Update app.py

app.py

@@ -1,113 +1,120 @@
 
+#!/usr/bin/env python
 
 import os
 import random
 import uuid
-import json
 
 import gradio as gr
 import numpy as np
 from PIL import Image
 import spaces
 import torch
-from diffusers import StableDiffusionXLPipeline, DPMSolverSinglestepScheduler
+from diffusers import DiffusionPipeline
 
-# Use environment variables for flexibility
-MODEL_ID = os.getenv("MODEL_ID", "sd-community/sdxl-flash")
-MAX_IMAGE_SIZE = int(os.getenv("MAX_IMAGE_SIZE", "4096"))
+DESCRIPTION = """# Playground v2.5"""
+if not torch.cuda.is_available():
+    DESCRIPTION += "\n<p>Running on CPU 🥶 This demo may not work on CPU.</p>"
+
+MAX_SEED = np.iinfo(np.int32).max
+CACHE_EXAMPLES = torch.cuda.is_available() and os.getenv("CACHE_EXAMPLES", "1") == "1"
+MAX_IMAGE_SIZE = int(os.getenv("MAX_IMAGE_SIZE", "1536"))
 USE_TORCH_COMPILE = os.getenv("USE_TORCH_COMPILE", "0") == "1"
 ENABLE_CPU_OFFLOAD = os.getenv("ENABLE_CPU_OFFLOAD", "0") == "1"
-BATCH_SIZE = int(os.getenv("BATCH_SIZE", "1"))  # Allow generating multiple images at once
 
-# Determine device and load model outside of function for efficiency
 device = torch.device("cuda:0" if torch.cuda.is_available() else "cpu")
-pipe = StableDiffusionXLPipeline.from_single_file(
-    "https://huggingface.co/kadirnar/Black-Hole/blob/main/tachyon.safetensors",
-    torch_dtype=torch.float16 if torch.cuda.is_available() else torch.float32,
-    use_safetensors=True,
-    add_watermarker=False,
-).to(device)
-pipe.scheduler = DPMSolverSinglestepScheduler(use_karras_sigmas=True).from_config(pipe.scheduler.config)
-
-# Torch compile for potential speedup (experimental)
-if USE_TORCH_COMPILE:
-    pipe.compile()
-
-# CPU offloading for larger RAM capacity (experimental)
-if ENABLE_CPU_OFFLOAD:
-    pipe.enable_model_cpu_offload()
 
-MAX_SEED = np.iinfo(np.int32).max
+NUM_IMAGES_PER_PROMPT = 1
+
+if torch.cuda.is_available():
+    pipe = DiffusionPipeline.from_pretrained(
+        "playgroundai/playground-v2.5-1024px-aesthetic",
+        torch_dtype=torch.float16,
+        use_safetensors=True,
+        add_watermarker=False,
+        variant="fp16"
+    )
+    if ENABLE_CPU_OFFLOAD:
+        pipe.enable_model_cpu_offload()
+    else:
+        pipe.to(device)    
+        print("Loaded on Device!")
+    
+    if USE_TORCH_COMPILE:
+        pipe.unet = torch.compile(pipe.unet, mode="reduce-overhead", fullgraph=True)
+        print("Model Compiled!")
+
+
+def save_image(img):
+    unique_name = str(uuid.uuid4()) + ".png"
+    img.save(unique_name)
+    return unique_name
+
 
 def randomize_seed_fn(seed: int, randomize_seed: bool) -> int:
     if randomize_seed:
         seed = random.randint(0, MAX_SEED)
     return seed
 
