extensions for ComfyUI ---")
-    controlnet_aux_path = os.path.join(APP_DIR, "custom_nodes", "comfyui_controlnet_aux")
-    if not os.path.exists(controlnet_aux_path):
-        os.system(f"git clone https://github.com/Fannovel16/comfyui_controlnet_aux.git {controlnet_aux_path}")
-        print("✅ comfyui_controlnet_aux extension cloned.")
-    else:
-        print("✅ comfyui_controlnet_aux extension already exists.")
 
+    # 1. ComfyUI_IPAdapter_plus
     ipadapter_plus_path = os.path.join(APP_DIR, "custom_nodes", "ComfyUI_IPAdapter_plus")
     if not os.path.exists(ipadapter_plus_path):
         os.system(f"git clone https://github.com/cubiq/ComfyUI_IPAdapter_plus.git {ipadapter_plus_path}")
@@ -55,6 +50,30 @@ def initialize_comfyui():
     else:
         print("✅ ComfyUI_IPAdapter_plus extension already exists.")
 
+    # 2. ComfyUI-InstantX-IPAdapter-SD3
+    ipadapter_plus_path = os.path.join(APP_DIR, "custom_nodes", "ComfyUI-InstantX-IPAdapter-SD3")
+    if not os.path.exists(ipadapter_plus_path):
+        os.system(f"git clone https://github.com/Slickytail/ComfyUI-InstantX-IPAdapter-SD3.git {ipadapter_plus_path}")
+        print("✅ ComfyUI-InstantX-IPAdapter-SD3 extension cloned.")
+    else:
+        print("✅ ComfyUI-InstantX-IPAdapter-SD3 extension already exists.")
+
+    # 3. ComfyUI-IPAdapter-Flux
+    ipadapter_flux_path = os.path.join(APP_DIR, "custom_nodes", "ComfyUI-IPAdapter-Flux")
+    if not os.path.exists(ipadapter_flux_path):
+        os.system(f"git clone https://github.com/Shakker-Labs/ComfyUI-IPAdapter-Flux.git {ipadapter_flux_path}")
+        print("✅ ComfyUI-IPAdapter-Flux extension cloned.")
+    else:
+        print("✅ ComfyUI-IPAdapter-Flux extension already exists.")
+
+    # 4. ComfyUI-Newbie-Nodes
+    newbie_nodes_path = os.path.join(APP_DIR, "custom_nodes", "ComfyUI-Newbie-Nodes")
+    if not os.path.exists(newbie_nodes_path):
+        os.system(f"git clone https://github.com/NewBieAI-Lab/ComfyUI-Newbie-Nodes.git {newbie_nodes_path}")
+        print("✅ ComfyUI-Newbie-Nodes extension cloned.")
+    else:
+        print("✅ ComfyUI-Newbie-Nodes extension already exists.")
+
     print(f"✅ Current working directory is: {os.getcwd()}")
     
     import comfy.model_management
@@ -62,12 +81,10 @@ def initialize_comfyui():
 
     print("✅ ComfyUI initialized with default attention mechanism.")
 
-    os.makedirs(os.path.join(APP_DIR, CHECKPOINT_DIR), exist_ok=True)
-    os.makedirs(os.path.join(APP_DIR, LORA_DIR), exist_ok=True)
-    os.makedirs(os.path.join(APP_DIR, EMBEDDING_DIR), exist_ok=True)
-    os.makedirs(os.path.join(APP_DIR, CONTROLNET_DIR), exist_ok=True)
-    os.makedirs(os.path.join(APP_DIR, DIFFUSION_MODELS_DIR), exist_ok=True)
-    os.makedirs(os.path.join(APP_DIR, VAE_DIR), exist_ok=True)
-    os.makedirs(os.path.join(APP_DIR, TEXT_ENCODERS_DIR), exist_ok=True)
+    for dir_path in CATEGORY_TO_DIR_MAP.values():
+        os.makedirs(os.path.join(APP_DIR, dir_path), exist_ok=True)
+    
     os.makedirs(os.path.join(APP_DIR, INPUT_DIR), exist_ok=True)
+    os.makedirs(os.path.join(APP_DIR, OUTPUT_DIR), exist_ok=True)
+    
     print("✅ All required model directories are present.")

core/generation_logic.py

@@ -1,25 +1,10 @@
 from typing import Any, Dict
 import gradio as gr
 
-from core.pipelines.controlnet_preprocessor import ControlNetPreprocessorPipeline
 from core.pipelines.sd_image_pipeline import SdImagePipeline
 
-controlnet_preprocessor_pipeline = ControlNetPreprocessorPipeline()
 sd_image_pipeline = SdImagePipeline()
 
 
-def build_reverse_map():
-    from nodes import NODE_DISPLAY_NAME_MAPPINGS
-    import core.pipelines.controlnet_preprocessor as cn_module
-    
-    if cn_module.REVERSE_DISPLAY_NAME_MAP is None:
-        cn_module.REVERSE_DISPLAY_NAME_MAP = {v: k for k, v in NODE_DISPLAY_NAME_MAPPINGS.items()}
-        if "Semantic Segmentor (legacy, alias for UniFormer)" not in cn_module.REVERSE_DISPLAY_NAME_MAP:
-             cn_module.REVERSE_DISPLAY_NAME_MAP["Semantic Segmentor (legacy, alias for UniFormer)"] = "SemSegPreprocessor"
-
-
-def run_cn_preprocessor_entry(*args, **kwargs):
-    return controlnet_preprocessor_pipeline.run(*args, **kwargs)
-
 def generate_image_wrapper(ui_inputs: dict, progress=gr.Progress(track_tqdm=True)):
     return sd_image_pipeline.run(ui_inputs=ui_inputs, progress=progress)

core/model_manager.py

@@ -1,9 +1,8 @@
 import gc
 from typing import List
 import gradio as gr
-
-from core.settings import ALL_MODEL_MAP
 from utils.app_utils import _ensure_model_downloaded
+from core.settings import ALL_MODEL_MAP
 
 class ModelManager:
     _instance = None
@@ -23,11 +22,11 @@ class ModelManager:
         print(f"--- [ModelManager] Ensuring models are downloaded: {required_models} ---")
         for i, display_name in enumerate(required_models):
             if progress and hasattr(progress, '__call__'):
-                progress(i / len(required_models), desc=f"Checking file: {display_name}")
+                progress(i / max(len(required_models), 1), desc=f"Checking file: {display_name}")
             try:
                 _ensure_model_downloaded(display_name, progress)
             except Exception as e:
                 raise gr.Error(f"Failed to download model '{display_name}'. Reason: {e}")
         print(f"--- [ModelManager] ✅ All required models are present on disk. ---")
-
+    
 model_manager = ModelManager()

core/pipelines/controlnet_preprocessor.py

@@ -1,143 +0,0 @@
-from typing import Dict, Any, List
-import imageio
-import tempfile
-import numpy as np
-import torch
-import gradio as gr
-from PIL import Image
-import spaces
-
-from .base_pipeline import BasePipeline
-from comfy_integration.nodes import NODE_CLASS_MAPPINGS
-from nodes import NODE_DISPLAY_NAME_MAPPINGS
-from utils.app_utils import get_value_at_index
-
-REVERSE_DISPLAY_NAME_MAP = None
-CPU_ONLY_PREPROCESSORS = {
-    "Binary Lines", "Canny Edge", "Color Pallete", "Fake Scribble Lines (aka scribble_hed)",
-    "Image Intensity", "Image Luminance", "Inpaint Preprocessor", "PyraCanny", "Scribble Lines",
-    "Scribble XDoG Lines", "Standard Lineart", "Content Shuffle", "Tile"
-}
-
-def run_node_by_function_name(node_instance: Any, **kwargs) -> Any:
-    node_class = type(node_instance)
-    function_name = getattr(node_class, 'FUNCTION', None)
-    if not function_name:
-        raise AttributeError(f"Node class '{node_class.__name__}' is missing the required 'FUNCTION' attribute.")
-    execution_method = getattr(node_instance, function_name, None)
-    if not callable(execution_method):
-        raise AttributeError(f"Method '{function_name}' not found or not callable on node '{node_class.__name__}'.")
-    return execution_method(**kwargs)
-
-class ControlNetPreprocessorPipeline(BasePipeline):
-    def get_required_models(self, **kwargs) -> List[str]:
-        return []
-
-    def _gpu_logic(
-        self, pil_images: List[Image.Image], preprocessor_name: str, model_name: str,
-        params: Dict[str, Any], progress=gr.Progress(track_tqdm=True)
-    ) -> List[Image.Image]:
-        global REVERSE_DISPLAY_NAME_MAP
-        if REVERSE_DISPLAY_NAME_MAP is None:
-            raise RuntimeError("REVERSE_DISPLAY_NAME_MAP has not been initialized. `build_reverse_map` must be called on startup.")
-            
-        class_name = REVERSE_DISPLAY_NAME_MAP.get(preprocessor_name)
-        if not class_name or class_name not in NODE_CLASS_MAPPINGS:
-            raise ValueError(f"Preprocessor '{preprocessor_name}' not found.")
-            
-        preprocessor_instance = NODE_CLASS_MAPPINGS[class_name]()
-        call_args = {**params, 'ckpt_name': model_name}
-        
-        processed_pil_images = []
-        total_frames = len(pil_images)
-
-        for i, frame_pil in enumerate(pil_images):
-            progress(i / total_frames, desc=f"Processing frame {i+1}/{total_frames} with {preprocessor_name}...")
-            
-            frame_tensor = torch.from_numpy(np.array(frame_pil).astype(np.float32) / 255.0).unsqueeze(0)
-            
-            resolution_arg = {'resolution': max(frame_tensor.shape[2], frame_tensor.shape[3])}
-
-            result_tuple = run_node_by_function_name(
-                preprocessor_instance, 
-                image=frame_tensor, 
-                **resolution_arg,
-                **call_args
-            )
-            
-            processed_tensor = get_value_at_index(result_tuple, 0)
-            processed_np = (processed_tensor.squeeze(0).cpu().numpy().clip(0, 1) * 255.0).astype(np.uint8)
-            processed_pil_images.append(Image.fromarray(processed_np))
-            
-        return processed_pil_images
-
-    def run(self, input_type, image_input, video_input, preprocessor_name, model_name, zero_gpu_duration, *args, progress=gr.Progress(track_tqdm=True)):
-        from utils import app_utils
-        pil_images, is_video, fps = [], False, 30
-        
-        progress(0, desc="Reading input file...")
-        if input_type == "Image":
-            if image_input is None: raise gr.Error("Please provide an input image.")
-            pil_images = [image_input]
-        elif input_type == "Video":
-            if video_input is None: raise gr.Error("Please provide an input video.")
-            try:
-                video_reader = imageio.get_reader(video_input)
-                meta = video_reader.get_meta_data()
-                fps = meta.get('fps', 30)
-                pil_images = [Image.fromarray(frame) for frame in video_reader]
-                is_video = True
-                video_reader.close()
-            except Exception as e: raise gr.Error(f"Failed to read video file: {e}")
-        else:
-            raise gr.Error("Invalid input type selected.")
-
-        if not pil_images: raise gr.Error("Could not extract any frames from the input.")
-        
-        if app_utils.PREPROCESSOR_PARAMETER_MAP is None:
-            raise RuntimeError("Preprocessor parameter map is not built. Check startup logs.")
-            
-        params_config = app_utils.PREPROCESSOR_PARAMETER_MAP.get(preprocessor_name, [])
-        sliders_params = [p for p in params_config if p['type'] in ["INT", "FLOAT"]]
-        dropdown_params = [p for p in params_config if isinstance(p['type'], list)]
-        checkbox_params = [p for p in params_config if p['type'] == "BOOLEAN"]
-        ordered_params_config = sliders_params + dropdown_params + checkbox_params
-        param_names = [p['name'] for p in ordered_params_config]
-        provided_params = {param_names[i]: args[i] for i in range(len(param_names))}
-        
-        if preprocessor_name not in CPU_ONLY_PREPROCESSORS:
-            print(f"--- '{preprocessor_name}' requires GPU, requesting ZeroGPU. ---")
-            try:
-                processed_pil_images = self._execute_gpu_logic(
-                    self._gpu_logic,
-                    duration=zero_gpu_duration,
-                    default_duration=60,
-                    task_name=f"Preprocessor '{preprocessor_name}'",
-                    pil_images=pil_images,
-                    preprocessor_name=preprocessor_name,
-                    model_name=model_name,
-                    params=provided_params,
-                    progress=progress
-                )
-            except Exception as e:
-                import traceback; traceback.print_exc()
-                raise gr.Error(f"Failed to run preprocessor '{preprocessor_name}' on GPU: {e}")
-        else:
-            print(f"--- Running '{preprocessor_name}' on CPU, no ZeroGPU requested. ---")
-            try:
-                processed_pil_images = self._gpu_logic(pil_images, preprocessor_name, model_name, provided_params, progress=progress)
-            except Exception as e:
-                import traceback; traceback.print_exc()
-                raise gr.Error(f"Failed to run preprocessor '{preprocessor_name}' on CPU: {e}")
-
-        if not processed_pil_images: raise gr.Error("Processing returned no frames.")
-        
-        progress(0.9, desc="Finalizing output...")
-        if is_video:
-            frames_np = [np.array(img) for img in processed_pil_images]
-            frames_tensor = torch.from_numpy(np.stack(frames_np)).to(torch.float32) / 255.0
-            video_path = self._encode_video_from_frames(frames_tensor, fps, progress)
-            return [video_path]
-        else:
-            progress(1.0, desc="Done!")
-            return processed_pil_images

core/pipelines/sd_image_pipeline.py

@@ -16,7 +16,15 @@ from core.workflow_assembler import WorkflowAssembler
 
 class SdImagePipeline(BasePipeline):
     def get_required_models(self, model_display_name: str, **kwargs) -> List[str]:
-        return [model_display_name]
+        model_info = ALL_MODEL_MAP.get(model_display_name)
+        if not model_info:
+            return [model_display_name]
+
+        path_or_components = model_info[1]
+        if isinstance(path_or_components, dict):
+            return [v for v in path_or_components.values() if v and v != "pixel_space"]
+        else:
+            return [model_display_name]
 
     def _topological_sort(self, workflow: Dict[str, Any]) -> List[str]:
         graph = defaultdict(list)
@@ -118,7 +126,7 @@ class SdImagePipeline(BasePipeline):
         progress(0.4, desc="Executing workflow...")
         
         initial_objects = {}
-        
+
         decoded_images_tensor = self._execute_workflow(workflow, initial_objects=initial_objects)
         
         output_images = []
@@ -134,6 +142,7 @@ class SdImagePipeline(BasePipeline):
             params_string = f"{ui_inputs['positive_prompt']}\nNegative prompt: {ui_inputs['negative_prompt']}\n"
             params_string += f"Steps: {ui_inputs['num_inference_steps']}, Sampler: {ui_inputs['sampler']}, Scheduler: {ui_inputs['scheduler']}, CFG scale: {ui_inputs['guidance_scale']}, Seed: {current_seed}, Size: {width_for_meta}x{height_for_meta}, Base Model: {model_display_name}"
             if ui_inputs['task_type'] != 'txt2img': params_string += f", Denoise: {ui_inputs['denoise']}"
+            if ui_inputs.get('clip_skip') and ui_inputs['clip_skip'] != 1: params_string += f", Clip skip: {abs(ui_inputs['clip_skip'])}"
             if loras_string: params_string += f", {loras_string}"
             
             pil_image.info = {'parameters': params_string.strip()}
@@ -145,26 +154,34 @@ class SdImagePipeline(BasePipeline):
         progress(0, desc="Preparing models...")
         
         task_type = ui_inputs['task_type']
+        model_display_name = ui_inputs['model_display_name']
+        model_type = MODEL_TYPE_MAP.get(model_display_name, 'sdxl')
+        
+        architectures_dict = ARCHITECTURES_CONFIG.get('architectures', {})
+        workflow_model_type = architectures_dict.get(model_type, {}).get("model_type", "sdxl")
         
         ui_inputs['positive_prompt'] = sanitize_prompt(ui_inputs.get('positive_prompt', ''))
         ui_inputs['negative_prompt'] = sanitize_prompt(ui_inputs.get('negative_prompt', ''))
         
-        required_models = self.get_required_models(model_display_name=ui_inputs['model_display_name'])
-        
+        if 'clip_skip' in ui_inputs and ui_inputs['clip_skip'] is not None:
+             ui_inputs['clip_skip'] = -int(ui_inputs['clip_skip'])
+        else:
+             ui_inputs['clip_skip'] = -1
+
+        required_models = self.get_required_models(model_display_name=model_display_name)
         self.model_manager.ensure_models_downloaded(required_models, progress=progress)
         
         lora_data = ui_inputs.get('lora_data', [])
         active_loras_for_gpu, active_loras_for_meta = [], []
         if lora_data:
             sources, ids, scales, files = lora_data[0::4], lora_data[1::4], lora_data[2::4], lora_data[3::4]
-            
             for i, (source, lora_id, scale, _) in enumerate(zip(sources, ids, scales, files)):
                 if scale > 0 and lora_id and lora_id.strip():
                     lora_filename = None
                     if source == "File":
                         lora_filename = sanitize_filename(lora_id)
                     elif source == "Civitai":
-                        local_path, status = get_lora_path(source, lora_id, ui_inputs['civitai_api_key'], progress)
+                        local_path, status = get_lora_path(source, lora_id, os.environ.get("CIVITAI_API_KEY", ""), progress)
                         if local_path: lora_filename = os.path.basename(local_path)
                         else: raise gr.Error(f"Failed to prepare LoRA {lora_id}: {status}")
                     
@@ -177,7 +194,6 @@ class SdImagePipeline(BasePipeline):
         elif task_type == 'hires_fix': ui_inputs['denoise'] = ui_inputs.get('hires_denoise', 0.55)
         
         temp_files_to_clean = []
-        
         if not os.path.exists(INPUT_DIR): os.makedirs(INPUT_DIR)
 
         if task_type == 'img2img':
@@ -196,7 +212,6 @@ class SdImagePipeline(BasePipeline):
                 raise gr.Error("Inpainting requires an input image and a drawn mask.")
             
             background_img = inpaint_dict['background'].convert("RGBA")
-            
             composite_mask_pil = Image.new('L', background_img.size, 0)
             for layer in inpaint_dict['layers']:
                 if layer:
@@ -210,7 +225,7 @@ class SdImagePipeline(BasePipeline):
             temp_file_path = os.path.join(INPUT_DIR, f"temp_inpaint_composite_{random.randint(1000, 9999)}.png")
             composite_image_with_mask.save(temp_file_path, "PNG")
             
-            ui_inputs['inpaint_image'] = os.path.basename(temp_file_path)
+            ui_inputs['input_image'] = os.path.basename(temp_file_path)
             temp_files_to_clean.append(temp_file_path)
             ui_inputs.pop('inpaint_mask', None)
 
@@ -221,6 +236,9 @@ class SdImagePipeline(BasePipeline):
                 input_image_pil.save(temp_file_path, "PNG")
                 ui_inputs['input_image'] = os.path.basename(temp_file_path)
                 temp_files_to_clean.append(temp_file_path)
+            
+            ui_inputs['megapixels'] = 0.25
+            ui_inputs['grow_mask_by'] = ui_inputs.get('feathering', 10)
 
         elif task_type == 'hires_fix':
             input_image_pil = ui_inputs.get('hires_image')
@@ -240,7 +258,7 @@ class SdImagePipeline(BasePipeline):
                     if source == "File":
                         emb_filename = sanitize_filename(emb_id)
                     elif source == "Civitai":
-                        local_path, status = get_embedding_path(source, emb_id, ui_inputs['civitai_api_key'], progress)
+                        local_path, status = get_embedding_path(source, emb_id, os.environ.get("CIVITAI_API_KEY", ""), progress)
                         if local_path: emb_filename = os.path.basename(local_path)
                         else: raise gr.Error(f"Failed to prepare Embedding {emb_id}: {status}")
                     
@@ -256,19 +274,36 @@ class SdImagePipeline(BasePipeline):
 
         controlnet_data = ui_inputs.get('controlnet_data', [])
         active_controlnets = []
-        (cn_images, _, _, cn_strengths, cn_filepaths) = [controlnet_data[i::5] for i in range(5)]
-        for i in range(len(cn_images)):
-            if cn_images[i] and cn_strengths[i] > 0 and cn_filepaths[i] and cn_filepaths[i] != "None":
-                ensure_controlnet_model_downloaded(cn_filepaths[i], progress)
-
-                if not os.path.exists(INPUT_DIR): os.makedirs(INPUT_DIR)
-                cn_temp_path = os.path.join(INPUT_DIR, f"temp_cn_{i}_{random.randint(1000, 9999)}.png")
-                cn_images[i].save(cn_temp_path, "PNG")
-                temp_files_to_clean.append(cn_temp_path)
-                active_controlnets.append({
-                    "image": os.path.basename(cn_temp_path), "strength": cn_strengths[i],
-                    "start_percent": 0.0, "end_percent": 1.0, "control_net_name": cn_filepaths[i]
-                })
+        if controlnet_data:
+            (cn_images, _, _, cn_strengths, cn_filepaths) = [controlnet_data[i::5] for i in range(5)]
+            for i in range(len(cn_images)):
+                if cn_images[i] and cn_strengths[i] > 0 and cn_filepaths[i] and cn_filepaths[i] != "None":
+                    ensure_controlnet_model_downloaded(cn_filepaths[i], progress)
+                    if not os.path.exists(INPUT_DIR): os.makedirs(INPUT_DIR)
+                    cn_temp_path = os.path.join(INPUT_DIR, f"temp_cn_{i}_{random.randint(1000, 9999)}.png")
+                    cn_images[i].save(cn_temp_path, "PNG")
+                    temp_files_to_clean.append(cn_temp_path)
+                    active_controlnets.append({
+                        "image": os.path.basename(cn_temp_path), "strength": cn_strengths[i],
+                        "start_percent": 0.0, "end_percent": 1.0, "control_net_name": cn_filepaths[i]
+                    })
+                    
+        diffsynth_controlnet_data = ui_inputs.get('diffsynth_controlnet_data', [])
+        active_diffsynth_controlnets = []
+        if diffsynth_controlnet_data:
+            (cn_images, _, _, cn_strengths, cn_filepaths) = [diffsynth_controlnet_data[i::5] for i in range(5)]
+            for i in range(len(cn_images)):
+                if cn_images[i] and cn_strengths[i] > 0 and cn_filepaths[i] and cn_filepaths[i] != "None":
+                    ensure_controlnet_model_downloaded(cn_filepaths[i], progress)
+
+                    if not os.path.exists(INPUT_DIR): os.makedirs(INPUT_DIR)
+                    cn_temp_path = os.path.join(INPUT_DIR, f"temp_diffsynth_cn_{i}_{random.randint(1000, 9999)}.png")
+                    cn_images[i].save(cn_temp_path, "PNG")
+                    temp_files_to_clean.append(cn_temp_path)
+                    active_diffsynth_controlnets.append({
+                        "image": os.path.basename(cn_temp_path), "strength": cn_strengths[i],
+                        "control_net_name": cn_filepaths[i]
+                    })
         
         ipadapter_data = ui_inputs.get('ipadapter_data', [])
         active_ipadapters = []
@@ -276,13 +311,10 @@ class SdImagePipeline(BasePipeline):
             num_ipa_units = (len(ipadapter_data) - 5) // 3
             final_preset, final_weight, final_lora_strength, final_embeds_scaling, final_combine_method = ipadapter_data[-5:]
             ipa_images, ipa_weights, ipa_lora_strengths = [ipadapter_data[i*num_ipa_units:(i+1)*num_ipa_units] for i in range(3)]
-            
             all_presets_to_download = set()
-            
             for i in range(num_ipa_units):
                 if ipa_images[i] and ipa_weights[i] > 0 and final_preset:
                     all_presets_to_download.add(final_preset)
-                    
                     if not os.path.exists(INPUT_DIR): os.makedirs(INPUT_DIR)
                     ipa_temp_path = os.path.join(INPUT_DIR, f"temp_ipa_{i}_{random.randint(1000, 9999)}.png")
                     ipa_images[i].save(ipa_temp_path, "PNG")
@@ -291,34 +323,111 @@ class SdImagePipeline(BasePipeline):
                         "image": os.path.basename(ipa_temp_path), "preset": final_preset,
                         "weight": ipa_weights[i], "lora_strength": ipa_lora_strengths[i]
                     })
-            
             if active_ipadapters and final_preset:
                 all_presets_to_download.add(final_preset)
-            
             for preset in all_presets_to_download:
                 ensure_ipadapter_models_downloaded(preset, progress)
-            
+                
+            model_type_key = 'sd15' if workflow_model_type == 'sd15' else 'sdxl'
             if active_ipadapters:
                 active_ipadapters.append({
-                    'is_final_settings': True, 'model_type': 'sdxl', 'final_preset': final_preset,
+                    'is_final_settings': True, 'model_type': model_type_key, 'final_preset': final_preset,
                     'final_weight': final_weight, 'final_lora_strength': final_lora_strength,
                     'final_embeds_scaling': final_embeds_scaling, 'final_combine_method': final_combine_method
                 })
+                
+        flux1_ipadapter_data = ui_inputs.get('flux1_ipadapter_data', [])
+        active_flux1_ipadapters = []
+        if flux1_ipadapter_data:
+            num_units = len(flux1_ipadapter_data) // 4
+            f_images = flux1_ipadapter_data[0*num_units : 1*num_units]
+            f_weights = flux1_ipadapter_data[1*num_units : 2*num_units]
+            f_starts = flux1_ipadapter_data[2*num_units : 3*num_units]
+            f_ends = flux1_ipadapter_data[3*num_units : 4*num_units]
+            for i in range(len(f_images)):
+                if f_images[i] and f_weights[i] > 0:
+                    from utils.app_utils import _ensure_model_downloaded
+                    for filename in ["ip-adapter.bin"]:
+                        _ensure_model_downloaded(filename, progress)
+                    
+                    from huggingface_hub import snapshot_download
+                    progress(0.5, desc="Caching HF SigLIP model...")
+                    snapshot_download(
+                        repo_id="google/siglip-so400m-patch14-384", 
+                        allow_patterns=["*.json", "*.safetensors", "*.txt"], 
+                        ignore_patterns=["*.msgpack", "*.h5", "*.bin"]
+                    )
+
+                    temp_path = os.path.join(INPUT_DIR, f"temp_fipa_{i}_{random.randint(1000, 9999)}.png")
+                    f_images[i].save(temp_path, "PNG")
+                    temp_files_to_clean.append(temp_path)
+                    active_flux1_ipadapters.append({
+                        "image": os.path.basename(temp_path),
+                        "weight": f_weights[i], "start_percent": f_starts[i], "end_percent": f_ends[i]
+                    })
+                    
+        sd3_ipadapter_data = ui_inputs.get('sd3_ipadapter_chain', [])
+        active_sd3_ipadapters = []
+        if sd3_ipadapter_data:
+            num_units = len(sd3_ipadapter_data) // 4
+            s_images = sd3_ipadapter_data[0*num_units : 1*num_units]
+            s_weights = sd3_ipadapter_data[1*num_units : 2*num_units]
+            s_starts = sd3_ipadapter_data[2*num_units : 3*num_units]
+            s_ends = sd3_ipadapter_data[3*num_units : 4*num_units]
+            sd3_ipa_downloaded = False
+            for i in range(len(s_images)):
+                if s_images[i] and s_weights[i] > 0:
+                    if not sd3_ipa_downloaded:
+                        from utils.app_utils import ensure_sd3_ipadapter_models_downloaded
+                        ensure_sd3_ipadapter_models_downloaded(progress)
+                        sd3_ipa_downloaded = True
+                    temp_path = os.path.join(INPUT_DIR, f"temp_s3ipa_{i}_{random.randint(1000, 9999)}.png")
+                    s_images[i].save(temp_path, "PNG")
+                    temp_files_to_clean.append(temp_path)
+                    active_sd3_ipadapters.append({
+                        "image": os.path.basename(temp_path),
+                        "weight": s_weights[i], "start_percent": s_starts[i], "end_percent": s_ends[i]
+                    })
+                    
+        style_data = ui_inputs.get('style_data', [])
+        active_styles = []
+        if style_data:
+            num_units = len(style_data) // 2
+            st_images = style_data[0*num_units : 1*num_units]
+            st_strengths = style_data[1*num_units : 2*num_units]
+            for i in range(len(st_images)):
+                if st_images[i] and st_strengths[i] > 0:
+                    from utils.app_utils import _ensure_model_downloaded
+                    _ensure_model_downloaded("sigclip_vision_patch14_384.safetensors", progress)
+                    temp_path = os.path.join(INPUT_DIR, f"temp_style_{i}_{random.randint(1000, 9999)}.png")
+                    st_images[i].save(temp_path, "PNG")
+                    temp_files_to_clean.append(temp_path)
+                    active_styles.append({
+                        "image": os.path.basename(temp_path), "strength": st_strengths[i]
+                    })
+        
+        reference_latent_data = ui_inputs.get('reference_latent_data', [])
+        active_reference_latents = []
+        if reference_latent_data:
+            for img in reference_latent_data:
+                if img:
+                    if not os.path.exists(INPUT_DIR): os.makedirs(INPUT_DIR)
+                    temp_path = os.path.join(INPUT_DIR, f"temp_ref_{random.randint(1000, 9999)}.png")
+                    img.save(temp_path, "PNG")
+                    temp_files_to_clean.append(temp_path)
+                    active_reference_latents.append(os.path.basename(temp_path))
         
         from utils.app_utils import get_vae_path
         vae_source = ui_inputs.get('vae_source')
         vae_id = ui_inputs.get('vae_id')
-        vae_file = ui_inputs.get('vae_file')
         vae_name_override = None
-
         if vae_source and vae_source != "None":
             if vae_source == "File":
                 vae_name_override = sanitize_filename(vae_id)
             elif vae_source == "Civitai" and vae_id and vae_id.strip():
-                local_path, status = get_vae_path(vae_source, vae_id, ui_inputs.get('civitai_api_key'), progress)
+                local_path, status = get_vae_path(vae_source, vae_id, os.environ.get("CIVITAI_API_KEY", ""), progress)
                 if local_path: vae_name_override = os.path.basename(local_path)
                 else: raise gr.Error(f"Failed to prepare VAE {vae_id}: {status}")
-        
         if vae_name_override:
             ui_inputs['vae_name'] = vae_name_override
 
@@ -326,22 +435,12 @@ class SdImagePipeline(BasePipeline):
         active_conditioning = []
         if conditioning_data:
             num_units = len(conditioning_data) // 6
-            prompts   = conditioning_data[0*num_units : 1*num_units]
-            widths    = conditioning_data[1*num_units : 2*num_units]
-            heights   = conditioning_data[2*num_units : 3*num_units]
-            xs        = conditioning_data[3*num_units : 4*num_units]
-            ys        = conditioning_data[4*num_units : 5*num_units]
-            strengths = conditioning_data[5*num_units : 6*num_units]
-
+            prompts, widths, heights, xs, ys, strengths = [conditioning_data[i*num_units : (i+1)*num_units] for i in range(6)]
             for i in range(num_units):
                 if prompts[i] and prompts[i].strip():
                     active_conditioning.append({
-                        "prompt": prompts[i],
-                        "width": int(widths[i]),
-                        "height": int(heights[i]),
-                        "x": int(xs[i]),
-                        "y": int(ys[i]),
-                        "strength": float(strengths[i])
+                        "prompt": prompts[i], "width": int(widths[i]), "height": int(heights[i]),
+                        "x": int(xs[i]), "y": int(ys[i]), "strength": float(strengths[i])
                     })
 
         loras_string = f"LoRAs: [{', '.join(active_loras_for_meta)}]" if active_loras_for_meta else ""
@@ -350,31 +449,62 @@ class SdImagePipeline(BasePipeline):
         
         if ui_inputs.get('seed') == -1:
             ui_inputs['seed'] = random.randint(0, 2**32 - 1)
-
-        dynamic_values = {'task_type': ui_inputs['task_type'], 'model_type': "sdxl"}
+        
+        model_info = ALL_MODEL_MAP[model_display_name]
+        path_or_components = model_info[1]
+        latent_type = model_info[3] if len(model_info) > 3 and model_info[3] else 'latent'
+        latent_generator_template = "EmptyLatentImage"
+        if latent_type == 'sd3_latent':
+            latent_generator_template = "EmptySD3LatentImage"
+        elif latent_type == 'chroma_radiance_latent':
+            latent_generator_template = "EmptyChromaRadianceLatentImage"
+        elif latent_type == 'hunyuan_latent':
+            latent_generator_template = "EmptyHunyuanImageLatent"
+
+        dynamic_values = {
+            'task_type': ui_inputs['task_type'], 
+            'model_type': workflow_model_type,
+            'latent_type': latent_type,
+            'latent_generator_template': latent_generator_template
+        }
         
         recipe_path = os.path.join(os.path.dirname(__file__), "workflow_recipes", "sd_unified_recipe.yaml")
         assembler = WorkflowAssembler(recipe_path, dynamic_values=dynamic_values)
         
         workflow_inputs = {
+            **ui_inputs,
             "positive_prompt": ui_inputs['positive_prompt'], "negative_prompt": ui_inputs['negative_prompt'],
             "seed": ui_inputs['seed'], "steps": ui_inputs['num_inference_steps'], "cfg": ui_inputs['guidance_scale'],
             "sampler_name": ui_inputs['sampler'], "scheduler": ui_inputs['scheduler'],
             "batch_size": ui_inputs['batch_size'],
-            "denoise": ui_inputs['denoise'], 
-            "input_image": ui_inputs.get('input_image'),
-            "inpaint_image": ui_inputs.get('inpaint_image'),
-            "inpaint_mask": ui_inputs.get('inpaint_mask'),
-            "left": ui_inputs.get('outpaint_left'), "top": ui_inputs.get('outpaint_top'),
-            "right": ui_inputs.get('outpaint_right'), "bottom": ui_inputs.get('outpaint_bottom'),
-            "hires_upscaler": ui_inputs.get('hires_upscaler'), "hires_scale_by": ui_inputs.get('hires_scale_by'),
-            "model_name": ALL_MODEL_MAP[ui_inputs['model_display_name']][1],
+            "clip_skip": ui_inputs['clip_skip'],
+            "denoise": ui_inputs['denoise'],
             "vae_name": ui_inputs.get('vae_name'),
+            "guidance": ui_inputs.get('guidance', 3.5),
             "lora_chain": active_loras_for_gpu,
             "controlnet_chain": active_controlnets,
+            "diffsynth_controlnet_chain": active_diffsynth_controlnets,
             "ipadapter_chain": active_ipadapters,
+            "flux1_ipadapter_chain": active_flux1_ipadapters,
+            "sd3_ipadapter_chain": active_sd3_ipadapters,
+            "style_chain": active_styles,
             "conditioning_chain": active_conditioning,
+            "reference_latent_chain": active_reference_latents,
         }
+
+        if isinstance(path_or_components, dict):
+            workflow_inputs.update({
+                'unet_name': path_or_components.get('unet'),
+                'vae_name': ui_inputs.get('vae_name') or path_or_components.get('vae'),
+                'clip_name': path_or_components.get('clip'),
+                'clip1_name': path_or_components.get('clip1'),
+                'clip2_name': path_or_components.get('clip2'),
+                'clip3_name': path_or_components.get('clip3'),
+                'clip4_name': path_or_components.get('clip4'),
+                'lora_name': path_or_components.get('lora'),
+            })
+        else:
+            workflow_inputs['model_name'] = path_or_components
         
         if task_type == 'txt2img':
             workflow_inputs['width'] = ui_inputs['width']
@@ -382,24 +512,19 @@ class SdImagePipeline(BasePipeline):
         
         workflow = assembler.assemble(workflow_inputs)
 
-        if workflow_inputs.get("vae_name"):
+        if ui_inputs.get("vae_name") and workflow_model_type not in ['flux1', 'hidream', 'lumina', 'omnigen2', 'chroma1-radiance', 'chroma1', 'hunyuanimage', 'ovis-image', 'longcat-image']:
             print("--- [Workflow Patch] VAE override provided. Adding VAELoader and rewiring connections. ---")
             vae_loader_id = assembler._get_unique_id()
             vae_loader_node = assembler._get_node_template("VAELoader")
-            vae_loader_node['inputs']['vae_name'] = workflow_inputs["vae_name"]
+            vae_loader_node['inputs']['vae_name'] = ui_inputs["vae_name"]
             workflow[vae_loader_id] = vae_loader_node
 
             vae_decode_id = assembler.node_map.get("vae_decode")
             if vae_decode_id and vae_decode_id in workflow:
                 workflow[vae_decode_id]['inputs']['vae'] = [vae_loader_id, 0]
-                print(f"  - Rewired 'vae_decode' (ID: {vae_decode_id}) to use new VAELoader.")
-
             vae_encode_id = assembler.node_map.get("vae_encode")
             if vae_encode_id and vae_encode_id in workflow:
                 workflow[vae_encode_id]['inputs']['vae'] = [vae_loader_id, 0]
-                print(f"  - Rewired 'vae_encode' (ID: {vae_encode_id}) to use new VAELoader.")
-        else:
-            print("--- [Workflow Info] No VAE override. Using VAE from checkpoint. ---")
         
         progress(1.0, desc="All models ready. Requesting GPU for generation...")
 

core/pipelines/workflow_recipes/_partials/{_base_sampler.yaml → _base_sampler_sd.yaml}

@@ -1,11 +1,21 @@
 nodes:
+  pos_prompt:
+    class_type: CLIPTextEncode
+    title: "CLIP Text Encode (Positive)"
+  neg_prompt:
+    class_type: CLIPTextEncode
+    title: "CLIP Text Encode (Negative)"
   ksampler:
     class_type: KSampler
-
+    title: "KSampler"
+    params:
+      denoise: 1.0
   vae_decode:
     class_type: VAEDecode
+    title: "VAE Decode"
   save_image:
     class_type: SaveImage
+    title: "Save Image"
     params: {}
 
 connections:
@@ -15,9 +25,12 @@ connections:
     to: "save_image:images"
 
 ui_map:
+  positive_prompt: "pos_prompt:text"
+  negative_prompt: "neg_prompt:text"
   seed: "ksampler:seed"
   steps: "ksampler:steps"
   cfg: "ksampler:cfg"
   sampler_name: "ksampler:sampler_name"
   scheduler: "ksampler:scheduler"
-  denoise: "ksampler:denoise"
+  denoise: "ksampler:denoise"
+  filename_prefix: "save_image:filename_prefix"

core/pipelines/workflow_recipes/_partials/conditioning/anima.yaml

@@ -0,0 +1,54 @@
+nodes:
+  unet_loader:
+    class_type: UNETLoader
+    title: "Load Diffusion Model"
+    params:
+      weight_dtype: "default"
+  vae_loader:
+    class_type: VAELoader
+    title: "Load VAE"
+  clip_loader:
+    class_type: CLIPLoader
+    title: "Load CLIP"
+    params:
+      type: "stable_diffusion"
+      device: "default"
+
+connections:
+  - from: "unet_loader:0"
+    to: "ksampler:model"
+  - from: "clip_loader:0"
+    to: "pos_prompt:clip"
+  - from: "clip_loader:0"
+    to: "neg_prompt:clip"
+  - from: "vae_loader:0"
+    to: "vae_decode:vae"
+  - from: "vae_loader:0"
+    to: "vae_encode:vae"
+  - from: "pos_prompt:0"
+    to: "ksampler:positive"
+  - from: "neg_prompt:0"
+    to: "ksampler:negative"
+
+dynamic_lora_chains:
+  lora_chain:
+    template: "LoraLoader"
+    output_map:
+      "unet_loader:0": "model"
+      "clip_loader:0": "clip"
+    input_map:
+      "model": "model"
+      "clip": "clip"
+    end_input_map:
+      "model": ["ksampler:model"]
+      "clip": ["pos_prompt:clip", "neg_prompt:clip"]
+
+dynamic_conditioning_chains:
+  conditioning_chain:
+    ksampler_node: "ksampler"
+    clip_source: "clip_loader:0"
+
+ui_map:
+  unet_name: "unet_loader:unet_name"
+  vae_name: "vae_loader:vae_name"
+  clip_name: "clip_loader:clip_name"

core/pipelines/workflow_recipes/_partials/conditioning/chroma1-radiance.yaml

@@ -0,0 +1,59 @@
+nodes:
+  unet_loader:
+    class_type: UNETLoader
+    title: "Load Diffusion Model"
+    params:
+      weight_dtype: "default"
+  vae_loader:
+    class_type: VAELoader
+    title: "Load VAE"
+    params:
+      vae_name: "pixel_space"
+  clip_loader:
+    class_type: CLIPLoader
+    title: "Load CLIP"
+    params:
+      type: "chroma"
+      device: "default"
+  t5_tokenizer:
+    class_type: T5TokenizerOptions
+    title: "T5TokenizerOptions"
+    params:
+      min_padding: 0
+      min_length: 3
+  model_sampler:
+    class_type: ModelSamplingAuraFlow
+    params:
+      shift: 3.0
+
+connections:
+  - from: "unet_loader:0"
+    to: "model_sampler:model"
+  - from: "model_sampler:0"
+    to: "ksampler:model"
+  
+  - from: "clip_loader:0"
+    to: "t5_tokenizer:clip"
+  - from: "t5_tokenizer:0"
+    to: "pos_prompt:clip"
+  - from: "t5_tokenizer:0"
+    to: "neg_prompt:clip"
+    
+  - from: "pos_prompt:0"
+    to: "ksampler:positive"
+  - from: "neg_prompt:0"
+    to: "ksampler:negative"
+  
+  - from: "vae_loader:0"
+    to: "vae_decode:vae"
+  - from: "vae_loader:0"
+    to: "vae_encode:vae"
+
+dynamic_conditioning_chains:
+  conditioning_chain:
+    ksampler_node: "ksampler"
+    clip_source: "t5_tokenizer:0"
+
+ui_map:
+  unet_name: "unet_loader:unet_name"
+  clip_name: "clip_loader:clip_name"

core/pipelines/workflow_recipes/_partials/conditioning/chroma1.yaml

@@ -0,0 +1,61 @@
+nodes:
+  unet_loader:
+    class_type: UNETLoader
+    title: "Load Diffusion Model"
+    params:
+      weight_dtype: "default"
+  vae_loader:
+    class_type: VAELoader
+    title: "Load VAE"
+  clip_loader:
+    class_type: CLIPLoader
+    title: "Load CLIP"
+    params:
+      type: "chroma"
+      device: "default"
+  t5_tokenizer:
+    class_type: T5TokenizerOptions
+    title: "T5TokenizerOptions"
+    params:
+      min_padding: 1
+      min_length: 0
+  fresca:
+    class_type: FreSca
+    title: "FreSca"
+    params:
+      scale_low: 1.0
+      scale_high: 2.5
+      freq_cutoff: 30
+
+connections:
+  - from: "unet_loader:0"
+    to: "fresca:model"
+  - from: "fresca:0"
+    to: "ksampler:model"
+  
+  - from: "clip_loader:0"
+    to: "t5_tokenizer:clip"
+  - from: "t5_tokenizer:0"
+    to: "pos_prompt:clip"
+  - from: "t5_tokenizer:0"
+    to: "neg_prompt:clip"
+    
+  - from: "pos_prompt:0"
+    to: "ksampler:positive"
+  - from: "neg_prompt:0"
+    to: "ksampler:negative"
+  
+  - from: "vae_loader:0"
+    to: "vae_decode:vae"
+  - from: "vae_loader:0"
+    to: "vae_encode:vae"
+
+dynamic_conditioning_chains:
+  conditioning_chain:
+    ksampler_node: "ksampler"
+    clip_source: "t5_tokenizer:0"
+
+ui_map:
+  unet_name: "unet_loader:unet_name"
+  vae_name: "vae_loader:vae_name"
+  clip_name: "clip_loader:clip_name"

core/pipelines/workflow_recipes/_partials/conditioning/ernie-image.yaml

@@ -0,0 +1,54 @@
+nodes:
+  unet_loader:
+    class_type: UNETLoader
+    title: "Load Diffusion Model"
+    params:
+      weight_dtype: "default"
+  clip_loader:
+    class_type: CLIPLoader
+    title: "Load CLIP"
+    params:
+      type: "flux2"
+      device: "default"
+  vae_loader:
+    class_type: VAELoader
+    title: "Load VAE"
+
+connections:
+  - from: "unet_loader:0"
+    to: "ksampler:model"
+  - from: "clip_loader:0"
+    to: "pos_prompt:clip"
+  - from: "clip_loader:0"
+    to: "neg_prompt:clip"
+  - from: "pos_prompt:0"
+    to: "ksampler:positive"
+  - from: "neg_prompt:0"
+    to: "ksampler:negative"
+  - from: "vae_loader:0"
+    to: "vae_decode:vae"
+  - from: "vae_loader:0"
+    to: "vae_encode:vae"
+
+dynamic_lora_chains:
+  lora_chain:
+    template: "LoraLoader"
+    output_map:
+      "unet_loader:0": "model"
+      "clip_loader:0": "clip"
+    input_map:
+      "model": "model"
+      "clip": "clip"
+    end_input_map:
+      "model": ["ksampler:model"]
+      "clip": ["pos_prompt:clip", "neg_prompt:clip"]
+
+dynamic_conditioning_chains:
+  conditioning_chain:
+    ksampler_node: "ksampler"
+    clip_source: "clip_loader:0"
+
+ui_map:
+  unet_name: "unet_loader:unet_name"
+  clip_name: "clip_loader:clip_name"
+  vae_name: "vae_loader:vae_name"

core/pipelines/workflow_recipes/_partials/conditioning/flux1.yaml

@@ -0,0 +1,64 @@
+nodes:
+  unet_loader:
+    class_type: UNETLoader
+    title: "Load FLUX UNET"
+    params:
+      weight_dtype: "default"
+  vae_loader:
+    class_type: VAELoader
+    title: "Load FLUX VAE"
+  clip_loader:
+    class_type: DualCLIPLoader
+    title: "Load FLUX Dual CLIP"
+    params:
+      type: "flux"
+      device: "default"
+  flux_guidance:
+    class_type: FluxGuidance
+    title: "FluxGuidance"
+
+connections:
+  - from: "unet_loader:0"
+    to: "ksampler:model"
+  - from: "clip_loader:0"
+    to: "pos_prompt:clip"
+  - from: "clip_loader:0"
+    to: "neg_prompt:clip"
+  - from: "vae_loader:0"
+    to: "vae_decode:vae"
+  - from: "vae_loader:0"
+    to: "vae_encode:vae"
+  - from: "pos_prompt:0"
+    to: "flux_guidance:conditioning"
+  - from: "flux_guidance:0"
+    to: "ksampler:positive"
+  - from: "neg_prompt:0"
+    to: "ksampler:negative"
+
+dynamic_controlnet_chains:
+  controlnet_chain:
+    template: "ControlNetApplyAdvanced"
+    ksampler_node: "ksampler"
+    vae_source: "vae_loader:0"
+
+dynamic_flux1_ipadapter_chains:
+  flux1_ipadapter_chain:
+    ksampler_node: "ksampler"
+
+dynamic_style_chains:
+  style_chain:
+    flux_guidance_node: "flux_guidance"
+    ksampler_node: "ksampler"
+
+dynamic_conditioning_chains:
+  conditioning_chain:
+    flux_guidance_node: "flux_guidance"
+    ksampler_node: "ksampler"
+    clip_source: "clip_loader:0"
+
+ui_map:
+  unet_name: "unet_loader:unet_name"
+  vae_name: "vae_loader:vae_name"
+  clip1_name: "clip_loader:clip_name1"
+  clip2_name: "clip_loader:clip_name2"
+  guidance: "flux_guidance:guidance"

core/pipelines/workflow_recipes/_partials/conditioning/flux2-kv.yaml

@@ -0,0 +1,104 @@
+nodes:
+  unet_loader:
+    class_type: UNETLoader
+    title: "Load Diffusion Model"
+    params:
+      weight_dtype: "default"
+  clip_loader:
+    class_type: CLIPLoader
+    title: "Load CLIP"
+    params:
+      type: "flux2"
+      device: "default"
+  vae_loader:
+    class_type: VAELoader
+    title: "Load VAE"
+    
+  flux_kv_cache:
+    class_type: FluxKVCache
+    title: "Flux KV Cache"
+  
+  pos_prompt:
+    class_type: CLIPTextEncode
+    title: "CLIP Text Encode (Positive)"
+  neg_prompt:
+    class_type: CLIPTextEncode
+    title: "CLIP Text Encode (Negative)"
+  
+  ksampler:
+    class_type: KSampler
+    title: "KSampler"
+    params:
+      denoise: 1.0
+
+  vae_decode:
+    class_type: VAEDecode
+    title: "VAE Decode"
+    
+  save_image:
+    class_type: SaveImage
+    title: "Save Image"
+
+connections:
+  - from: "unet_loader:0"
+    to: "flux_kv_cache:model"
+  - from: "flux_kv_cache:0"
+    to: "ksampler:model"
+    
+  - from: "clip_loader:0"
+    to: "pos_prompt:clip"
+  - from: "clip_loader:0"
+    to: "neg_prompt:clip"
+    
+  - from: "vae_loader:0"
+    to: "vae_decode:vae"
+  - from: "vae_loader:0"
+    to: "vae_encode:vae"
+  
+  - from: "pos_prompt:0"
+    to: "ksampler:positive"
+  - from: "neg_prompt:0"
+    to: "ksampler:negative"
+    
+  - from: "latent_source:0"
+    to: "ksampler:latent_image"
+    
+  - from: "ksampler:0"
+    to: "vae_decode:samples"
+  - from: "vae_decode:0"
+    to: "save_image:images"
+
+dynamic_lora_chains:
+  lora_chain:
+    template: "LoraLoader"
+    output_map:
+      "unet_loader:0": "model"
+      "clip_loader:0": "clip"
+    input_map:
+      "model": "model"
+      "clip": "clip"
+    end_input_map:
+      "model": ["flux_kv_cache:model"]
+      "clip": ["pos_prompt:clip", "neg_prompt:clip"]
+
+dynamic_reference_latent_chains:
+  reference_latent_chain:
+    ksampler_node: "ksampler"
+    vae_node: "vae_loader"
+
+ui_map:
+  unet_name: "unet_loader:unet_name"
+  clip_name: "clip_loader:clip_name"
+  vae_name: "vae_loader:vae_name"
+  
+  positive_prompt: "pos_prompt:text"
+  negative_prompt: "neg_prompt:text"
+  
+  seed: "ksampler:seed"
+  steps: "ksampler:steps"
+  cfg: "ksampler:cfg"
+  sampler_name: "ksampler:sampler_name"
+  scheduler: "ksampler:scheduler"
+  denoise: "ksampler:denoise"
+  
+  filename_prefix: "save_image:filename_prefix"

core/pipelines/workflow_recipes/_partials/conditioning/flux2.yaml

@@ -0,0 +1,96 @@
+nodes:
+  unet_loader:
+    class_type: UNETLoader
+    title: "Load Diffusion Model"
+    params:
+      weight_dtype: "default"
+  clip_loader:
+    class_type: CLIPLoader
+    title: "Load CLIP"
+    params:
+      type: "flux2"
+      device: "default"
+  vae_loader:
+    class_type: VAELoader
+    title: "Load VAE"
+  
+  pos_prompt:
+    class_type: CLIPTextEncode
+    title: "CLIP Text Encode (Positive)"
+  neg_prompt:
+    class_type: CLIPTextEncode
+    title: "CLIP Text Encode (Negative)"
+  
+  ksampler:
+    class_type: KSampler
+    title: "KSampler"
+    params:
+      denoise: 1.0
+
+  vae_decode:
+    class_type: VAEDecode
+    title: "VAE Decode"
+    
+  save_image:
+    class_type: SaveImage
+    title: "Save Image"
+
+connections:
+  - from: "unet_loader:0"
+    to: "ksampler:model"
+  - from: "clip_loader:0"
+    to: "pos_prompt:clip"
+  - from: "clip_loader:0"
+    to: "neg_prompt:clip"
+  - from: "vae_loader:0"
+    to: "vae_decode:vae"
+  - from: "vae_loader:0"
+    to: "vae_encode:vae"
+  
+  - from: "pos_prompt:0"
+    to: "ksampler:positive"
+  - from: "neg_prompt:0"
+    to: "ksampler:negative"
+    
+  - from: "latent_source:0"
+    to: "ksampler:latent_image"
+    
+  - from: "ksampler:0"
+    to: "vae_decode:samples"
+  - from: "vae_decode:0"
+    to: "save_image:images"
+
+dynamic_lora_chains:
+  lora_chain:
+    template: "LoraLoader"
+    output_map:
+      "unet_loader:0": "model"
+      "clip_loader:0": "clip"
+    input_map:
+      "model": "model"
+      "clip": "clip"
+    end_input_map:
+      "model": ["ksampler:model"]
+      "clip": ["pos_prompt:clip", "neg_prompt:clip"]
+
+dynamic_reference_latent_chains:
+  reference_latent_chain:
+    ksampler_node: "ksampler"
+    vae_node: "vae_loader"
+
+ui_map:
+  unet_name: "unet_loader:unet_name"
+  clip_name: "clip_loader:clip_name"
+  vae_name: "vae_loader:vae_name"
+  
+  positive_prompt: "pos_prompt:text"
+  negative_prompt: "neg_prompt:text"
+  
+  seed: "ksampler:seed"
+  steps: "ksampler:steps"
+  cfg: "ksampler:cfg"
+  sampler_name: "ksampler:sampler_name"
+  scheduler: "ksampler:scheduler"
+  denoise: "ksampler:denoise"
+  
+  filename_prefix: "save_image:filename_prefix"

core/pipelines/workflow_recipes/_partials/conditioning/hidream.yaml

@@ -0,0 +1,53 @@
+nodes:
+  unet_loader:
+    class_type: UNETLoader
+    title: "Load HiDream UNET"
+    params:
+      weight_dtype: "default"
+  vae_loader:
+    class_type: VAELoader
+    title: "Load HiDream VAE"
+  clip_loader:
+    class_type: QuadrupleCLIPLoader
+    title: "Load HiDream Quadruple CLIP"
+  
+  model_sampler:
+    class_type: ModelSamplingSD3
+    title: "ModelSamplingSD3"
+    params:
+      shift: 6.0
+
+connections:
+  - from: "unet_loader:0"
+    to: "model_sampler:model"
+    
+  - from: "model_sampler:0"
+    to: "ksampler:model"
+    
+  - from: "clip_loader:0"
+    to: "pos_prompt:clip"
+  - from: "clip_loader:0"
+    to: "neg_prompt:clip"
+
+  - from: "pos_prompt:0"
+    to: "ksampler:positive"
+  - from: "neg_prompt:0"
+    to: "ksampler:negative"
+    
+  - from: "vae_loader:0"
+    to: "vae_decode:vae"
+  - from: "vae_loader:0"
+    to: "vae_encode:vae"
+
+dynamic_conditioning_chains:
+  conditioning_chain:
+    ksampler_node: "ksampler"
+    clip_source: "clip_loader:0"
+
+ui_map:
+  unet_name: "unet_loader:unet_name"
+  vae_name: "vae_loader:vae_name"
+  clip1_name: "clip_loader:clip_name1"
+  clip2_name: "clip_loader:clip_name2"
+  clip3_name: "clip_loader:clip_name3"
+  clip4_name: "clip_loader:clip_name4"

core/pipelines/workflow_recipes/_partials/conditioning/hunyuanimage.yaml

@@ -0,0 +1,42 @@
+nodes:
+  unet_loader:
+    class_type: UNETLoader
+    title: "Load Hunyuan UNET"
+    params:
+      weight_dtype: "default"
+  vae_loader:
+    class_type: VAELoader
+    title: "Load Hunyuan VAE"
+  clip_loader:
+    class_type: DualCLIPLoader
+    title: "Load Hunyuan Dual CLIP"
+    params:
+      type: "hunyuan_image"
+      device: "default"
+
+connections:
+  - from: "unet_loader:0"
+    to: "ksampler:model"
+  - from: "clip_loader:0"
+    to: "pos_prompt:clip"
+  - from: "clip_loader:0"
+    to: "neg_prompt:clip"
+  - from: "vae_loader:0"
+    to: "vae_decode:vae"
+  - from: "vae_loader:0"
+    to: "vae_encode:vae"
+  - from: "pos_prompt:0"
+    to: "ksampler:positive"
+  - from: "neg_prompt:0"
+    to: "ksampler:negative"
+
+dynamic_conditioning_chains:
+  conditioning_chain:
+    ksampler_node: "ksampler"
+    clip_source: "clip_loader:0"
+
+ui_map:
+  unet_name: "unet_loader:unet_name"
+  vae_name: "vae_loader:vae_name"
+  clip1_name: "clip_loader:clip_name1"
+  clip2_name: "clip_loader:clip_name2"

core/pipelines/workflow_recipes/_partials/conditioning/longcat-image.yaml

@@ -0,0 +1,83 @@
+nodes:
+  unet_loader:
+    class_type: UNETLoader
+    title: "Load Diffusion Model"
+    params:
+      weight_dtype: "default"
+  vae_loader:
+    class_type: VAELoader
+    title: "Load VAE"
+  clip_loader:
+    class_type: CLIPLoader
+    title: "Load CLIP"
+    params:
+      type: "longcat_image"
+      device: "default"
+  
+  cfg_norm:
+    class_type: CFGNorm
+    title: "CFGNorm"
+    params:
+      strength: 1.0
+
+  flux_guidance_pos:
+    class_type: FluxGuidance
+    title: "FluxGuidance (Positive)"
+    params:
+      guidance: 4.0
+
+  flux_guidance_neg:
+    class_type: FluxGuidance
+    title: "FluxGuidance (Negative)"
+    params:
+      guidance: 4.0
+
+connections:
+  - from: "unet_loader:0"
+    to: "cfg_norm:model"
+  - from: "cfg_norm:0"
+    to: "ksampler:model"
+  
+  - from: "clip_loader:0"
+    to: "pos_prompt:clip"
+  - from: "clip_loader:0"
+    to: "neg_prompt:clip"
+    
+  - from: "pos_prompt:0"
+    to: "flux_guidance_pos:conditioning"
+  - from: "neg_prompt:0"
+    to: "flux_guidance_neg:conditioning"
+    
+  - from: "flux_guidance_pos:0"
+    to: "ksampler:positive"
+  - from: "flux_guidance_neg:0"
+    to: "ksampler:negative"
+  
+  - from: "vae_loader:0"
+    to: "vae_decode:vae"
+  - from: "vae_loader:0"
+    to: "vae_encode:vae"
+
+dynamic_lora_chains:
+  lora_chain:
+    template: "LoraLoader"
+    output_map:
+      "unet_loader:0": "model"
+      "clip_loader:0": "clip"
+    input_map:
+      "model": "model"
+      "clip": "clip"
+    end_input_map:
+      "model": ["cfg_norm:model"]
+      "clip": ["pos_prompt:clip", "neg_prompt:clip"]
+
+dynamic_conditioning_chains:
+  conditioning_chain:
+    flux_guidance_node: "flux_guidance_pos"
+    ksampler_node: "ksampler"
+    clip_source: "clip_loader:0"
+
+ui_map:
+  unet_name: "unet_loader:unet_name"
+  vae_name: "vae_loader:vae_name"
+  clip_name: "clip_loader:clip_name"

core/pipelines/workflow_recipes/_partials/conditioning/lumina.yaml

@@ -0,0 +1,51 @@
+nodes:
+  ckpt_loader:
+    class_type: CheckpointLoaderSimple
+    title: "Load Checkpoint"
+  model_sampler:
+    class_type: ModelSamplingAuraFlow
+    title: "ModelSamplingAuraFlow"
+    params:
+      shift: 4.0
+
+connections:
+  - from: "ckpt_loader:0"
+    to: "model_sampler:model"
+  - from: "model_sampler:0"
+    to: "ksampler:model"
+  
+  - from: "ckpt_loader:1"
+    to: "pos_prompt:clip"
+  - from: "ckpt_loader:1"
+    to: "neg_prompt:clip"
+  - from: "pos_prompt:0"
+    to: "ksampler:positive"
+  - from: "neg_prompt:0"
+    to: "ksampler:negative"
+    
+  - from: "ckpt_loader:2"
+    to: "vae_decode:vae"
+  - from: "ckpt_loader:2"
+    to: "vae_encode:vae"
+
+dynamic_lora_chains:
+  lora_chain:
+    template: "LoraLoader"
+    start: "ckpt_loader" 
+    output_map:
+      "0": "model" 
+      "1": "clip"  
+    input_map:
+      "model": "model"
+      "clip": "clip"
+    end_input_map:
+      "model": ["model_sampler:model"]
+      "clip": ["pos_prompt:clip", "neg_prompt:clip"]
+
+dynamic_conditioning_chains:
+  conditioning_chain:
+    ksampler_node: "ksampler"
+    clip_source: "ckpt_loader:1"
+
+ui_map:
+  model_name: "ckpt_loader:ckpt_name"

core/pipelines/workflow_recipes/_partials/conditioning/newbie-image.yaml

@@ -0,0 +1,65 @@
+nodes:
+  unet_loader:
+    class_type: UNETLoader
+    title: "Load Diffusion Model"
+    params:
+      weight_dtype: "default"
+  vae_loader:
+    class_type: VAELoader
+    title: "Load VAE"
+  clip_loader:
+    class_type: DualCLIPLoader
+    title: "Load Dual CLIP"
+    params:
+      type: "newbie"
+      device: "default"
+  model_sampler:
+    class_type: ModelSamplingAuraFlow
+    title: "ModelSamplingAuraFlow"
+    params:
+      shift: 6
+
+connections:
+  - from: "unet_loader:0"
+    to: "model_sampler:model"
+  - from: "model_sampler:0"
+    to: "ksampler:model"
+  
+  - from: "clip_loader:0"
+    to: "pos_prompt:clip"
+  - from: "clip_loader:0"
+    to: "neg_prompt:clip"
+    
+  - from: "pos_prompt:0"
+    to: "ksampler:positive"
+  - from: "neg_prompt:0"
+    to: "ksampler:negative"
+  
+  - from: "vae_loader:0"
+    to: "vae_decode:vae"
+  - from: "vae_loader:0"
+    to: "vae_encode:vae"
+
+dynamic_newbie_lora_chains:
+  lora_chain:
+    template: "NewBieLoraLoader"
+    output_map:
+      "unet_loader:0": "model"
+      "clip_loader:0": "clip"
+    input_map:
+      "model": "model"
+      "clip": "clip"
+    end_input_map:
+      "model": ["model_sampler:model"]
+      "clip": ["pos_prompt:clip", "neg_prompt:clip"]
+
+dynamic_conditioning_chains:
+  conditioning_chain:
+    ksampler_node: "ksampler"
+    clip_source: "clip_loader:0"
+
+ui_map:
+  unet_name: "unet_loader:unet_name"
+  vae_name: "vae_loader:vae_name"
+  clip1_name: "clip_loader:clip_name1"
+  clip2_name: "clip_loader:clip_name2"

core/pipelines/workflow_recipes/_partials/conditioning/omnigen2.yaml

@@ -0,0 +1,59 @@
+nodes:
+  unet_loader:
+    class_type: UNETLoader
+    title: "Load Diffusion Model"
+    params:
+      weight_dtype: "default"
+  vae_loader:
+    class_type: VAELoader
+    title: "Load VAE"
+  clip_loader:
+    class_type: CLIPLoader
+    title: "Load CLIP"
+    params:
+      type: "omnigen2"
+      device: "default"
+
+connections:
+  - from: "unet_loader:0"
+    to: "ksampler:model"
+  - from: "clip_loader:0"
+    to: "pos_prompt:clip"
+  - from: "clip_loader:0"
+    to: "neg_prompt:clip"
+  - from: "pos_prompt:0"
+    to: "ksampler:positive"
+  - from: "neg_prompt:0"
+    to: "ksampler:negative"
+  - from: "vae_loader:0"
+    to: "vae_decode:vae"
+  - from: "vae_loader:0"
+    to: "vae_encode:vae"
+
+dynamic_lora_chains:
+  lora_chain:
+    template: "LoraLoader"
+    output_map:
+      "unet_loader:0": "model"
+      "clip_loader:0": "clip"
+    input_map:
+      "model": "model"
+      "clip": "clip"
+    end_input_map:
+      "model": ["ksampler:model"]
+      "clip": ["pos_prompt:clip", "neg_prompt:clip"]
+
+dynamic_conditioning_chains:
+  conditioning_chain:
+    ksampler_node: "ksampler"
+    clip_source: "clip_loader:0"
+
+dynamic_reference_latent_chains:
+  reference_latent_chain:
+    ksampler_node: "ksampler"
+    vae_node: "vae_loader"
+
+ui_map:
+  unet_name: "unet_loader:unet_name"
+  vae_name: "vae_loader:vae_name"
+  clip_name: "clip_loader:clip_name"

core/pipelines/workflow_recipes/_partials/conditioning/ovis-image.yaml

@@ -0,0 +1,50 @@
+nodes:
+  unet_loader:
+    class_type: UNETLoader
+    title: "Load Diffusion Model"
+    params:
+      weight_dtype: "default"
+  vae_loader:
+    class_type: VAELoader
+    title: "Load VAE"
+  clip_loader:
+    class_type: CLIPLoader
+    title: "Load CLIP"
+    params:
+      type: "ovis"
+      device: "default"
+  model_sampler:
+    class_type: ModelSamplingAuraFlow
+    params:
+      shift: 3.0
+
+connections:
+  - from: "unet_loader:0"
+    to: "model_sampler:model"
+  - from: "model_sampler:0"
+    to: "ksampler:model"
+  
+  - from: "clip_loader:0"
+    to: "pos_prompt:clip"
+  - from: "clip_loader:0"
+    to: "neg_prompt:clip"
+    
+  - from: "pos_prompt:0"
+    to: "ksampler:positive"
+  - from: "neg_prompt:0"
+    to: "ksampler:negative"
+  
+  - from: "vae_loader:0"
+    to: "vae_decode:vae"
+  - from: "vae_loader:0"
+    to: "vae_encode:vae"
+
+dynamic_conditioning_chains:
+  conditioning_chain:
+    ksampler_node: "ksampler"
+    clip_source: "clip_loader:0"
+
+ui_map:
+  unet_name: "unet_loader:unet_name"
+  vae_name: "vae_loader:vae_name"
+  clip_name: "clip_loader:clip_name"

core/pipelines/workflow_recipes/_partials/conditioning/qwen-image.yaml

@@ -0,0 +1,80 @@
+nodes:
+  unet_loader:
+    class_type: UNETLoader
+    title: "Load Qwen UNET"
+    params:
+      weight_dtype: "default"
+  vae_loader:
+    class_type: VAELoader
+    title: "Load Qwen VAE"
+  clip_loader:
+    class_type: CLIPLoader
+    title: "Load Qwen CLIP"
+    params:
+      type: "qwen_image"
+      device: "default"
+  
+  lora_loader:
+    class_type: LoraLoaderModelOnly
+    title: "Load Qwen Lightning LoRA"
+    params:
+      strength_model: 1.0
+  model_sampler:
+    class_type: ModelSamplingAuraFlow
+    title: "ModelSamplingAuraFlow"
+    params:
+      shift: 3.1
+
+connections:
+  - from: "unet_loader:0"
+    to: "lora_loader:model"
+  - from: "lora_loader:0"
+    to: "model_sampler:model"
+
+  - from: "model_sampler:0"
+    to: "ksampler:model"
+    
+  - from: "clip_loader:0"
+    to: "pos_prompt:clip"
+  - from: "clip_loader:0"
+    to: "neg_prompt:clip"
+    
+  - from: "vae_loader:0"
+    to: "vae_decode:vae"
+  - from: "vae_loader:0"
+    to: "vae_encode:vae"
+
+  - from: "pos_prompt:0"
+    to: "ksampler:positive"
+  - from: "neg_prompt:0"
+    to: "ksampler:negative"
+
+dynamic_lora_chains:
+  lora_chain:
+    template: "LoraLoader"
+    output_map:
+      "lora_loader:0": "model"
+      "clip_loader:0": "clip"
+    input_map:
+      "model": "model"
+      "clip": "clip"
+    end_input_map:
+      "model": ["model_sampler:model"]
+      "clip": ["pos_prompt:clip", "neg_prompt:clip"]
+
+dynamic_controlnet_chains:
+  controlnet_chain:
+    template: "ControlNetApplyAdvanced"
+    ksampler_node: "ksampler"
+    vae_source: "vae_loader:0"
+
+dynamic_conditioning_chains:
+  conditioning_chain:
+    ksampler_node: "ksampler"
+    clip_source: "clip_loader:0"
+
+ui_map:
+  unet_name: "unet_loader:unet_name"
+  vae_name: "vae_loader:vae_name"
+  clip_name: "clip_loader:clip_name"
+  lora_name: "lora_loader:lora_name"

core/pipelines/workflow_recipes/_partials/conditioning/sd15.yaml

@@ -0,0 +1,63 @@
+nodes:
+  ckpt_loader:
+    class_type: CheckpointLoaderSimple
+    title: "Load Checkpoint"
+  clip_set_last_layer:
+    class_type: CLIPSetLastLayer
+    title: "CLIP Set Last Layer"
+
+connections:
+  - from: "ckpt_loader:0"
+    to: "ksampler:model"
+  - from: "ckpt_loader:1"
+    to: "clip_set_last_layer:clip"
+  - from: "clip_set_last_layer:0"
+    to: "pos_prompt:clip"
+  - from: "clip_set_last_layer:0"
+    to: "neg_prompt:clip"
+  - from: "pos_prompt:0"
+    to: "ksampler:positive"
+  - from: "neg_prompt:0"
+    to: "ksampler:negative"
+  - from: "ckpt_loader:2"
+    to: "vae_decode:vae"
+  - from: "ckpt_loader:2"
+    to: "vae_encode:vae"
+
+dynamic_lora_chains:
+  lora_chain:
+    template: "LoraLoader"
+    start: "clip_set_last_layer"
+    output_map:
+      "ckpt_loader:0": "model" 
+      "0": "clip"             
+    input_map:
+      "model": "model"
+      "clip": "clip"
+    end_input_map:
+      "model": ["ksampler:model"]
+      "clip": ["pos_prompt:clip", "neg_prompt:clip"]
+
+dynamic_controlnet_chains:
+  controlnet_chain:
+    template: "ControlNetApplyAdvanced"
+    ksampler_node: "ksampler"
+    vae_source: "ckpt_loader:2"
+
+dynamic_ipadapter_chains:
+  ipadapter_chain:
+    end: "ksampler"
+    final_preset: "{{ ipadapter_final_preset }}"
+    final_weight: "{{ ipadapter_final_weight }}"
+    final_embeds_scaling: "{{ ipadapter_embeds_scaling }}"
+    final_loader_type: "{{ ipadapter_final_loader_type }}"
+    final_lora_strength: "{{ ipadapter_final_lora_strength }}"
+
+dynamic_conditioning_chains:
+  conditioning_chain:
+    ksampler_node: "ksampler"
+    clip_source: "clip_set_last_layer:0"
+
+ui_map:
+  model_name: "ckpt_loader:ckpt_name"
+  clip_skip: "clip_set_last_layer:stop_at_clip_layer"

core/pipelines/workflow_recipes/_partials/conditioning/sd35.yaml

@@ -0,0 +1,52 @@
+nodes:
+  ckpt_loader:
+    class_type: CheckpointLoaderSimple
+    title: "Load Checkpoint"
+
+connections:
+  - from: "ckpt_loader:0"
+    to: "ksampler:model"
+  - from: "ckpt_loader:1"
+    to: "pos_prompt:clip"
+  - from: "ckpt_loader:1"
+    to: "neg_prompt:clip"
+  - from: "pos_prompt:0"
+    to: "ksampler:positive"
+  - from: "neg_prompt:0"
+    to: "ksampler:negative"
+  - from: "ckpt_loader:2"
+    to: "vae_decode:vae"
+  - from: "ckpt_loader:2"
+    to: "vae_encode:vae"
+
+dynamic_lora_chains:
+  lora_chain:
+    template: "LoraLoader"
+    start: "ckpt_loader" 
+    output_map:
+      "0": "model" 
+      "1": "clip"  
+    input_map:
+      "model": "model"
+      "clip": "clip"
+    end_input_map:
+      "model": ["ksampler:model"]
+      "clip": ["pos_prompt:clip", "neg_prompt:clip"]
+
+dynamic_controlnet_chains:
+  controlnet_chain:
+    template: "ControlNetApplyAdvanced"
+    ksampler_node: "ksampler"
+    vae_source: "ckpt_loader:2"
+
+dynamic_sd3_ipadapter_chains:
+  sd3_ipadapter_chain:
+    ksampler_node: "ksampler"
+
+dynamic_conditioning_chains:
+  conditioning_chain:
+    ksampler_node: "ksampler"
+    clip_source: "ckpt_loader:1"
+
+ui_map:
+  model_name: "ckpt_loader:ckpt_name"

core/pipelines/workflow_recipes/_partials/conditioning/sdxl.yaml

@@ -1,15 +1,7 @@
 nodes:
   ckpt_loader:
     class_type: CheckpointLoaderSimple
-    title: "Load SDXL Checkpoint"
-  
-  pos_prompt:
-    class_type: CLIPTextEncode
-    title: "Positive Prompt Encoder"
-  
-  neg_prompt:
-    class_type: CLIPTextEncode
-    title: "Negative Prompt Encoder"
+    title: "Load Checkpoint"
 
 connections:
   - from: "ckpt_loader:0"
@@ -18,26 +10,22 @@ connections:
     to: "pos_prompt:clip"
   - from: "ckpt_loader:1"
     to: "neg_prompt:clip"
-  
-  - from: "ckpt_loader:2"
-    to: "vae_decode:vae"
-  
   - from: "pos_prompt:0"
     to: "ksampler:positive"
   - from: "neg_prompt:0"
     to: "ksampler:negative"
-    
-ui_map:
-  model_name: "ckpt_loader:ckpt_name"
-  positive_prompt: "pos_prompt:text"
-  negative_prompt: "neg_prompt:text"
-  
+  - from: "ckpt_loader:2"
+    to: "vae_decode:vae"
+  - from: "ckpt_loader:2"
+    to: "vae_encode:vae"
+
 dynamic_lora_chains:
   lora_chain:
     template: "LoraLoader"
+    start: "ckpt_loader" 
     output_map:
-      "ckpt_loader:0": "model"
-      "ckpt_loader:1": "clip"
+      "0": "model" 
+      "1": "clip"  
     input_map:
       "model": "model"
       "clip": "clip"
@@ -57,8 +45,13 @@ dynamic_ipadapter_chains:
     final_preset: "{{ ipadapter_final_preset }}"
     final_weight: "{{ ipadapter_final_weight }}"
     final_embeds_scaling: "{{ ipadapter_embeds_scaling }}"
+    final_loader_type: "{{ ipadapter_final_loader_type }}"
+    final_lora_strength: "{{ ipadapter_final_lora_strength }}"
 
 dynamic_conditioning_chains:
   conditioning_chain:
     ksampler_node: "ksampler"
-    clip_source: "ckpt_loader:1"
+    clip_source: "ckpt_loader:1"
+
+ui_map:
+  model_name: "ckpt_loader:ckpt_name"

core/pipelines/workflow_recipes/_partials/conditioning/z-image.yaml

@@ -0,0 +1,65 @@
+nodes:
+  unet_loader:
+    class_type: UNETLoader
+    title: "Load Diffusion Model"
+    params:
+      weight_dtype: "default"
+  vae_loader:
+    class_type: VAELoader
+    title: "Load VAE"
+  clip_loader:
+    class_type: CLIPLoader
+    title: "Load CLIP"
+    params:
+      type: "lumina2"
+      device: "default"
+  model_sampler:
+    class_type: ModelSamplingAuraFlow
+    params:
+      shift: 3.0
+
+connections:
+  - from: "unet_loader:0"
+    to: "model_sampler:model"
+  - from: "model_sampler:0"
+    to: "ksampler:model"
+  
+  - from: "clip_loader:0"
+    to: "pos_prompt:clip"
+  - from: "clip_loader:0"
+    to: "neg_prompt:clip"
+    
+  - from: "pos_prompt:0"
+    to: "ksampler:positive"
+  - from: "neg_prompt:0"
+    to: "ksampler:negative"
+  
+  - from: "vae_loader:0"
+    to: "vae_decode:vae"
+  - from: "vae_loader:0"
+    to: "vae_encode:vae"
+
+dynamic_lora_chains:
+  lora_chain:
+    template: "LoraLoader"
+    output_map:
+      "unet_loader:0": "model"
+      "clip_loader:0": "clip"
+    input_map:
+      "model": "model"
+      "clip": "clip"
+    end_input_map:
+      "model": ["model_sampler:model"]
+      "clip": ["pos_prompt:clip", "neg_prompt:clip"]
+
+dynamic_diffsynth_controlnet_chains:
+  diffsynth_controlnet_chain:
+    template: "QwenImageDiffsynthControlnet"
+    model_sampler_node: "model_sampler"
+    ksampler_node: "ksampler"
+    vae_source: "vae_loader:0"
+
+ui_map:
+  unet_name: "unet_loader:unet_name"
+  vae_name: "vae_loader:vae_name"
+  clip_name: "clip_loader:clip_name"

core/pipelines/workflow_recipes/_partials/input/hires_fix.yaml

@@ -1,15 +1,16 @@
 nodes:
   input_image_loader:
     class_type: LoadImage
-    
+    title: "Load Input Image"
   vae_encode:
     class_type: VAEEncode
-    
+    title: "VAE Encode (Hires Pre-step)"
   latent_upscaler:
     class_type: LatentUpscaleBy
-
+    title: "Upscale Latent By"
   latent_source:
     class_type: RepeatLatentBatch
+    title: "Repeat Latent Batch for Hires"
 
 connections:
   - from: "input_image_loader:0"

core/pipelines/workflow_recipes/_partials/input/img2img.yaml

@@ -1,12 +1,13 @@
 nodes:
   input_image_loader:
     class_type: LoadImage
-  
+    title: "Load Input Image"
   vae_encode:
     class_type: VAEEncode
-
+    title: "VAE Encode (Img2Img)"
   latent_source:
     class_type: RepeatLatentBatch
+    title: "Repeat Latent Batch"
 
 connections:
   - from: "input_image_loader:0"

core/pipelines/workflow_recipes/_partials/input/inpaint.yaml

@@ -2,24 +2,22 @@ nodes:
   inpaint_loader:
     class_type: LoadImage
     title: "Load Inpaint Image+Mask"
-      
   vae_encode:
     class_type: VAEEncodeForInpaint
-    params:
-      grow_mask_by: 6
-      
+    title: "VAE Encode (for Inpainting)"
   latent_source:
     class_type: RepeatLatentBatch
-      
+    title: "Repeat Latent Batch"
+
 connections:
   - from: "inpaint_loader:0"
     to: "vae_encode:pixels"
   - from: "inpaint_loader:1"
     to: "vae_encode:mask"
-    
   - from: "vae_encode:0"
     to: "latent_source:samples"
 
 ui_map:
-  inpaint_image: "inpaint_loader:image"
-  batch_size: "latent_source:amount"
+  input_image: "inpaint_loader:image"
+  batch_size: "latent_source:amount"
+  grow_mask_by: "vae_encode:grow_mask_by"

core/pipelines/workflow_recipes/_partials/input/outpaint.yaml

@@ -1,38 +1,41 @@
 nodes:
   input_image_loader:
     class_type: LoadImage
-    
+    title: "Load Image for Outpaint"
+  scale_image:
+    class_type: ImageScaleToTotalPixels
+    title: "Scale Image to Total Pixels"
+    params:
+      upscale_method: "nearest-exact"
   pad_image:
     class_type: ImagePadForOutpaint
-    params:
-      feathering: 10
-      
+    title: "Pad Image for Outpainting"
   vae_encode:
     class_type: VAEEncodeForInpaint
-    params:
-      grow_mask_by: 6
-
+    title: "VAE Encode (for Inpainting)"
   latent_source:
     class_type: RepeatLatentBatch
+    title: "Repeat Latent Batch"
 
 connections:
   - from: "input_image_loader:0"
+    to: "scale_image:image"
+  - from: "scale_image:0"
     to: "pad_image:image"
-    
   - from: "pad_image:0"
     to: "vae_encode:pixels"
   - from: "pad_image:1"
     to: "vae_encode:mask"
-
   - from: "vae_encode:0"
     to: "latent_source:samples"
 
 ui_map:
   input_image: "input_image_loader:image"
-  
+  megapixels: "scale_image:megapixels"
   left: "pad_image:left"
   top: "pad_image:top"
   right: "pad_image:right"
   bottom: "pad_image:bottom"
-  
+  feathering: "pad_image:feathering"
+  grow_mask_by: "vae_encode:grow_mask_by"
   batch_size: "latent_source:amount"

core/pipelines/workflow_recipes/_partials/input/txt2img.yaml

@@ -1,8 +1,2 @@
-nodes:
-  latent_source:
-    class_type: EmptyLatentImage
-
-ui_map:
-  width: "latent_source:width"
-  height: "latent_source:height"
-  batch_size: "latent_source:batch_size"
+imports:
+  - "txt2img_{{ latent_type }}.yaml"

core/pipelines/workflow_recipes/_partials/input/txt2img_chroma_radiance_latent.yaml

@@ -0,0 +1,11 @@
+nodes:
+  latent_source:
+    class_type: "EmptyChromaRadianceLatentImage"
+    title: "EmptyChromaRadianceLatentImage"
+
+connections: []
+
+ui_map:
+  width: "latent_source:width"
+  height: "latent_source:height"
+  batch_size: "latent_source:batch_size"

core/pipelines/workflow_recipes/_partials/input/txt2img_flux2_latent.yaml

@@ -0,0 +1,11 @@
+nodes:
+  latent_source:
+    class_type: "EmptyFlux2LatentImage"
+    title: "Empty Flux 2 Latent"
+
+connections: []
+
+ui_map:
+  width: "latent_source:width"
+  height: "latent_source:height"
+  batch_size: "latent_source:batch_size"

core/pipelines/workflow_recipes/_partials/input/txt2img_hunyuan_latent.yaml

@@ -0,0 +1,11 @@
+nodes:
+  latent_source:
+    class_type: "EmptyHunyuanImageLatent"
+    title: "EmptyHunyuanImageLatent"
+
+connections: []
+
+ui_map:
+  width: "latent_source:width"
+  height: "latent_source:height"
+  batch_size: "latent_source:batch_size"

core/pipelines/workflow_recipes/_partials/input/txt2img_latent.yaml

@@ -0,0 +1,11 @@
+nodes:
+  latent_source:
+    class_type: "{{ latent_generator_template }}"
+    title: "Empty Latent Image"
+
+connections: []
+
+ui_map:
+  width: "latent_source:width"
+  height: "latent_source:height"
+  batch_size: "latent_source:batch_size"

core/pipelines/workflow_recipes/_partials/input/txt2img_sd3_latent.yaml

@@ -0,0 +1,11 @@
+nodes:
+  latent_source:
+    class_type: "EmptySD3LatentImage"
+    title: "EmptySD3LatentImage"
+
+connections: []
+
+ui_map:
+  width: "latent_source:width"
+  height: "latent_source:height"
+  batch_size: "latent_source:batch_size"

core/pipelines/workflow_recipes/sd_unified_recipe.yaml

@@ -1,10 +1,8 @@
 imports:
-  - "_partials/_base_sampler.yaml"
+  - "_partials/_base_sampler_sd.yaml"
   - "_partials/input/{{ task_type }}.yaml"
-  - "_partials/conditioning/sdxl.yaml"
+  - "_partials/conditioning/{{ model_type }}.yaml"
 
 connections:
   - from: "latent_source:0"
-    to: "ksampler:latent_image"
-  - from: "ckpt_loader:2"
-    to: "vae_encode:vae"
+    to: "ksampler:latent_image"

core/settings.py

@@ -6,18 +6,40 @@ CHECKPOINT_DIR = "models/checkpoints"
 LORA_DIR = "models/loras"
 EMBEDDING_DIR = "models/embeddings" 
 CONTROLNET_DIR = "models/controlnet"
+MODEL_PATCHES_DIR = "models/model_patches"
 DIFFUSION_MODELS_DIR = "models/diffusion_models"
 VAE_DIR = "models/vae"
 TEXT_ENCODERS_DIR = "models/text_encoders"
+STYLE_MODELS_DIR = "models/style_models"
+CLIP_VISION_DIR = "models/clip_vision"
+IPADAPTER_DIR = "models/ipadapter"
+IPADAPTER_FLUX_DIR = "models/ipadapter-flux"
 INPUT_DIR = "input"
 OUTPUT_DIR = "output"
 
+CATEGORY_TO_DIR_MAP = {
+    "diffusion_models": DIFFUSION_MODELS_DIR,
+    "text_encoders": TEXT_ENCODERS_DIR,
+    "vae": VAE_DIR,
+    "checkpoints": CHECKPOINT_DIR,
+    "loras": LORA_DIR,
+    "controlnet": CONTROLNET_DIR,
+    "model_patches": MODEL_PATCHES_DIR,
+    "embeddings": EMBEDDING_DIR,
+    "style_models": STYLE_MODELS_DIR,
+    "clip_vision": CLIP_VISION_DIR,
+    "ipadapter": IPADAPTER_DIR,
+    "ipadapter-flux": IPADAPTER_FLUX_DIR
+}
+
 _PROJECT_ROOT = os.path.dirname(os.path.dirname(os.path.abspath(__file__)))
 _MODEL_LIST_PATH = os.path.join(_PROJECT_ROOT, 'yaml', 'model_list.yaml')
 _FILE_LIST_PATH = os.path.join(_PROJECT_ROOT, 'yaml', 'file_list.yaml')
 _IPADAPTER_LIST_PATH = os.path.join(_PROJECT_ROOT, 'yaml', 'ipadapter.yaml')
 _CONSTANTS_PATH = os.path.join(_PROJECT_ROOT, 'yaml', 'constants.yaml')
-
+_MODEL_ARCHITECTURES_PATH = os.path.join(_PROJECT_ROOT, 'yaml', 'model_architectures.yaml')
+_IMAGE_GEN_FEATURES_PATH = os.path.join(_PROJECT_ROOT, 'yaml', 'image_gen_features.yaml')
+_MODEL_DEFAULTS_PATH = os.path.join(_PROJECT_ROOT, 'yaml', 'model_defaults.yaml')
 
 def load_constants_from_yaml(filepath=_CONSTANTS_PATH):
     if not os.path.exists(filepath):
@@ -26,6 +48,27 @@ def load_constants_from_yaml(filepath=_CONSTANTS_PATH):
     with open(filepath, 'r', encoding='utf-8') as f:
         return yaml.safe_load(f)
 
+def load_architectures_config(filepath=_MODEL_ARCHITECTURES_PATH):
+    if not os.path.exists(filepath):
+        print(f"Warning: Architectures file not found at {filepath}.")
+        return {}
+    with open(filepath, 'r', encoding='utf-8') as f:
+        return yaml.safe_load(f)
+
+def load_features_config(filepath=_IMAGE_GEN_FEATURES_PATH):
+    if not os.path.exists(filepath):
+        print(f"Warning: Features file not found at {filepath}.")
+        return {}
+    with open(filepath, 'r', encoding='utf-8') as f:
+        return yaml.safe_load(f)
+
+def load_model_defaults(filepath=_MODEL_DEFAULTS_PATH):
+    if not os.path.exists(filepath):
+        print(f"Warning: Model defaults file not found at {filepath}.")
+        return {}
+    with open(filepath, 'r', encoding='utf-8') as f:
+        return yaml.safe_load(f)
+
 def load_file_download_map(filepath=_FILE_LIST_PATH):
     if not os.path.exists(filepath):
         raise FileNotFoundError(f"The file list (for downloads) was not found at: {filepath}")
@@ -58,27 +101,40 @@ def load_models_from_yaml(model_list_filepath=_MODEL_LIST_PATH, download_map=Non
     }
     category_map_names = {
         "Checkpoint": "MODEL_MAP_CHECKPOINT",
+        "Checkpoints": "MODEL_MAP_CHECKPOINT"
     }
     
-    for category, models in model_data.items():
+    for category, architectures in model_data.items():
         if category in category_map_names:
             map_name = category_map_names[category]
-            if not isinstance(models, list): continue
-            for model in models:
-                display_name = model['display_name']
-                filename = model['path']
+            if not isinstance(architectures, dict): continue
+            
+            for arch, arch_data in architectures.items():
+                if not isinstance(arch_data, dict): continue
+                
+                latent_type = arch_data.get('latent_type', 'latent')
+                models = arch_data.get('models', [])
+                if not isinstance(models, list): continue
                 
-                download_info = download_map.get(filename, {})
-                repo_id = download_info.get('repo_id', '')
-
-                model_tuple = (
-                    repo_id,
-                    filename,
-                    "SDXL",
-                    None
-                )
-                model_maps[map_name][display_name] = model_tuple
-                model_maps["ALL_MODEL_MAP"][display_name] = model_tuple
+                for model in models:
+                    display_name = model['display_name']
+                    path_or_components = model.get('path') or model.get('components')
+                    mod_category = model.get('category', None)
+                    
+                    repo_id = ''
+                    if isinstance(path_or_components, str):
+                        download_info = download_map.get(path_or_components, {})
+                        repo_id = download_info.get('repo_id', '')
+                    
+                    model_tuple = (
+                        repo_id,
+                        path_or_components,
+                        arch,
+                        latent_type,
+                        mod_category
+                    )
+                    model_maps[map_name][display_name] = model_tuple
+                    model_maps["ALL_MODEL_MAP"][display_name] = model_tuple
     
     return model_maps
 
@@ -88,13 +144,43 @@ try:
     MODEL_MAP_CHECKPOINT = loaded_maps["MODEL_MAP_CHECKPOINT"]
     ALL_MODEL_MAP = loaded_maps["ALL_MODEL_MAP"]
     
+    category_to_model_type = {
+        "diffusion_models": "UNET",
+        "text_encoders": "TEXT_ENCODER",
+        "vae": "VAE",
+        "checkpoints": "SDXL",
+        "loras": "LORA",
+        "controlnet": "CONTROLNET",
+        "model_patches": "MODEL_PATCH",
+        "style_models": "STYLE",
+        "clip_vision": "CLIP_VISION",
+        "ipadapter": "IPADAPTER",
+        "ipadapter-flux": "IPADAPTER_FLUX"
+    }
+    for filename, file_info in ALL_FILE_DOWNLOAD_MAP.items():
+        if filename not in ALL_MODEL_MAP:
+            category = file_info.get('category')
+            model_type = category_to_model_type.get(category, 'UNKNOWN')
+            repo_id = file_info.get('repo_id', '')
+            ALL_MODEL_MAP[filename] = (repo_id, filename, model_type, None, None)
+
     MODEL_TYPE_MAP = {k: v[2] for k, v in ALL_MODEL_MAP.items()}
 
+    ARCH_CATEGORIES_MAP = {}
+    for display_name, info in MODEL_MAP_CHECKPOINT.items():
+        arch = info[2]
+        cat = info[4] if len(info) > 4 else None
+        if arch not in ARCH_CATEGORIES_MAP:
+            ARCH_CATEGORIES_MAP[arch] = []
+        if cat and cat not in ARCH_CATEGORIES_MAP[arch]:
+            ARCH_CATEGORIES_MAP[arch].append(cat)
+
 except Exception as e:
     print(f"FATAL: Could not load model configuration from YAML. Error: {e}")
     ALL_FILE_DOWNLOAD_MAP = {}
     MODEL_MAP_CHECKPOINT, ALL_MODEL_MAP = {}, {}
     MODEL_TYPE_MAP = {}
+    ARCH_CATEGORIES_MAP = {}
 
 
 try:
@@ -104,13 +190,16 @@ try:
     MAX_CONDITIONINGS = _constants.get('MAX_CONDITIONINGS', 10)
     MAX_CONTROLNETS = _constants.get('MAX_CONTROLNETS', 5)
     MAX_IPADAPTERS = _constants.get('MAX_IPADAPTERS', 5)
-    LORA_SOURCE_CHOICES = _constants.get('LORA_SOURCE_CHOICES', ["Civitai", "Custom URL", "File"])
+    LORA_SOURCE_CHOICES = _constants.get('LORA_SOURCE_CHOICES', ["Civitai", "File"])
     RESOLUTION_MAP = _constants.get('RESOLUTION_MAP', {})
+    ARCHITECTURES_CONFIG = load_architectures_config()
+    FEATURES_CONFIG = load_features_config()
+    MODEL_DEFAULTS_CONFIG = load_model_defaults()
 except Exception as e:
     print(f"FATAL: Could not load constants from YAML. Error: {e}")
     MAX_LORAS, MAX_EMBEDDINGS, MAX_CONDITIONINGS, MAX_CONTROLNETS, MAX_IPADAPTERS = 5, 5, 10, 5, 5
-    LORA_SOURCE_CHOICES = ["Civitai", "Custom URL", "File"]
+    LORA_SOURCE_CHOICES = ["Civitai", "File"]
     RESOLUTION_MAP = {}
-
-
-DEFAULT_NEGATIVE_PROMPT = "monochrome, (low quality, worst quality:1.2), 3d, watermark, signature, ugly, poorly drawn,"
+    ARCHITECTURES_CONFIG = {}
+    FEATURES_CONFIG = {}
+    MODEL_DEFAULTS_CONFIG = {}

requirements.txt

@@ -58,4 +58,5 @@ svglib
 trimesh[easy]
 yacs
 yapf
-onnxruntime-gpu
+onnxruntime-gpu
+diffusers

ui/events.py

@@ -8,8 +8,7 @@ from utils.app_utils import *
 from core.generation_logic import *
 from comfy_integration.nodes import SAMPLER_CHOICES, SCHEDULER_CHOICES
 
-from core.pipelines.controlnet_preprocessor import CPU_ONLY_PREPROCESSORS
-from utils.app_utils import PREPROCESSOR_MODEL_MAP, PREPROCESSOR_PARAMETER_MAP, save_uploaded_file_with_hash
+from utils.app_utils import save_uploaded_file_with_hash
 from ui.shared.ui_components import RESOLUTION_MAP, MAX_CONTROLNETS, MAX_IPADAPTERS, MAX_EMBEDDINGS, MAX_CONDITIONINGS, MAX_LORAS
 
 
@@ -22,10 +21,74 @@ def load_controlnet_config():
         with open(_CN_MODEL_LIST_PATH, 'r', encoding='utf-8') as f:
             config = yaml.safe_load(f)
         print("--- ✅ controlnet_models.yaml loaded successfully ---")
-        return config.get("ControlNet", {}).get("SDXL", [])
+        return config.get("ControlNet", {})
     except Exception as e:
         print(f"Error loading controlnet_models.yaml: {e}")
-        return []
+        return {}
+
+
+def get_cn_defaults(arch_val):
+    cn_full_config = load_controlnet_config()
+    cn_config = cn_full_config.get(arch_val, [])
+    
+    if not cn_config:
+        return [], None, [], None, "None"
+        
+    all_types = sorted(list(set(t for model in cn_config for t in model.get("Type", []))))
+    default_type = all_types[0] if all_types else None
+    
+    series_choices = []
+    if default_type:
+        series_choices = sorted(list(set(model.get("Series", "Default") for model in cn_config if default_type in model.get("Type", []))))
+    default_series = series_choices[0] if series_choices else None
+    
+    filepath = "None"
+    if default_series and default_type:
+        for model in cn_config:
+            if model.get("Series") == default_series and default_type in model.get("Type", []):
+                filepath = model.get("Filepath")
+                break
+                
+    return all_types, default_type, series_choices, default_series, filepath
+
+@lru_cache(maxsize=1)
+def load_diffsynth_controlnet_config():
+    _PROJECT_ROOT = os.path.dirname(os.path.dirname(os.path.abspath(__file__)))
+    _CN_MODEL_LIST_PATH = os.path.join(_PROJECT_ROOT, 'yaml', 'diffsynth_controlnet_models.yaml')
+    try:
+        print("--- Loading diffsynth_controlnet_models.yaml ---")
+        with open(_CN_MODEL_LIST_PATH, 'r', encoding='utf-8') as f:
+            config = yaml.safe_load(f)
+        print("--- ✅ diffsynth_controlnet_models.yaml loaded successfully ---")
+        return config.get("DiffSynth_ControlNet", {})
+    except Exception as e:
+        print(f"Error loading diffsynth_controlnet_models.yaml: {e}")
+        return {}
+
+def get_diffsynth_cn_defaults(arch_val):
+    cn_full_config = load_diffsynth_controlnet_config()
+    cn_config = cn_full_config.get(arch_val, [])
+    
+    if not cn_config:
+        return [], None, [], None, "None"
+        
+    all_types = sorted(list(set(t for model in cn_config for t in model.get("Type", []))))
+    default_type = all_types[0] if all_types else None
+    
+    series_choices = []
+    if default_type:
+        series_choices = sorted(list(set(model.get("Series", "Default") for model in cn_config if default_type in model.get("Type", []))))
+    default_series = series_choices[0] if series_choices else None
+    
+    filepath = "None"
+    if default_series and default_type:
+        for model in cn_config:
+            if model.get("Series") == default_series and default_type in model.get("Type", []):
+                filepath = model.get("Filepath")
+                break
+                
+    return all_types, default_type, series_choices, default_series, filepath
+
 
 @lru_cache(maxsize=1)
 def load_ipadapter_config():
@@ -42,118 +105,96 @@ def load_ipadapter_config():
         return {}
 
 
-def attach_event_handlers(ui_components, demo):
-    def update_cn_input_visibility(choice):
-        return {
-            ui_components["cn_image_input"]: gr.update(visible=choice == "Image"),
-            ui_components["cn_video_input"]: gr.update(visible=choice == "Video")
-        }
-    ui_components["cn_input_type"].change(
-        fn=update_cn_input_visibility,
-        inputs=[ui_components["cn_input_type"]],
-        outputs=[ui_components["cn_image_input"], ui_components["cn_video_input"]]
-    )
+def apply_data_to_ui(data, prefix, ui_components):
+    final_sampler = data.get('sampler') if data.get('sampler') in SAMPLER_CHOICES else SAMPLER_CHOICES[0]
+    default_scheduler = 'normal' if 'normal' in SCHEDULER_CHOICES else SCHEDULER_CHOICES[0]
+    final_scheduler = data.get('scheduler') if data.get('scheduler') in SCHEDULER_CHOICES else default_scheduler
+
+    updates = {}
+    base_model_name = data.get('base_model')
     
-    def update_preprocessor_models_dropdown(preprocessor_name):
-        models = PREPROCESSOR_MODEL_MAP.get(preprocessor_name)
-        if models:
-            model_filenames = [m[1] for m in models]
-            return gr.update(choices=model_filenames, value=model_filenames[0], visible=True)
-        else:
-            return gr.update(choices=[], value=None, visible=False)
-
-    def update_preprocessor_settings_ui(preprocessor_name):
-        from ui.layout import MAX_DYNAMIC_CONTROLS
-        params = PREPROCESSOR_PARAMETER_MAP.get(preprocessor_name, [])
-        
-        slider_updates, dropdown_updates, checkbox_updates = [], [], []
-        
-        s_idx, d_idx, c_idx = 0, 0, 0
-
-        for param in params:
-            if s_idx + d_idx + c_idx >= MAX_DYNAMIC_CONTROLS: break
-            
-            name = param["name"]
-            ptype = param["type"]
-            config = param["config"]
-            label = name.replace('_', ' ').title()
-
-            if ptype == "INT" or ptype == "FLOAT":
-                if s_idx < MAX_DYNAMIC_CONTROLS:
-                    slider_updates.append(gr.update(
-                        label=label,
-                        minimum=config.get('min', 0),
-                        maximum=config.get('max', 255),
-                        step=config.get('step', 0.1 if ptype == "FLOAT" else 1),
-                        value=config.get('default', 0),
-                        visible=True
-                    ))
-                    s_idx += 1
-            elif isinstance(ptype, list):
-                if d_idx < MAX_DYNAMIC_CONTROLS:
-                    dropdown_updates.append(gr.update(
-                        label=label,
-                        choices=ptype,
-                        value=config.get('default', ptype[0] if ptype else None),
-                        visible=True
-                    ))
-                    d_idx += 1
-            elif ptype == "BOOLEAN":
-                if c_idx < MAX_DYNAMIC_CONTROLS:
-                    checkbox_updates.append(gr.update(
-                        label=label,
-                        value=config.get('default', False),
-                        visible=True
-                    ))
-                    c_idx += 1
-
-        for _ in range(s_idx, MAX_DYNAMIC_CONTROLS): slider_updates.append(gr.update(visible=False))
-        for _ in range(d_idx, MAX_DYNAMIC_CONTROLS): dropdown_updates.append(gr.update(visible=False))
-        for _ in range(c_idx, MAX_DYNAMIC_CONTROLS): checkbox_updates.append(gr.update(visible=False))
-
-        return slider_updates + dropdown_updates + checkbox_updates
-
-    def update_run_button_for_cpu(preprocessor_name):
-        if preprocessor_name in CPU_ONLY_PREPROCESSORS:
-            return gr.update(value="Run Preprocessor CPU Only", variant="primary"), gr.update(visible=False)
+    model_map = MODEL_MAP_CHECKPOINT
+
+    if f'base_model_{prefix}' in ui_components:
+        model_dropdown_component = ui_components[f'base_model_{prefix}']
+        if base_model_name and base_model_name in model_map:
+            updates[model_dropdown_component] = base_model_name
+            if f'model_arch_{prefix}' in ui_components:
+                m_type = MODEL_TYPE_MAP.get(base_model_name, "SDXL")
+                updates[ui_components[f'model_arch_{prefix}']] = m_type
+            if f'model_cat_{prefix}' in ui_components:
+                m_info = model_map.get(base_model_name)
+                m_cat = m_info[4] if m_info and len(m_info) > 4 else None
+                updates[ui_components[f'model_cat_{prefix}']] = m_cat if m_cat else "ALL"
         else:
-            return gr.update(value="Run Preprocessor", variant="primary"), gr.update(visible=True)
-
-    ui_components["preprocessor_cn"].change(
-        fn=update_preprocessor_models_dropdown,
-        inputs=[ui_components["preprocessor_cn"]],
-        outputs=[ui_components["preprocessor_model_cn"]]
-    ).then(
-        fn=update_preprocessor_settings_ui,
-        inputs=[ui_components["preprocessor_cn"]],
-        outputs=ui_components["cn_sliders"] + ui_components["cn_dropdowns"] + ui_components["cn_checkboxes"]
-    ).then(
-        fn=update_run_button_for_cpu,
-        inputs=[ui_components["preprocessor_cn"]],
-        outputs=[ui_components["run_cn"], ui_components["zero_gpu_cn"]]
-    )
-
-    all_dynamic_inputs = (
-        ui_components["cn_sliders"] + 
-        ui_components["cn_dropdowns"] + 
-        ui_components["cn_checkboxes"]
-    )
-
-    ui_components["run_cn"].click(
-        fn=run_cn_preprocessor_entry,
-        inputs=[
-            ui_components["cn_input_type"],
-            ui_components["cn_image_input"],
-            ui_components["cn_video_input"],
-            ui_components["preprocessor_cn"],
-            ui_components["preprocessor_model_cn"],
-            ui_components["zero_gpu_cn"],
-        ] + all_dynamic_inputs,
-        outputs=[ui_components["output_gallery_cn"]]
-    )
+            updates[model_dropdown_component] = gr.update()
+    
+    common_params = {
+        f'prompt_{prefix}': data.get('prompt', ''),
+        f'neg_prompt_{prefix}': data.get('negative_prompt', ''),
+        f'seed_{prefix}': data.get('seed', -1),
+        f'cfg_{prefix}': data.get('cfg_scale', 7.5),
+        f'steps_{prefix}': data.get('steps', 28),
+        f'sampler_{prefix}': final_sampler,
+        f'scheduler_{prefix}': final_scheduler,
+    }
+    
+    for comp_name, value in common_params.items():
+        if comp_name in ui_components:
+            updates[ui_components[comp_name]] = value
+    
+    if prefix == 'txt2img':
+         if f'width_{prefix}' in ui_components:
+             updates[ui_components[f'width_{prefix}']] = data.get('width', 1024)
+         if f'height_{prefix}' in ui_components:
+             updates[ui_components[f'height_{prefix}']] = data.get('height', 1024)
+
+    tab_indices = {"txt2img": 0, "img2img": 1, "inpaint": 2, "outpaint": 3, "hires_fix": 4}
+    tab_index = tab_indices.get(prefix, 0)
+
+    updates[ui_components['tabs']] = gr.Tabs(selected=tab_index)
+    
+    return updates
+
+
+def send_info_to_tab(image, prefix, ui_components):
+    if not image or not image.info.get('parameters', ''):
+        all_comps = [comp for comp_or_list in ui_components.values() for comp in (comp_or_list if isinstance(comp_or_list, list) else [comp_or_list])]
+        return {comp: gr.update() for comp in all_comps}
+    
+    data = parse_parameters(image.info['parameters'])
+    
+    image_input_map = {
+        "img2img": 'input_image_img2img',
+        "inpaint": 'input_image_dict_inpaint',
+        "outpaint": 'input_image_outpaint',
+        "hires_fix": 'input_image_hires_fix'
+    }
+    
+    updates = apply_data_to_ui(data, prefix, ui_components)
+    
+    if prefix in image_input_map and image_input_map[prefix] in ui_components:
+        component_key = image_input_map[prefix]
+        updates[ui_components[component_key]] = gr.update(value=image)
+        
+    return updates
+
+
+def send_info_by_hash(image, ui_components):
+    if not image or not image.info.get('parameters', ''):
+        all_comps = [comp for comp_or_list in ui_components.values() for comp in (comp_or_list if isinstance(comp_or_list, list) else [comp_or_list])]
+        return {comp: gr.update() for comp in all_comps}
+
+    data = parse_parameters(image.info['parameters'])
+    
+    return apply_data_to_ui(data, "txt2img", ui_components)
+
+
+def attach_event_handlers(ui_components, demo):
     
     def create_lora_event_handlers(prefix):
-        lora_rows = ui_components[f'lora_rows_{prefix}']
+        lora_rows = ui_components.get(f'lora_rows_{prefix}')
+        if not lora_rows: return
         lora_ids = ui_components[f'lora_ids_{prefix}']
         lora_scales = ui_components[f'lora_scales_{prefix}']
         lora_uploads = ui_components[f'lora_uploads_{prefix}']
@@ -193,7 +234,8 @@ def attach_event_handlers(ui_components, demo):
         del_button.click(del_lora_row, [count_state], del_outputs, show_progress=False)
         
     def create_controlnet_event_handlers(prefix):
-        cn_rows = ui_components[f'controlnet_rows_{prefix}']
+        cn_rows = ui_components.get(f'controlnet_rows_{prefix}')
+        if not cn_rows: return
         cn_types = ui_components[f'controlnet_types_{prefix}']
         cn_series = ui_components[f'controlnet_series_{prefix}']
         cn_filepaths = ui_components[f'controlnet_filepaths_{prefix}']
@@ -205,6 +247,114 @@ def attach_event_handlers(ui_components, demo):
         del_button = ui_components[f'delete_controlnet_button_{prefix}']
         accordion = ui_components[f'controlnet_accordion_{prefix}']
 
+        arch_comp = ui_components.get(f'model_arch_{prefix}')
+        actual_arch_comp = arch_comp if arch_comp else gr.State("SDXL")
+
+        def add_cn_row(c):
+            c += 1
+            updates = {
+                count_state: c,
+                cn_rows[c-1]: gr.update(visible=True),
+                add_button: gr.update(visible=c < MAX_CONTROLNETS),
+                del_button: gr.update(visible=True)
+            }
+            return updates
+
+        def del_cn_row(c):
+            c -= 1
+            updates = {
+                count_state: c,
+                cn_rows[c]: gr.update(visible=False),
+                cn_images[c]: None,
+                cn_strengths[c]: 1.0,
+                add_button: gr.update(visible=True),
+                del_button: gr.update(visible=c > 0)
+            }
+            return updates
+            
+        add_outputs = [count_state, add_button, del_button] + cn_rows
+        del_outputs = [count_state, add_button, del_button] + cn_rows + cn_images + cn_strengths
+        add_button.click(fn=add_cn_row, inputs=[count_state], outputs=add_outputs, show_progress=False)
+        del_button.click(fn=del_cn_row, inputs=[count_state], outputs=del_outputs, show_progress=False)
+        
+        def on_cn_type_change(selected_type, arch_val):
+            cn_full_config = load_controlnet_config()
+            
+            architectures_dict = ARCHITECTURES_CONFIG.get('architectures', {})
+            controlnet_key = architectures_dict.get(arch_val, {}).get("controlnet_key", arch_val)
+
+            cn_config = cn_full_config.get(controlnet_key, [])
+            series_choices = []
+            if selected_type:
+                series_choices = sorted(list(set(
+                    model.get("Series", "Default") for model in cn_config
+                    if selected_type in model.get("Type", [])
+                )))
+            default_series = series_choices[0] if series_choices else None
+            filepath = "None"
+            if default_series:
+                for model in cn_config:
+                    if model.get("Series") == default_series and selected_type in model.get("Type", []):
+                        filepath = model.get("Filepath")
+                        break
+            return gr.update(choices=series_choices, value=default_series), filepath
+
+        def on_cn_series_change(selected_series, selected_type, arch_val):
+            cn_full_config = load_controlnet_config()
+            
+            architectures_dict = ARCHITECTURES_CONFIG.get('architectures', {})
+            controlnet_key = architectures_dict.get(arch_val, {}).get("controlnet_key", arch_val)
+
+            cn_config = cn_full_config.get(controlnet_key, [])
+            filepath = "None"
+            if selected_series and selected_type:
+                for model in cn_config:
+                    if model.get("Series") == selected_series and selected_type in model.get("Type", []):
+                        filepath = model.get("Filepath")
+                        break
+            return filepath
+            
+        for i in range(MAX_CONTROLNETS):
+            cn_types[i].change(
+                fn=on_cn_type_change,
+                inputs=[cn_types[i], actual_arch_comp],
+                outputs=[cn_series[i], cn_filepaths[i]],
+                show_progress=False
+            )
+            cn_series[i].change(
+                fn=on_cn_series_change,
+                inputs=[cn_series[i], cn_types[i], actual_arch_comp],
+                outputs=[cn_filepaths[i]],
+                show_progress=False
+            )
+
+        def on_accordion_expand(*images):
+            return [gr.update() for _ in images]
+        
+        accordion.expand(
+            fn=on_accordion_expand,
+            inputs=cn_images,
+            outputs=cn_images,
+            show_progress=False
+        )
+
+    def create_diffsynth_controlnet_event_handlers(prefix):
+        cn_rows = ui_components.get(f'diffsynth_controlnet_rows_{prefix}')
+        if not cn_rows: return
+        cn_types = ui_components[f'diffsynth_controlnet_types_{prefix}']
+        cn_series = ui_components[f'diffsynth_controlnet_series_{prefix}']
+        cn_filepaths = ui_components[f'diffsynth_controlnet_filepaths_{prefix}']
+        cn_images = ui_components[f'diffsynth_controlnet_images_{prefix}']
+        cn_strengths = ui_components[f'diffsynth_controlnet_strengths_{prefix}']
+        
+        count_state = ui_components[f'diffsynth_controlnet_count_state_{prefix}']
+        add_button = ui_components[f'add_diffsynth_controlnet_button_{prefix}']
+        del_button = ui_components[f'delete_diffsynth_controlnet_button_{prefix}']
+        accordion = ui_components[f'diffsynth_controlnet_accordion_{prefix}']
+
+        arch_comp = ui_components.get(f'model_arch_{prefix}')
+        actual_arch_comp = arch_comp if arch_comp else gr.State("Z-Image")
+
         def add_cn_row(c):
             c += 1
             updates = {
@@ -232,8 +382,13 @@ def attach_event_handlers(ui_components, demo):
         add_button.click(fn=add_cn_row, inputs=[count_state], outputs=add_outputs, show_progress=False)
         del_button.click(fn=del_cn_row, inputs=[count_state], outputs=del_outputs, show_progress=False)
         
-        def on_cn_type_change(selected_type):
-            cn_config = load_controlnet_config()
+        def on_cn_type_change(selected_type, arch_val):
+            cn_full_config = load_diffsynth_controlnet_config()
+            
+            architectures_dict = ARCHITECTURES_CONFIG.get('architectures', {})
+            controlnet_key = architectures_dict.get(arch_val, {}).get("controlnet_key", arch_val)
+
+            cn_config = cn_full_config.get(controlnet_key, [])
             series_choices = []
             if selected_type:
                 series_choices = sorted(list(set(
@@ -249,8 +404,13 @@ def attach_event_handlers(ui_components, demo):
                         break
             return gr.update(choices=series_choices, value=default_series), filepath
 
-        def on_cn_series_change(selected_series, selected_type):
-            cn_config = load_controlnet_config()
+        def on_cn_series_change(selected_series, selected_type, arch_val):
+            cn_full_config = load_diffsynth_controlnet_config()
+            
+            architectures_dict = ARCHITECTURES_CONFIG.get('architectures', {})
+            controlnet_key = architectures_dict.get(arch_val, {}).get("controlnet_key", arch_val)
+
+            cn_config = cn_full_config.get(controlnet_key, [])
             filepath = "None"
             if selected_series and selected_type:
                 for model in cn_config:
@@ -262,13 +422,13 @@ def attach_event_handlers(ui_components, demo):
         for i in range(MAX_CONTROLNETS):
             cn_types[i].change(
                 fn=on_cn_type_change,
-                inputs=[cn_types[i]],
+                inputs=[cn_types[i], actual_arch_comp],
                 outputs=[cn_series[i], cn_filepaths[i]],
                 show_progress=False
             )
             cn_series[i].change(
                 fn=on_cn_series_change,
-                inputs=[cn_series[i], cn_types[i]],
+                inputs=[cn_series[i], cn_types[i], actual_arch_comp],
                 outputs=[cn_filepaths[i]],
                 show_progress=False
             )
@@ -283,8 +443,69 @@ def attach_event_handlers(ui_components, demo):
             show_progress=False
         )
 
+    def create_flux1_ipadapter_event_handlers(prefix):
+        fipa_rows = ui_components.get(f'flux1_ipadapter_rows_{prefix}')
+        if not fipa_rows: return
+        count_state = ui_components[f'flux1_ipadapter_count_state_{prefix}']
+        add_button = ui_components[f'add_flux1_ipadapter_button_{prefix}']
+        del_button = ui_components[f'delete_flux1_ipadapter_button_{prefix}']
+
+        def add_fipa_row(c):
+            c += 1
+            return {
+                count_state: c,
+                fipa_rows[c - 1]: gr.update(visible=True),
+                add_button: gr.update(visible=c < MAX_IPADAPTERS),
+                del_button: gr.update(visible=True),
+            }
+
+        def del_fipa_row(c):
+            c -= 1
+            return {
+                count_state: c,
+                fipa_rows[c]: gr.update(visible=False),
+                add_button: gr.update(visible=True),
+                del_button: gr.update(visible=c > 0),
+            }
+
+        add_outputs = [count_state, add_button, del_button] + fipa_rows
+        del_outputs = [count_state, add_button, del_button] + fipa_rows
+        add_button.click(fn=add_fipa_row, inputs=[count_state], outputs=add_outputs, show_progress=False)
+        del_button.click(fn=del_fipa_row, inputs=[count_state], outputs=del_outputs, show_progress=False)
+
+    def create_style_event_handlers(prefix):
+        style_rows = ui_components.get(f'style_rows_{prefix}')
+        if not style_rows: return
+        count_state = ui_components[f'style_count_state_{prefix}']
+        add_button = ui_components[f'add_style_button_{prefix}']
+        del_button = ui_components[f'delete_style_button_{prefix}']
+
+        def add_style_row(c):
+            c += 1
+            return {
+                count_state: c,
+                style_rows[c - 1]: gr.update(visible=True),
+                add_button: gr.update(visible=c < 5),
+                del_button: gr.update(visible=True),
+            }
+
+        def del_style_row(c):
+            c -= 1
+            return {
+                count_state: c,
+                style_rows[c]: gr.update(visible=False),
+                add_button: gr.update(visible=True),
+                del_button: gr.update(visible=c > 0),
+            }
+
+        add_outputs = [count_state, add_button, del_button] + style_rows
+        del_outputs = [count_state, add_button, del_button] + style_rows
+        add_button.click(fn=add_style_row, inputs=[count_state], outputs=add_outputs, show_progress=False)
+        del_button.click(fn=del_style_row, inputs=[count_state], outputs=del_outputs, show_progress=False)
+
     def create_ipadapter_event_handlers(prefix):
-        ipa_rows = ui_components[f'ipadapter_rows_{prefix}']
+        ipa_rows = ui_components.get(f'ipadapter_rows_{prefix}')
+        if not ipa_rows: return
         ipa_lora_strengths = ui_components[f'ipadapter_lora_strengths_{prefix}']
         ipa_final_preset = ui_components[f'ipadapter_final_preset_{prefix}']
         ipa_final_lora_strength = ui_components[f'ipadapter_final_lora_strength_{prefix}']
@@ -319,11 +540,10 @@ def attach_event_handlers(ui_components, demo):
         def on_preset_change(preset_value):
             config = load_ipadapter_config()
             faceid_presets = []
-            if isinstance(config, list):
-                faceid_presets = [
-                    p.get('preset_name', '') for p in config 
-                    if 'FACE' in p.get('preset_name', '') or 'FACEID' in p.get('preset_name', '')
-                ]
+            if config:
+                faceid_presets.extend(config.get("IPAdapter_FaceID_presets", {}).get("SDXL", []))
+                faceid_presets.extend(config.get("IPAdapter_FaceID_presets", {}).get("SD1.5", []))
+
             is_visible = preset_value in faceid_presets
             updates = [gr.update(visible=is_visible)] * (MAX_IPADAPTERS + 1)
             return updates
@@ -333,9 +553,42 @@ def attach_event_handlers(ui_components, demo):
         
         accordion.expand(fn=lambda *imgs: [gr.update() for _ in imgs], inputs=ui_components[f'ipadapter_images_{prefix}'], outputs=ui_components[f'ipadapter_images_{prefix}'], show_progress=False)
 
+    def create_reference_latent_event_handlers(prefix):
+        ref_rows = ui_components.get(f'reference_latent_rows_{prefix}')
+        if not ref_rows: return
+        count_state = ui_components[f'reference_latent_count_state_{prefix}']
+        add_button = ui_components[f'add_reference_latent_button_{prefix}']
+        del_button = ui_components[f'delete_reference_latent_button_{prefix}']
+        images = ui_components[f'reference_latent_images_{prefix}']
+
+        def add_ref_row(c):
+            c += 1
+            return {
+                count_state: c,
+                ref_rows[c - 1]: gr.update(visible=True),
+                add_button: gr.update(visible=c < 10),
+                del_button: gr.update(visible=True),
+            }
+
+        def del_ref_row(c):
+            c -= 1
+            return {
+                count_state: c,
+                ref_rows[c]: gr.update(visible=False),
+                images[c]: None,
+                add_button: gr.update(visible=True),
+                del_button: gr.update(visible=c > 0),
+            }
+
+        add_outputs = [count_state, add_button, del_button] + ref_rows
+        del_outputs = [count_state, add_button, del_button] + ref_rows + images
+        add_button.click(fn=add_ref_row, inputs=[count_state], outputs=add_outputs, show_progress=False)
+        del_button.click(fn=del_ref_row, inputs=[count_state], outputs=del_outputs, show_progress=False)
+
 
     def create_embedding_event_handlers(prefix):
-        rows = ui_components[f'embedding_rows_{prefix}']
+        rows = ui_components.get(f'embedding_rows_{prefix}')
+        if not rows: return
         ids = ui_components[f'embeddings_ids_{prefix}']
         files = ui_components[f'embeddings_files_{prefix}']
         count_state = ui_components[f'embedding_count_state_{prefix}']
@@ -368,7 +621,8 @@ def attach_event_handlers(ui_components, demo):
         del_button.click(fn=del_row, inputs=[count_state], outputs=del_outputs, show_progress=False)
     
     def create_conditioning_event_handlers(prefix):
-        rows = ui_components[f'conditioning_rows_{prefix}']
+        rows = ui_components.get(f'conditioning_rows_{prefix}')
+        if not rows: return
         prompts = ui_components[f'conditioning_prompts_{prefix}']
         count_state = ui_components[f'conditioning_count_state_{prefix}']
         add_button = ui_components[f'add_conditioning_button_{prefix}']
@@ -423,38 +677,48 @@ def attach_event_handlers(ui_components, demo):
     def create_run_event(prefix: str, task_type: str):
         run_inputs_map = {
             'model_display_name': ui_components[f'base_model_{prefix}'],
-            'positive_prompt': ui_components[f'prompt_{prefix}'],
-            'negative_prompt': ui_components[f'neg_prompt_{prefix}'],
-            'seed': ui_components[f'seed_{prefix}'],
-            'batch_size': ui_components[f'batch_size_{prefix}'],
-            'guidance_scale': ui_components[f'cfg_{prefix}'],
-            'num_inference_steps': ui_components[f'steps_{prefix}'],
-            'sampler': ui_components[f'sampler_{prefix}'],
-            'scheduler': ui_components[f'scheduler_{prefix}'],
-            'zero_gpu_duration': ui_components[f'zero_gpu_{prefix}'],
-            'civitai_api_key': ui_components.get(f'civitai_api_key_{prefix}'),
-            'clip_skip': ui_components[f'clip_skip_{prefix}'],
+            'positive_prompt': ui_components.get(f'prompt_{prefix}') or ui_components.get(f'{prefix}_positive_prompt'),
+            'negative_prompt': ui_components.get(f'neg_prompt_{prefix}') or ui_components.get(f'{prefix}_negative_prompt'),
+            'seed': ui_components.get(f'seed_{prefix}') or ui_components.get(f'{prefix}_seed'),
+            'batch_size': ui_components.get(f'batch_size_{prefix}') or ui_components.get(f'{prefix}_batch_size'),
+            'guidance_scale': ui_components.get(f'cfg_{prefix}') or ui_components.get(f'{prefix}_cfg'),
+            'num_inference_steps': ui_components.get(f'steps_{prefix}') or ui_components.get(f'{prefix}_steps'),
+            'sampler': ui_components.get(f'sampler_{prefix}') or ui_components.get(f'{prefix}_sampler_name'),
+            'scheduler': ui_components.get(f'scheduler_{prefix}') or ui_components.get(f'{prefix}_scheduler'),
+            'zero_gpu_duration': ui_components.get(f'zero_gpu_{prefix}'),
+
+            'clip_skip': ui_components.get(f'clip_skip_{prefix}'),
+            'guidance': ui_components.get(f'guidance_{prefix}'),
             'task_type': gr.State(task_type)
         }
         
         if task_type not in ['img2img', 'inpaint']:
-            run_inputs_map.update({'width': ui_components[f'width_{prefix}'], 'height': ui_components[f'height_{prefix}']})
+            run_inputs_map.update({
+                'width': ui_components.get(f'width_{prefix}') or ui_components.get(f'{prefix}_width'), 
+                'height': ui_components.get(f'height_{prefix}') or ui_components.get(f'{prefix}_height')
+            })
         
         task_specific_map = {
             'img2img': {'img2img_image': f'input_image_{prefix}', 'img2img_denoise': f'denoise_{prefix}'},
-            'inpaint': {'inpaint_image_dict': f'input_image_dict_{prefix}'},
-            'outpaint': {'outpaint_image': f'input_image_{prefix}', 'outpaint_left': f'outpaint_left_{prefix}', 'outpaint_top': f'outpaint_top_{prefix}', 'outpaint_right': f'outpaint_right_{prefix}', 'outpaint_bottom': f'outpaint_bottom_{prefix}'},
+            'inpaint': {'inpaint_image_dict': f'input_image_dict_{prefix}', 'grow_mask_by': f'grow_mask_by_{prefix}'},
+            'outpaint': {'outpaint_image': f'input_image_{prefix}', 'left': f'left_{prefix}', 'top': f'top_{prefix}', 'right': f'right_{prefix}', 'bottom': f'bottom_{prefix}', 'feathering': f'feathering_{prefix}'},
             'hires_fix': {'hires_image': f'input_image_{prefix}', 'hires_upscaler': f'hires_upscaler_{prefix}', 'hires_scale_by': f'hires_scale_by_{prefix}', 'hires_denoise': f'denoise_{prefix}'}
         }
         if task_type in task_specific_map:
             for key, comp_name in task_specific_map[task_type].items():
-                run_inputs_map[key] = ui_components[comp_name]
+                if comp_name in ui_components:
+                    run_inputs_map[key] = ui_components[comp_name]
         
         lora_data_components = ui_components.get(f'all_lora_components_flat_{prefix}', [])
         controlnet_data_components = ui_components.get(f'all_controlnet_components_flat_{prefix}', [])
+        diffsynth_controlnet_data_components = ui_components.get(f'all_diffsynth_controlnet_components_flat_{prefix}', [])
         ipadapter_data_components = ui_components.get(f'all_ipadapter_components_flat_{prefix}', [])
+        sd3_ipadapter_data_components = ui_components.get(f'all_sd3_ipadapter_components_flat_{prefix}', [])
+        flux1_ipadapter_data_components = ui_components.get(f'all_flux1_ipadapter_components_flat_{prefix}', [])
+        style_data_components = ui_components.get(f'all_style_components_flat_{prefix}', [])
         embedding_data_components = ui_components.get(f'all_embedding_components_flat_{prefix}', [])
         conditioning_data_components = ui_components.get(f'all_conditioning_components_flat_{prefix}', [])
+        reference_latent_data_components = ui_components.get(f'all_reference_latent_components_flat_{prefix}', [])
         
         run_inputs_map['vae_source'] = ui_components.get(f'vae_source_{prefix}')
         run_inputs_map['vae_id'] = ui_components.get(f'vae_id_{prefix}')
@@ -462,133 +726,441 @@ def attach_event_handlers(ui_components, demo):
 
         input_keys = list(run_inputs_map.keys())
         input_list_flat = [v for v in run_inputs_map.values() if v is not None]
-        input_list_flat += lora_data_components + controlnet_data_components + ipadapter_data_components + embedding_data_components + conditioning_data_components
+        all_chains = [
+            lora_data_components, controlnet_data_components, diffsynth_controlnet_data_components, ipadapter_data_components,
+            sd3_ipadapter_data_components, flux1_ipadapter_data_components, style_data_components,
+            embedding_data_components, conditioning_data_components, reference_latent_data_components
+        ]
+        for chain in all_chains:
+            if chain:
+                input_list_flat.extend(chain)
 
         def create_ui_inputs_dict(*args):
             valid_keys = [k for k in input_keys if run_inputs_map[k] is not None]
             ui_dict = dict(zip(valid_keys, args[:len(valid_keys)]))
             arg_idx = len(valid_keys)
-            
-            ui_dict['lora_data'] = list(args[arg_idx : arg_idx + len(lora_data_components)])
-            arg_idx += len(lora_data_components)
-            ui_dict['controlnet_data'] = list(args[arg_idx : arg_idx + len(controlnet_data_components)])
-            arg_idx += len(controlnet_data_components)
-            ui_dict['ipadapter_data'] = list(args[arg_idx : arg_idx + len(ipadapter_data_components)])
-            arg_idx += len(ipadapter_data_components)
-            ui_dict['embedding_data'] = list(args[arg_idx : arg_idx + len(embedding_data_components)])
-            arg_idx += len(embedding_data_components)
-            ui_dict['conditioning_data'] = list(args[arg_idx : arg_idx + len(conditioning_data_components)])
+
+            def assign_chain_data(chain_key, components_list):
+                nonlocal arg_idx
+                if components_list:
+                    ui_dict[chain_key] = list(args[arg_idx : arg_idx + len(components_list)])
+                    arg_idx += len(components_list)
+
+            assign_chain_data('lora_data', lora_data_components)
+            assign_chain_data('controlnet_data', controlnet_data_components)
+            assign_chain_data('diffsynth_controlnet_data', diffsynth_controlnet_data_components)
+            assign_chain_data('ipadapter_data', ipadapter_data_components)
+            assign_chain_data('sd3_ipadapter_chain', sd3_ipadapter_data_components)
+            assign_chain_data('flux1_ipadapter_data', flux1_ipadapter_data_components)
+            assign_chain_data('style_data', style_data_components)
+            assign_chain_data('embedding_data', embedding_data_components)
+            assign_chain_data('conditioning_data', conditioning_data_components)
+            assign_chain_data('reference_latent_data', reference_latent_data_components)
 
             return ui_dict
 
-        ui_components[f'run_{prefix}'].click(
-            fn=lambda *args, progress=gr.Progress(track_tqdm=True): generate_image_wrapper(create_ui_inputs_dict(*args), progress),
-            inputs=input_list_flat,
-            outputs=[ui_components[f'result_{prefix}']]
-        )
+        run_btn = ui_components.get(f'run_{prefix}') or ui_components.get(f'{prefix}_run_button')
+        res_gal = ui_components.get(f'result_{prefix}') or ui_components.get(f'{prefix}_output_gallery')
+        if run_btn and res_gal:
+            run_btn.click(
+                fn=lambda *args, progress=gr.Progress(track_tqdm=True): generate_image_wrapper(create_ui_inputs_dict(*args), progress),
+                inputs=input_list_flat,
+                outputs=[res_gal]
+            )
+
+    def make_update_fn(m_comp, cat_comp, cs_comp, ar_comp, width_comp, height_comp, cn_types, cn_series, cn_filepaths, diffsynth_cn_types, diffsynth_cn_series, diffsynth_cn_filepaths, ipa_preset, lora_acc, cn_acc, diffsynth_cn_acc, ipa_acc, sd3_ipa_acc, flux1_ipa_acc, style_acc, embed_acc, cond_acc, ref_latent_acc, guidance_comp, prompt_comp, neg_prompt_comp, steps_comp, cfg_comp, sampler_comp, scheduler_comp):
+        def update_fn(*args):
+            arch = args[0]
+            category = args[1]
+            current_ar = args[2] if len(args) > 2 else None
+            from core.settings import MODEL_TYPE_MAP, MODEL_MAP_CHECKPOINT, FEATURES_CONFIG, ARCHITECTURES_CONFIG, MODEL_DEFAULTS_CONFIG, ARCH_CATEGORIES_MAP
+            from utils.app_utils import get_model_generation_defaults
+            
+            if arch == "ALL":
+                valid_cats = list(set(cat for cats in ARCH_CATEGORIES_MAP.values() for cat in cats))
+            else:
+                valid_cats = ARCH_CATEGORIES_MAP.get(arch, [])
+            
+            cat_choices = ["ALL"] + sorted(valid_cats)
+            new_category = category if category in cat_choices else "ALL"
+
+            choices = []
+            for name, info in MODEL_MAP_CHECKPOINT.items():
+                m_arch = info[2]
+                m_cat = info[4] if len(info) > 4 else None
+                arch_match = (arch == "ALL" or m_arch == arch)
+                cat_match = (new_category == "ALL" or m_cat == new_category)
+                if arch_match and cat_match:
+                    choices.append(name)
+            
+            val = choices[0] if choices else None
+            
+            updates = {
+                m_comp: gr.update(choices=choices, value=val),
+                cat_comp: gr.update(choices=cat_choices, value=new_category)
+            }
+
+            m_type = MODEL_TYPE_MAP.get(val, "SDXL") if val else "SDXL"
+            
+            architectures_dict = ARCHITECTURES_CONFIG.get('architectures', {})
+            arch_model_type = architectures_dict.get(m_type, {}).get("model_type", m_type.lower().replace(" ", "").replace(".", ""))
+
+            arch_features = FEATURES_CONFIG.get(arch_model_type, FEATURES_CONFIG.get('default', {}))
+            enabled_chains = arch_features.get('enabled_chains', [])
+
+            if lora_acc: updates[lora_acc] = gr.update(visible=('lora' in enabled_chains))
+            if cn_acc: updates[cn_acc] = gr.update(visible=('controlnet' in enabled_chains))
+            if diffsynth_cn_acc: updates[diffsynth_cn_acc] = gr.update(visible=('controlnet_model_patch' in enabled_chains))
+            if ipa_acc: updates[ipa_acc] = gr.update(visible=('ipadapter' in enabled_chains))
+            if flux1_ipa_acc: updates[flux1_ipa_acc] = gr.update(visible=('flux1_ipadapter' in enabled_chains))
+            if sd3_ipa_acc: updates[sd3_ipa_acc] = gr.update(visible=('sd3_ipadapter' in enabled_chains))
+            if style_acc: updates[style_acc] = gr.update(visible=('style' in enabled_chains))
+            if embed_acc: updates[embed_acc] = gr.update(visible=('embedding' in enabled_chains))
+            if cond_acc: updates[cond_acc] = gr.update(visible=('conditioning' in enabled_chains))
+            if ref_latent_acc: updates[ref_latent_acc] = gr.update(visible=('reference_latent' in enabled_chains))
+
+            if cs_comp:
+                updates[cs_comp] = gr.update(visible=(arch_model_type == "sd15"))
+            if guidance_comp:
+                updates[guidance_comp] = gr.update(visible=(arch_model_type == "flux1"))
+            
+            if ar_comp:
+                res_key = arch_model_type
+                if res_key not in RESOLUTION_MAP:
+                    res_key = 'sdxl'
+                res_map = RESOLUTION_MAP.get(res_key, {})
+                target_ar = current_ar if current_ar in res_map else (list(res_map.keys())[0] if res_map else "1:1 (Square)")
+                updates[ar_comp] = gr.update(choices=list(res_map.keys()), value=target_ar)
+                if width_comp and height_comp and target_ar in res_map:
+                    updates[width_comp] = gr.update(value=res_map[target_ar][0])
+                    updates[height_comp] = gr.update(value=res_map[target_ar][1])
+            
+            controlnet_key = architectures_dict.get(m_type, {}).get("controlnet_key", m_type)
+            
+            all_types, default_type, series_choices, default_series, filepath = get_cn_defaults(controlnet_key)
+            for t_comp in cn_types:
+                updates[t_comp] = gr.update(choices=all_types, value=default_type)
+            for s_comp in cn_series:
+                updates[s_comp] = gr.update(choices=series_choices, value=default_series)
+            for f_comp in cn_filepaths:
+                updates[f_comp] = filepath
+                
+            diffsynth_all_types, diffsynth_default_type, diffsynth_series_choices, diffsynth_default_series, diffsynth_filepath = get_diffsynth_cn_defaults(controlnet_key)
+            for t_comp in diffsynth_cn_types:
+                updates[t_comp] = gr.update(choices=diffsynth_all_types, value=diffsynth_default_type)
+            for s_comp in diffsynth_cn_series:
+                updates[s_comp] = gr.update(choices=diffsynth_series_choices, value=diffsynth_default_series)
+            for f_comp in diffsynth_cn_filepaths:
+                updates[f_comp] = diffsynth_filepath
+            
+            if ipa_preset and (arch_model_type in ["sdxl", "sd15", "sd35"]):
+                config = load_ipadapter_config()
+                ipa_arch_key = "SDXL" if arch_model_type in ["sdxl", "sd35"] else "SD1.5"
+                std_presets = config.get("IPAdapter_presets", {}).get(ipa_arch_key, [])
+                face_presets = config.get("IPAdapter_FaceID_presets", {}).get(ipa_arch_key, [])
+                all_ipa_presets = std_presets + face_presets
+                default_ipa = all_ipa_presets[0] if all_ipa_presets else None
+                updates[ipa_preset] = gr.update(choices=all_ipa_presets, value=default_ipa)
+            
+            defaults = get_model_generation_defaults(val, arch_model_type, MODEL_DEFAULTS_CONFIG)
+            if steps_comp: updates[steps_comp] = gr.update(value=defaults.get('steps'))
+            if cfg_comp: updates[cfg_comp] = gr.update(value=defaults.get('cfg'))
+            if sampler_comp: updates[sampler_comp] = gr.update(value=defaults.get('sampler_name'))
+            if scheduler_comp: updates[scheduler_comp] = gr.update(value=defaults.get('scheduler'))
+            if prompt_comp: updates[prompt_comp] = gr.update(value=defaults.get('positive_prompt'))
+            if neg_prompt_comp: updates[neg_prompt_comp] = gr.update(value=defaults.get('negative_prompt'))
+                
+            return updates
+        return update_fn
+
+    def make_model_change_fn(cat_comp_ref, cs_comp, ar_comp, width_comp, height_comp, cn_types, cn_series, cn_filepaths, diffsynth_cn_types, diffsynth_cn_series, diffsynth_cn_filepaths, arch_comp_ref, ipa_preset, lora_acc, cn_acc, diffsynth_cn_acc, ipa_acc, sd3_ipa_acc, flux1_ipa_acc, style_acc, embed_acc, cond_acc, ref_latent_acc, guidance_comp, prompt_comp, neg_prompt_comp, steps_comp, cfg_comp, sampler_comp, scheduler_comp):
+        def change_fn(*args):
+            model_name = args[0]
+            idx = 1
+            current_arch = args[idx] if arch_comp_ref and idx < len(args) else None
+            if arch_comp_ref: idx += 1
+            current_cat = args[idx] if cat_comp_ref and idx < len(args) else None
+            if cat_comp_ref: idx += 1
+            current_ar = args[idx] if idx < len(args) else None
+            from core.settings import MODEL_TYPE_MAP, FEATURES_CONFIG, ARCHITECTURES_CONFIG, MODEL_DEFAULTS_CONFIG, ARCH_CATEGORIES_MAP, MODEL_MAP_CHECKPOINT
+            from utils.app_utils import get_model_generation_defaults
+            m_type = MODEL_TYPE_MAP.get(model_name, "SDXL")
+            
+            m_info = MODEL_MAP_CHECKPOINT.get(model_name)
+            m_cat = m_info[4] if m_info and len(m_info) > 4 else None
+            if not m_cat: m_cat = "ALL"
+            
+            updates = {}
+            target_arch = m_type
+            if arch_comp_ref:
+                if current_arch == "ALL":
+                    updates[arch_comp_ref] = gr.update()
+                    target_arch = "ALL"
+                else:
+                    updates[arch_comp_ref] = m_type
+                    
+            if cat_comp_ref:
+                if target_arch == "ALL":
+                    valid_cats = list(set(cat for cats in ARCH_CATEGORIES_MAP.values() for cat in cats))
+                else:
+                    valid_cats = ARCH_CATEGORIES_MAP.get(target_arch, [])
+                cat_choices = ["ALL"] + sorted(valid_cats)
+                
+                if current_cat == "ALL":
+                    updates[cat_comp_ref] = gr.update(choices=cat_choices)
+                else:
+                    updates[cat_comp_ref] = gr.update(choices=cat_choices, value=m_cat)
+            
+            architectures_dict = ARCHITECTURES_CONFIG.get('architectures', {})
+            arch_model_type = architectures_dict.get(m_type, {}).get("model_type", m_type.lower().replace(" ", "").replace(".", ""))
+
+            arch_features = FEATURES_CONFIG.get(arch_model_type, FEATURES_CONFIG.get('default', {}))
+            enabled_chains = arch_features.get('enabled_chains', [])
+
+            if lora_acc: updates[lora_acc] = gr.update(visible=('lora' in enabled_chains))
+            if cn_acc: updates[cn_acc] = gr.update(visible=('controlnet' in enabled_chains))
+            if diffsynth_cn_acc: updates[diffsynth_cn_acc] = gr.update(visible=('controlnet_model_patch' in enabled_chains))
+            if ipa_acc: updates[ipa_acc] = gr.update(visible=('ipadapter' in enabled_chains))
+            if flux1_ipa_acc: updates[flux1_ipa_acc] = gr.update(visible=('flux1_ipadapter' in enabled_chains))
+            if sd3_ipa_acc: updates[sd3_ipa_acc] = gr.update(visible=('sd3_ipadapter' in enabled_chains))
+            if style_acc: updates[style_acc] = gr.update(visible=('style' in enabled_chains))
+            if embed_acc: updates[embed_acc] = gr.update(visible=('embedding' in enabled_chains))
+            if cond_acc: updates[cond_acc] = gr.update(visible=('conditioning' in enabled_chains))
+            if ref_latent_acc: updates[ref_latent_acc] = gr.update(visible=('reference_latent' in enabled_chains))
+
+            if cs_comp:
+                updates[cs_comp] = gr.update(visible=(arch_model_type == "sd15"))
+            if guidance_comp:
+                updates[guidance_comp] = gr.update(visible=(arch_model_type == "flux1"))
+            
+            if ar_comp:
+                res_key = arch_model_type
+                if res_key not in RESOLUTION_MAP:
+                    res_key = 'sdxl'
+                res_map = RESOLUTION_MAP.get(res_key, {})
+                target_ar = current_ar if current_ar in res_map else (list(res_map.keys())[0] if res_map else "1:1 (Square)")
+                updates[ar_comp] = gr.update(choices=list(res_map.keys()), value=target_ar)
+                if width_comp and height_comp and target_ar in res_map:
+                    updates[width_comp] = gr.update(value=res_map[target_ar][0])
+                    updates[height_comp] = gr.update(value=res_map[target_ar][1])
+            
+            controlnet_key = architectures_dict.get(m_type, {}).get("controlnet_key", m_type)
+            
+            all_types, default_type, series_choices, default_series, filepath = get_cn_defaults(controlnet_key)
+            for t_comp in cn_types:
+                updates[t_comp] = gr.update(choices=all_types, value=default_type)
+            for s_comp in cn_series:
+                updates[s_comp] = gr.update(choices=series_choices, value=default_series)
+            for f_comp in cn_filepaths:
+                updates[f_comp] = filepath
+                
+            diffsynth_all_types, diffsynth_default_type, diffsynth_series_choices, diffsynth_default_series, diffsynth_filepath = get_diffsynth_cn_defaults(controlnet_key)
+            for t_comp in diffsynth_cn_types:
+                updates[t_comp] = gr.update(choices=diffsynth_all_types, value=diffsynth_default_type)
+            for s_comp in diffsynth_cn_series:
+                updates[s_comp] = gr.update(choices=diffsynth_series_choices, value=diffsynth_default_series)
+            for f_comp in diffsynth_cn_filepaths:
+                updates[f_comp] = diffsynth_filepath
+            
+            if ipa_preset and (arch_model_type in ["sdxl", "sd15", "sd35"]):
+                config = load_ipadapter_config()
+                ipa_arch_key = "SDXL" if arch_model_type in ["sdxl", "sd35"] else "SD1.5"
+                std_presets = config.get("IPAdapter_presets", {}).get(ipa_arch_key, [])
+                face_presets = config.get("IPAdapter_FaceID_presets", {}).get(ipa_arch_key, [])
+                all_ipa_presets = std_presets + face_presets
+                default_ipa = all_ipa_presets[0] if all_ipa_presets else None
+                updates[ipa_preset] = gr.update(choices=all_ipa_presets, value=default_ipa)
+            
+            defaults = get_model_generation_defaults(model_name, arch_model_type, MODEL_DEFAULTS_CONFIG)
+            if steps_comp: updates[steps_comp] = gr.update(value=defaults.get('steps'))
+            if cfg_comp: updates[cfg_comp] = gr.update(value=defaults.get('cfg'))
+            if sampler_comp: updates[sampler_comp] = gr.update(value=defaults.get('sampler_name'))
+            if scheduler_comp: updates[scheduler_comp] = gr.update(value=defaults.get('scheduler'))
+            if prompt_comp: updates[prompt_comp] = gr.update(value=defaults.get('positive_prompt'))
+            if neg_prompt_comp: updates[neg_prompt_comp] = gr.update(value=defaults.get('negative_prompt'))
+                
+            return updates
+        return change_fn
 
 
     for prefix, task_type in [
         ("txt2img", "txt2img"), ("img2img", "img2img"), ("inpaint", "inpaint"),
         ("outpaint", "outpaint"), ("hires_fix", "hires_fix"),
     ]:
-        if f'add_lora_button_{prefix}' in ui_components: 
-            create_lora_event_handlers(prefix)
-            lora_uploads = ui_components[f'lora_uploads_{prefix}']
-            lora_ids = ui_components[f'lora_ids_{prefix}']
-            lora_sources = ui_components[f'lora_sources_{prefix}']
-            for i in range(MAX_LORAS):
-                lora_uploads[i].upload(
-                    fn=on_lora_upload,
-                    inputs=[lora_uploads[i]],
-                    outputs=[lora_ids[i], lora_sources[i]],
-                    show_progress=False
-                )
-        if f'add_controlnet_button_{prefix}' in ui_components: create_controlnet_event_handlers(prefix)
-        if f'add_ipadapter_button_{prefix}' in ui_components: create_ipadapter_event_handlers(prefix)
-        if f'add_embedding_button_{prefix}' in ui_components: 
-            create_embedding_event_handlers(prefix)
-            if f'embeddings_uploads_{prefix}' in ui_components:
-                emb_uploads = ui_components[f'embeddings_uploads_{prefix}']
-                emb_ids = ui_components[f'embeddings_ids_{prefix}']
-                emb_sources = ui_components[f'embeddings_sources_{prefix}']
-                emb_files = ui_components[f'embeddings_files_{prefix}']
-                for i in range(MAX_EMBEDDINGS):
-                    emb_uploads[i].upload(
-                        fn=on_embedding_upload,
-                        inputs=[emb_uploads[i]],
-                        outputs=[emb_ids[i], emb_sources[i], emb_files[i]],
-                        show_progress=False
-                    )
-        if f'add_conditioning_button_{prefix}' in ui_components: create_conditioning_event_handlers(prefix)
-        if f'vae_source_{prefix}' in ui_components:
-            upload_button = ui_components.get(f'vae_upload_button_{prefix}')
-            if upload_button:
-                 upload_button.upload(
-                    fn=on_vae_upload, 
-                    inputs=[upload_button], 
-                    outputs=[
-                        ui_components[f'vae_id_{prefix}'], 
-                        ui_components[f'vae_source_{prefix}'], 
-                        ui_components[f'vae_file_{prefix}']
-                    ]
+
+        arch_comp = ui_components.get(f'model_arch_{prefix}')
+        cat_comp = ui_components.get(f'model_cat_{prefix}')
+        model_comp = ui_components.get(f'base_model_{prefix}')
+        clip_skip_comp = ui_components.get(f'clip_skip_{prefix}') or ui_components.get(f'{prefix}_clip_skip')
+        guidance_comp = ui_components.get(f'guidance_{prefix}') or ui_components.get(f'{prefix}_guidance')
+        aspect_ratio_comp = ui_components.get(f'aspect_ratio_{prefix}') or ui_components.get(f'{prefix}_aspect_ratio_dropdown')
+        width_comp = ui_components.get(f'width_{prefix}') or ui_components.get(f'{prefix}_width')
+        height_comp = ui_components.get(f'height_{prefix}') or ui_components.get(f'{prefix}_height')
+        
+        cn_types_list = ui_components.get(f'controlnet_types_{prefix}', [])
+        cn_series_list = ui_components.get(f'controlnet_series_{prefix}', [])
+        cn_filepaths_list = ui_components.get(f'controlnet_filepaths_{prefix}', [])
+        
+        diffsynth_cn_types_list = ui_components.get(f'diffsynth_controlnet_types_{prefix}', [])
+        diffsynth_cn_series_list = ui_components.get(f'diffsynth_controlnet_series_{prefix}', [])
+        diffsynth_cn_filepaths_list = ui_components.get(f'diffsynth_controlnet_filepaths_{prefix}', [])
+        
+        lora_accordion = ui_components.get(f'lora_accordion_{prefix}')
+        cn_accordion = ui_components.get(f'controlnet_accordion_{prefix}')
+        diffsynth_cn_accordion = ui_components.get(f'diffsynth_controlnet_accordion_{prefix}')
+        ipa_accordion = ui_components.get(f'ipadapter_accordion_{prefix}')
+        sd3_ipa_accordion = ui_components.get(f'sd3_ipadapter_accordion_{prefix}')
+        flux1_ipa_accordion = ui_components.get(f'flux1_ipadapter_accordion_{prefix}')
+        style_accordion = ui_components.get(f'style_accordion_{prefix}')
+        embedding_accordion = ui_components.get(f'embedding_accordion_{prefix}')
+        conditioning_accordion = ui_components.get(f'conditioning_accordion_{prefix}')
+        ref_latent_accordion = ui_components.get(f'reference_latent_accordion_{prefix}')
+        
+        ipa_preset_list = ui_components.get(f'ipadapter_final_preset_{prefix}')
+        
+        prompt_comp = ui_components.get(f'prompt_{prefix}') or ui_components.get(f'{prefix}_positive_prompt')
+        neg_prompt_comp = ui_components.get(f'neg_prompt_{prefix}') or ui_components.get(f'{prefix}_negative_prompt')
+        steps_comp = ui_components.get(f'steps_{prefix}') or ui_components.get(f'{prefix}_steps')
+        cfg_comp = ui_components.get(f'cfg_{prefix}') or ui_components.get(f'{prefix}_cfg')
+        sampler_comp = ui_components.get(f'sampler_{prefix}') or ui_components.get(f'{prefix}_sampler_name')
+        scheduler_comp = ui_components.get(f'scheduler_{prefix}') or ui_components.get(f'{prefix}_scheduler')
+        
+        extra_comps = [prompt_comp, neg_prompt_comp, steps_comp, cfg_comp, sampler_comp, scheduler_comp, width_comp, height_comp]
+        valid_extra_comps = [c for c in extra_comps if c is not None]
+        
+        if arch_comp and cat_comp and model_comp:
+            outputs = [model_comp, cat_comp]
+            if clip_skip_comp: outputs.append(clip_skip_comp)
+            if guidance_comp: outputs.append(guidance_comp)
+            if aspect_ratio_comp: outputs.append(aspect_ratio_comp)
+            outputs.extend(cn_types_list + cn_series_list + cn_filepaths_list)
+            outputs.extend(diffsynth_cn_types_list + diffsynth_cn_series_list + diffsynth_cn_filepaths_list)
+            if lora_accordion: outputs.append(lora_accordion)
+            if cn_accordion: outputs.append(cn_accordion)
+            if diffsynth_cn_accordion: outputs.append(diffsynth_cn_accordion)
+            if ipa_accordion: outputs.append(ipa_accordion)
+            if sd3_ipa_accordion: outputs.append(sd3_ipa_accordion)
+            if flux1_ipa_accordion: outputs.append(flux1_ipa_accordion)
+            if style_accordion: outputs.append(style_accordion)
+            if embedding_accordion: outputs.append(embedding_accordion)
+            if conditioning_accordion: outputs.append(conditioning_accordion)
+            if ref_latent_accordion: outputs.append(ref_latent_accordion)
+            if ipa_preset_list: outputs.append(ipa_preset_list)
+            
+            outputs.extend(valid_extra_comps)
+            
+            update_fn = make_update_fn(
+                model_comp, cat_comp, clip_skip_comp, aspect_ratio_comp, width_comp, height_comp, 
+                cn_types_list, cn_series_list, cn_filepaths_list, 
+                diffsynth_cn_types_list, diffsynth_cn_series_list, diffsynth_cn_filepaths_list, 
+                ipa_preset_list, lora_accordion, cn_accordion, diffsynth_cn_accordion, ipa_accordion, sd3_ipa_accordion, flux1_ipa_accordion, style_accordion, embedding_accordion, conditioning_accordion,
+                ref_latent_accordion, guidance_comp, prompt_comp, neg_prompt_comp, steps_comp, cfg_comp, sampler_comp, scheduler_comp
+            )
+            inputs = [arch_comp, cat_comp]
+            if aspect_ratio_comp:
+                inputs.append(aspect_ratio_comp)
+            arch_comp.change(fn=update_fn, inputs=inputs, outputs=outputs)
+            cat_comp.change(fn=update_fn, inputs=inputs, outputs=outputs)
+
+        if model_comp:
+            outputs2 = []
+            if arch_comp: outputs2.append(arch_comp)
+            if cat_comp: outputs2.append(cat_comp)
+            if clip_skip_comp: outputs2.append(clip_skip_comp)
+            if guidance_comp: outputs2.append(guidance_comp)
+            if aspect_ratio_comp: outputs2.append(aspect_ratio_comp)
+            outputs2.extend(cn_types_list + cn_series_list + cn_filepaths_list)
+            outputs2.extend(diffsynth_cn_types_list + diffsynth_cn_series_list + diffsynth_cn_filepaths_list)
+            if lora_accordion: outputs2.append(lora_accordion)
+            if cn_accordion: outputs2.append(cn_accordion)
+            if diffsynth_cn_accordion: outputs2.append(diffsynth_cn_accordion)
+            if ipa_accordion: outputs2.append(ipa_accordion)
+            if sd3_ipa_accordion: outputs2.append(sd3_ipa_accordion)
+            if flux1_ipa_accordion: outputs2.append(flux1_ipa_accordion)
+            if style_accordion: outputs2.append(style_accordion)
+            if embedding_accordion: outputs2.append(embedding_accordion)
+            if conditioning_accordion: outputs2.append(conditioning_accordion)
+            if ref_latent_accordion: outputs2.append(ref_latent_accordion)
+            if ipa_preset_list: outputs2.append(ipa_preset_list)
+            
+            outputs2.extend(valid_extra_comps)
+            
+            if outputs2:
+                inputs2 = [model_comp]
+                if arch_comp: inputs2.append(arch_comp)
+                if cat_comp: inputs2.append(cat_comp)
+                if aspect_ratio_comp: inputs2.append(aspect_ratio_comp)
+                change_fn = make_model_change_fn(
+                    cat_comp, clip_skip_comp, aspect_ratio_comp, width_comp, height_comp, 
+                    cn_types_list, cn_series_list, cn_filepaths_list, 
+                    diffsynth_cn_types_list, diffsynth_cn_series_list, diffsynth_cn_filepaths_list, 
+                    arch_comp, ipa_preset_list, lora_accordion, cn_accordion, diffsynth_cn_accordion, ipa_accordion, sd3_ipa_accordion, flux1_ipa_accordion, style_accordion, embedding_accordion, conditioning_accordion,
+                    ref_latent_accordion, guidance_comp, prompt_comp, neg_prompt_comp, steps_comp, cfg_comp, sampler_comp, scheduler_comp
                 )
+                model_comp.change(fn=change_fn, inputs=inputs2, outputs=outputs2)
 
+        create_lora_event_handlers(prefix)
+        create_controlnet_event_handlers(prefix)
+        create_diffsynth_controlnet_event_handlers(prefix)
+        create_ipadapter_event_handlers(prefix)
+        create_embedding_event_handlers(prefix)
+        create_conditioning_event_handlers(prefix)
+        create_flux1_ipadapter_event_handlers(prefix)
+        create_style_event_handlers(prefix)
+        create_reference_latent_event_handlers(prefix)
         create_run_event(prefix, task_type)
 
-    def on_aspect_ratio_change(ratio_key, model_display_name):
-        model_type = MODEL_TYPE_MAP.get(model_display_name, 'sdxl').lower()
-        res_map = RESOLUTION_MAP.get(model_type, RESOLUTION_MAP.get("sdxl", {}))
-        w, h = res_map.get(ratio_key, (1024, 1024))
-        return w, h
 
-    for prefix in ["txt2img", "img2img", "inpaint", "outpaint", "hires_fix"]:
-        if f'aspect_ratio_{prefix}' in ui_components:
-            aspect_ratio_dropdown = ui_components[f'aspect_ratio_{prefix}']
-            width_component = ui_components[f'width_{prefix}']
-            height_component = ui_components[f'height_{prefix}']
-            model_dropdown = ui_components[f'base_model_{prefix}']
-            aspect_ratio_dropdown.change(fn=on_aspect_ratio_change, inputs=[aspect_ratio_dropdown, model_dropdown], outputs=[width_component, height_component], show_progress=False)
-            
     if 'view_mode_inpaint' in ui_components:
         def toggle_inpaint_fullscreen_view(view_mode):
             is_fullscreen = (view_mode == "Fullscreen View")
             other_elements_visible = not is_fullscreen
             editor_height = 800 if is_fullscreen else 272
-            return {
-                ui_components['model_and_run_row_inpaint']: gr.update(visible=other_elements_visible),
+            
+            updates = {
                 ui_components['prompts_column_inpaint']: gr.update(visible=other_elements_visible),
                 ui_components['params_and_gallery_row_inpaint']: gr.update(visible=other_elements_visible),
                 ui_components['accordion_wrapper_inpaint']: gr.update(visible=other_elements_visible),
                 ui_components['input_image_dict_inpaint']: gr.update(height=editor_height),
             }
+            
+            model_and_run_rows = ui_components.get('model_and_run_row_inpaint', [])
+            for row in model_and_run_rows:
+                updates[row] = gr.update(visible=other_elements_visible)
+            
+            return updates
         
-        output_components = [
-            ui_components['model_and_run_row_inpaint'], ui_components['prompts_column_inpaint'],
-            ui_components['params_and_gallery_row_inpaint'], ui_components['accordion_wrapper_inpaint'],
+        output_components = []
+        model_and_run_rows = ui_components.get('model_and_run_row_inpaint', [])
+        if isinstance(model_and_run_rows, list):
+            output_components.extend(model_and_run_rows)
+        else:
+            output_components.append(model_and_run_rows)
+        
+        output_components.extend([
+            ui_components['prompts_column_inpaint'],
+            ui_components['params_and_gallery_row_inpaint'], 
+            ui_components['accordion_wrapper_inpaint'],
             ui_components['input_image_dict_inpaint']
-        ]
-        ui_components['view_mode_inpaint'].change(fn=toggle_inpaint_fullscreen_view, inputs=[ui_components['view_mode_inpaint']], outputs=output_components, show_progress=False)
+        ])
+        
+        ui_components['view_mode_inpaint'].change(
+            fn=toggle_inpaint_fullscreen_view, 
+            inputs=[ui_components['view_mode_inpaint']], 
+            outputs=output_components, 
+            show_progress=False
+        )
 
     def initialize_all_cn_dropdowns():
-        cn_config = load_controlnet_config()
-        if not cn_config: return {}
-
-        all_types = sorted(list(set(t for model in cn_config for t in model.get("Type", []))))
-        default_type = all_types[0] if all_types else None
+        from core.settings import MODEL_TYPE_MAP, MODEL_MAP_CHECKPOINT, ARCHITECTURES_CONFIG
+        default_model_name = list(MODEL_MAP_CHECKPOINT.keys())[0] if MODEL_MAP_CHECKPOINT else None
+        default_m_type = MODEL_TYPE_MAP.get(default_model_name, "SDXL") if default_model_name else "SDXL"
+        architectures_dict = ARCHITECTURES_CONFIG.get('architectures', {})
+        controlnet_key = architectures_dict.get(default_m_type, {}).get("controlnet_key", default_m_type)
         
-        series_choices = []
-        if default_type:
-            series_choices = sorted(list(set(model.get("Series", "Default") for model in cn_config if default_type in model.get("Type", []))))
-        default_series = series_choices[0] if series_choices else None
+        all_types, default_type, series_choices, default_series, filepath = get_cn_defaults(controlnet_key)
+        diffsynth_all_types, diffsynth_default_type, diffsynth_series_choices, diffsynth_default_series, diffsynth_filepath = get_diffsynth_cn_defaults(controlnet_key)
         
-        filepath = "None"
-        if default_series and default_type:
-            for model in cn_config:
-                if model.get("Series") == default_series and default_type in model.get("Type", []):
-                    filepath = model.get("Filepath")
-                    break
-
         updates = {}
         for prefix in ["txt2img", "img2img", "inpaint", "outpaint", "hires_fix"]:
             if f'controlnet_types_{prefix}' in ui_components:
@@ -598,22 +1170,30 @@ def attach_event_handlers(ui_components, demo):
                     updates[series_dd] = gr.update(choices=series_choices, value=default_series)
                 for filepath_state in ui_components[f'controlnet_filepaths_{prefix}']:
                     updates[filepath_state] = filepath
+            
+            if f'diffsynth_controlnet_types_{prefix}' in ui_components:
+                for type_dd in ui_components[f'diffsynth_controlnet_types_{prefix}']:
+                    updates[type_dd] = gr.update(choices=diffsynth_all_types, value=diffsynth_default_type)
+                for series_dd in ui_components[f'diffsynth_controlnet_series_{prefix}']:
+                    updates[series_dd] = gr.update(choices=diffsynth_series_choices, value=diffsynth_default_series)
+                for filepath_state in ui_components[f'diffsynth_controlnet_filepaths_{prefix}']:
+                    updates[filepath_state] = diffsynth_filepath
+                    
         return updates
 
     def initialize_all_ipa_dropdowns():
         config = load_ipadapter_config()
-        if not config or not isinstance(config, list): return {}
-        
-        unified_presets = []
-        faceid_presets = []
-        for preset_info in config:
-            name = preset_info.get("preset_name")
-            if not name:
-                continue
-            if "FACEID" in name or "FACE" in name:
-                faceid_presets.append(name)
-            else:
-                unified_presets.append(name)
+        if not config: return {}
+        
+        from core.settings import MODEL_TYPE_MAP, MODEL_MAP_CHECKPOINT, ARCHITECTURES_CONFIG
+        default_model_name = list(MODEL_MAP_CHECKPOINT.keys())[0] if MODEL_MAP_CHECKPOINT else None
+        default_m_type = MODEL_TYPE_MAP.get(default_model_name, "SDXL") if default_model_name else "SDXL"
+        architectures_dict = ARCHITECTURES_CONFIG.get('architectures', {})
+        arch_model_type = architectures_dict.get(default_m_type, {}).get("model_type", default_m_type.lower().replace(" ", "").replace(".", ""))
+        ipa_arch_key = "SDXL" if arch_model_type in ["sdxl", "sd35"] else "SD1.5"
+
+        unified_presets = config.get("IPAdapter_presets", {}).get(ipa_arch_key, [])
+        faceid_presets = config.get("IPAdapter_FaceID_presets", {}).get(ipa_arch_key, [])
         
         all_presets = unified_presets + faceid_presets
         default_preset = all_presets[0] if all_presets else None
@@ -636,19 +1216,6 @@ def attach_event_handlers(ui_components, demo):
         
         all_updates = {**cn_updates, **ipa_updates}
         
-        default_preprocessor = "Canny Edge" 
-        model_update = update_preprocessor_models_dropdown(default_preprocessor)
-        all_updates[ui_components["preprocessor_model_cn"]] = model_update
-        
-        settings_outputs = update_preprocessor_settings_ui(default_preprocessor)
-        dynamic_outputs = ui_components["cn_sliders"] + ui_components["cn_dropdowns"] + ui_components["cn_checkboxes"]
-        for i, comp in enumerate(dynamic_outputs):
-            all_updates[comp] = settings_outputs[i]
-
-        run_button_update, zero_gpu_update = update_run_button_for_cpu(default_preprocessor)
-        all_updates[ui_components["run_cn"]] = run_button_update
-        all_updates[ui_components["zero_gpu_cn"]] = zero_gpu_update
-        
         return all_updates
     
     all_load_outputs = []
@@ -657,22 +1224,35 @@ def attach_event_handlers(ui_components, demo):
             all_load_outputs.extend(ui_components[f'controlnet_types_{prefix}'])
             all_load_outputs.extend(ui_components[f'controlnet_series_{prefix}'])
             all_load_outputs.extend(ui_components[f'controlnet_filepaths_{prefix}'])
+        if f'diffsynth_controlnet_types_{prefix}' in ui_components:
+            all_load_outputs.extend(ui_components[f'diffsynth_controlnet_types_{prefix}'])
+            all_load_outputs.extend(ui_components[f'diffsynth_controlnet_series_{prefix}'])
+            all_load_outputs.extend(ui_components[f'diffsynth_controlnet_filepaths_{prefix}'])
         if f'ipadapter_final_preset_{prefix}' in ui_components:
             all_load_outputs.extend(ui_components[f'ipadapter_lora_strengths_{prefix}'])
             all_load_outputs.append(ui_components[f'ipadapter_final_preset_{prefix}'])
             all_load_outputs.append(ui_components[f'ipadapter_final_lora_strength_{prefix}'])
 
-    all_load_outputs.extend([
-        ui_components["preprocessor_model_cn"],
-        *ui_components["cn_sliders"],
-        *ui_components["cn_dropdowns"],
-        *ui_components["cn_checkboxes"],
-        ui_components["run_cn"],
-        ui_components["zero_gpu_cn"]
-    ])
-    
     if all_load_outputs:
         demo.load(
             fn=run_on_load,
             outputs=all_load_outputs
-        )
+        )
+
+    def on_aspect_ratio_change(ratio_key, model_display_name):
+        from core.settings import MODEL_TYPE_MAP, ARCHITECTURES_CONFIG
+        m_type = MODEL_TYPE_MAP.get(model_display_name, 'SDXL')
+        architectures_dict = ARCHITECTURES_CONFIG.get('architectures', {})
+        arch_model_type = architectures_dict.get(m_type, {}).get("model_type", m_type.lower().replace(" ", "").replace(".", ""))
+        
+        res_map = RESOLUTION_MAP.get(arch_model_type, RESOLUTION_MAP.get("sdxl", {}))
+        w, h = res_map.get(ratio_key, (1024, 1024))
+        return w, h
+
+    for prefix in ["txt2img", "img2img", "inpaint", "outpaint", "hires_fix"]:
+        aspect_ratio_dropdown = ui_components.get(f'aspect_ratio_{prefix}') or ui_components.get(f'{prefix}_aspect_ratio_dropdown')
+        width_component = ui_components.get(f'width_{prefix}') or ui_components.get(f'{prefix}_width')
+        height_component = ui_components.get(f'height_{prefix}') or ui_components.get(f'{prefix}_height')
+        model_dropdown = ui_components.get(f'base_model_{prefix}')
+        if aspect_ratio_dropdown and width_component and height_component and model_dropdown:
+            aspect_ratio_dropdown.change(fn=on_aspect_ratio_change, inputs=[aspect_ratio_dropdown, model_dropdown], outputs=[width_component, height_component], show_progress=False)

ui/layout.py

@@ -6,83 +6,40 @@ from .shared import txt2img_ui, img2img_ui, inpaint_ui, outpaint_ui, hires_fix_u
 
 MAX_DYNAMIC_CONTROLS = 10
 
-def get_preprocessor_choices():
-    from nodes import NODE_DISPLAY_NAME_MAPPINGS
-    
-    preprocessor_names = [
-        display_name for class_name, display_name in NODE_DISPLAY_NAME_MAPPINGS.items()
-        if "Preprocessor" in class_name or "Segmentor" in class_name or 
-           "Estimator" in class_name or "Detector" in class_name
-    ]
-    return sorted(list(set(preprocessor_names)))
-
-
 def build_ui(event_handler_function):
     ui_components = {}
 
     with gr.Blocks() as demo:
-        gr.Markdown("# ImageGen - SDXL")
+        gr.Markdown("# ImageGen")
         gr.Markdown(
-            "This demo is a streamlined version of the [Comfy web UI](https://github.com/RioShiina47/comfy-webui)'s ImgGen functionality. "
+            "This demo is a streamlined version of the [Comfy web UI](https://github.com/RioShiina47/comfy-webui)'s ImageGen functionality. "
             "Other versions are also available: "
             "[FLUX.2](https://huggingface.co/spaces/RioShiina/ImageGen-FLUX.2), "
             "[Z-Image](https://huggingface.co/spaces/RioShiina/ImageGen-Z-Image), "
             "[Qwen-Image](https://huggingface.co/spaces/RioShiina/ImageGen-Qwen-Image), "
-            "[Anima](https://huggingface.co/spaces/RioShiina/ImageGen-Anima), "
+            "[Anime](https://huggingface.co/spaces/RioShiina/ImageGen-Anime), "
             "[Illustrious](https://huggingface.co/spaces/RioShiina/ImageGen-Illustrious), "
             "[NoobAI](https://huggingface.co/spaces/RioShiina/ImageGen-NoobAI), "
             "[Pony](https://huggingface.co/spaces/RioShiina/ImageGen-Pony)"
         )
         with gr.Tabs(elem_id="tabs_container") as tabs:
-            with gr.TabItem("SDXL", id=0):
-                with gr.Tabs(elem_id="image_gen_tabs") as image_gen_tabs:
-                    with gr.TabItem("Txt2Img", id=0):
-                        ui_components.update(txt2img_ui.create_ui())
-                    
-                    with gr.TabItem("Img2Img", id=1):
-                        ui_components.update(img2img_ui.create_ui())
+            with gr.TabItem("Txt2Img", id=0):
+                ui_components.update(txt2img_ui.create_ui())
+            
+            with gr.TabItem("Img2Img", id=1):
+                ui_components.update(img2img_ui.create_ui())
 
-                    with gr.TabItem("Inpaint", id=2):
-                        ui_components.update(inpaint_ui.create_ui())
+            with gr.TabItem("Inpaint", id=2):
+                ui_components.update(inpaint_ui.create_ui())
 
-                    with gr.TabItem("Outpaint", id=3):
-                        ui_components.update(outpaint_ui.create_ui())
+            with gr.TabItem("Outpaint", id=3):
+                ui_components.update(outpaint_ui.create_ui())
 
-                    with gr.TabItem("Hires. Fix", id=4):
-                        ui_components.update(hires_fix_ui.create_ui())
-                
-                ui_components['image_gen_tabs'] = image_gen_tabs
+            with gr.TabItem("Hires. Fix", id=4):
+                ui_components.update(hires_fix_ui.create_ui())
 
-            with gr.TabItem("Controlnet Preprocessors", id=1):
-                gr.Markdown("## ControlNet Auxiliary Preprocessors")
-                gr.Markdown("Powered by [Fannovel16/comfyui_controlnet_aux](https://github.com/Fannovel16/comfyui_controlnet_aux).")
-                gr.Markdown("Upload an image or video to process it with a ControlNet preprocessor.")
-                with gr.Row():
-                    with gr.Column(scale=1):
-                        cn_input_type = gr.Radio(["Image", "Video"], label="Input Type", value="Image")
-                        cn_image_input = gr.Image(type="pil", label="Input Image", visible=True, height=384)
-                        cn_video_input = gr.Video(label="Input Video", visible=False)
-                        preprocessor_cn = gr.Dropdown(label="Preprocessor", choices=get_preprocessor_choices(), value="Canny Edge")
-                        preprocessor_model_cn = gr.Dropdown(label="Preprocessor Model", choices=[], value=None, visible=False)
-                        with gr.Column() as preprocessor_settings_ui:
-                            cn_sliders, cn_dropdowns, cn_checkboxes = [], [], []
-                            for i in range(MAX_DYNAMIC_CONTROLS):
-                                cn_sliders.append(gr.Slider(visible=False, label=f"dyn_slider_{i}"))
-                                cn_dropdowns.append(gr.Dropdown(visible=False, label=f"dyn_dropdown_{i}"))
-                                cn_checkboxes.append(gr.Checkbox(visible=False, label=f"dyn_checkbox_{i}"))
-                        run_cn = gr.Button("Run Preprocessor", variant="primary")
-                    with gr.Column(scale=1):
-                        output_gallery_cn = gr.Gallery(label="Output", show_label=False, object_fit="contain", height=512)
-                        zero_gpu_cn = gr.Number(label="ZeroGPU Duration (s)", value=None, placeholder="Default: 60s, Max: 120s", info="Optional")
-                ui_components.update({
-                    "cn_input_type": cn_input_type, "cn_image_input": cn_image_input, "cn_video_input": cn_video_input,
-                    "preprocessor_cn": preprocessor_cn, "preprocessor_model_cn": preprocessor_model_cn, "run_cn": run_cn,
-                    "zero_gpu_cn": zero_gpu_cn, "output_gallery_cn": output_gallery_cn,
-                    "preprocessor_settings_ui": preprocessor_settings_ui, "cn_sliders": cn_sliders,
-                    "cn_dropdowns": cn_dropdowns, "cn_checkboxes": cn_checkboxes
-                })
-            
             ui_components["tabs"] = tabs
+            ui_components["image_gen_tabs"] = tabs
         
         gr.Markdown("<div style='text-align: center; margin-top: 20px;'>Made by RioShiina with ❤️<br><a href='https://github.com/RioShiina47' target='_blank'>GitHub</a> | <a href='https://huggingface.co/RioShiina' target='_blank'>Hugging Face</a> | <a href='https://civitai.com/user/RioShiina' target='_blank'>Civitai</a></div>")
         

ui/shared/hires_fix_ui.py

@@ -4,7 +4,10 @@ from comfy_integration.nodes import SAMPLER_CHOICES, SCHEDULER_CHOICES
 from .ui_components import (
     create_lora_settings_ui,
     create_controlnet_ui, create_ipadapter_ui, create_embedding_ui,
-    create_conditioning_ui, create_vae_override_ui, create_api_key_ui
+    create_conditioning_ui, create_vae_override_ui,
+    create_model_architecture_filter_ui, create_category_filter_ui,
+    create_sd3_ipadapter_ui, create_flux1_ipadapter_ui, create_style_ui,
+    create_reference_latent_ui
 )
 
 def create_ui():
@@ -12,7 +15,10 @@ def create_ui():
     components = {}
     
     with gr.Column():
+        components.update(create_model_architecture_filter_ui(prefix))
+        
         with gr.Row():
+            components.update(create_category_filter_ui(prefix))
             components[f'base_model_{prefix}'] = gr.Dropdown(
                 label="Base Model", 
                 choices=list(MODEL_MAP_CHECKPOINT.keys()), 
@@ -26,8 +32,8 @@ def create_ui():
             with gr.Column(scale=1):
                 components[f'input_image_{prefix}'] = gr.Image(type="pil", label="Input Image", height=255)
             with gr.Column(scale=2):
-                components[f'prompt_{prefix}'] = gr.Text(label="Prompt", lines=3, placeholder="Describe the final image...")
-                components[f'neg_prompt_{prefix}'] = gr.Text(label="Negative prompt", lines=3, value="monochrome, (low quality, worst quality:1.2), 3d, watermark, signature, ugly, poorly drawn,")
+                components[f'prompt_{prefix}'] = gr.Text(label="Prompt", lines=3)
+                components[f'neg_prompt_{prefix}'] = gr.Text(label="Negative prompt", lines=3)
 
         with gr.Row():
             with gr.Column(scale=1):
@@ -54,21 +60,26 @@ def create_ui():
                     components[f'seed_{prefix}'] = gr.Number(label="Seed (-1 for random)", value=-1, precision=0)
                     components[f'batch_size_{prefix}'] = gr.Slider(label="Batch Size", minimum=1, maximum=16, step=1, value=1)
                 with gr.Row():
+                    components[f'clip_skip_{prefix}'] = gr.Slider(label="Clip Skip", minimum=1, maximum=2, step=1, value=1, visible=False, interactive=True)
+                    components[f'guidance_{prefix}'] = gr.Slider(label="Guidance (FLUX)", minimum=1.0, maximum=10.0, step=0.1, value=3.5, visible=False, interactive=True)
                     components[f'zero_gpu_{prefix}'] = gr.Number(label="ZeroGPU Duration (s)", value=None, placeholder="Default: 60s, Max: 120s", info="Optional: Set how long to reserve the GPU.")
 
-                components[f'clip_skip_{prefix}'] = gr.State(value=1)
                 components[f'width_{prefix}'] = gr.State(value=512)
                 components[f'height_{prefix}'] = gr.State(value=512)
 
             with gr.Column(scale=1):
                 components[f'result_{prefix}'] = gr.Gallery(label="Result", show_label=False, columns=1, object_fit="contain", height=610)
 
-        components.update(create_api_key_ui(prefix))
+
         components.update(create_lora_settings_ui(prefix))
         components.update(create_controlnet_ui(prefix))
         components.update(create_ipadapter_ui(prefix))
+        components.update(create_flux1_ipadapter_ui(prefix))
+        components.update(create_sd3_ipadapter_ui(prefix))
+        components.update(create_style_ui(prefix))
         components.update(create_embedding_ui(prefix))
         components.update(create_conditioning_ui(prefix))
+        components.update(create_reference_latent_ui(prefix))
         components.update(create_vae_override_ui(prefix))
             
     return components