-@spaces.GPU()
+
+@spaces.GPU(enable_queue=True)
 def generate(
     prompt: str,
     negative_prompt: str = "",
     use_negative_prompt: bool = False,
-    seed: int = 1,
+    seed: int = 0,
     width: int = 1024,
     height: int = 1024,
     guidance_scale: float = 3,
-    num_inference_steps: int = 30,
     randomize_seed: bool = False,
-    use_resolution_binning: bool = True, 
-    num_images: int = 1,  # Number of images to generate
+    use_resolution_binning: bool = True,
     progress=gr.Progress(track_tqdm=True),
 ):
+    pipe.to(device)
     seed = int(randomize_seed_fn(seed, randomize_seed))
-    generator = torch.Generator(device=device).manual_seed(seed)
-
-    # Improved options handling
-    options = {
-        "prompt": [prompt] * num_images,
-        "negative_prompt": [negative_prompt] * num_images if use_negative_prompt else None,
-        "width": width,
-        "height": height,
-        "guidance_scale": guidance_scale,
-        "num_inference_steps": num_inference_steps,
-        "generator": generator,
-        "output_type": "pil",
-    }
-
-    # Use resolution binning for faster generation with less VRAM usage
-    if use_resolution_binning:
-        options["use_resolution_binning"] = True
-
-    # Generate images potentially in batches
-    images = []
-    for i in range(0, num_images, BATCH_SIZE):
-        batch_options = options.copy()
-        batch_options["prompt"] = options["prompt"][i:i+BATCH_SIZE]
-        if "negative_prompt" in batch_options:
-            batch_options["negative_prompt"] = options["negative_prompt"][i:i+BATCH_SIZE]
-        images.extend(pipe(**batch_options).images)
-
-    return images
+    generator = torch.Generator().manual_seed(seed)
+
+    if not use_negative_prompt:
+        negative_prompt = None  # type: ignore
+    
+    images = pipe(
+        prompt=prompt,
+        negative_prompt=negative_prompt,
+        width=width,
+        height=height,
+        guidance_scale=guidance_scale,
+        num_inference_steps=25,
+        generator=generator,
+        num_images_per_prompt=NUM_IMAGES_PER_PROMPT,
+        use_resolution_binning=use_resolution_binning,
+        output_type="pil",
+    ).images
+
+    image_paths = [save_image(img) for img in images]
+    print(image_paths)
+    return image_paths, seed
+
 
 examples = [
+    "neon holography crystal cat",
     "a cat eating a piece of cheese",
-    "a ROBOT riding a BLUE horse on Mars, photorealistic, 4k",
-    "Ironman VS Hulk, ultrarealistic",
-    "Astronaut in a jungle, cold color palette, oil pastel, detailed, 8k",
-    "An alien holding a sign board containing the word 'Flash', futuristic, neonpunk",
-    "Kids going to school, Anime style"
+    "an astronaut riding a horse in space",
+    "a cartoon of a boy playing with a tiger",
+    "a cute robot artist painting on an easel, concept art",
+    "a close up of a woman wearing a transparent, prismatic, elaborate nemeses headdress, over the should pose, brown skin-tone"
 ]
 
 css = '''
-.gradio-container{max-width: 700px !important}
+.gradio-container{max-width: 560px !important}
 h1{text-align:center}
-footer {
-    visibility: hidden
-}
 '''
-
 with gr.Blocks(css=css) as demo:
-    gr.Markdown("""# Black Hole SDXL-Lightning""")
+    gr.Markdown(DESCRIPTION)
+    gr.DuplicateButton(
+        value="Duplicate Space for private use",
+        elem_id="duplicate-button",
+        visible=os.getenv("SHOW_DUPLICATE_BUTTON") == "1",
+    )
     with gr.Group():
         with gr.Row():
             prompt = gr.Text(
@@ -118,24 +125,14 @@ with gr.Blocks(css=css) as demo:
                 container=False,
             )
             run_button = gr.Button("Run", scale=0)
-        result = gr.Gallery(elem_id="gallery", label="Result", show_label=False) 
-        
+        result = gr.Gallery(label="Result", columns=NUM_IMAGES_PER_PROMPT, show_label=False)
     with gr.Accordion("Advanced options", open=False):
-        num_images = gr.Slider(
-            label="Number of Images",
-            minimum=1,
-            maximum=4,
-            step=1,
-            value=1,
-        )
         with gr.Row():
-            use_negative_prompt = gr.Checkbox(label="Use negative prompt", value=True)
+            use_negative_prompt = gr.Checkbox(label="Use negative prompt", value=False)
             negative_prompt = gr.Text(
                 label="Negative prompt",
-                max_lines=5,
-                lines=4,
+                max_lines=1,
                 placeholder="Enter a negative prompt",
-                value="(deformed, distorted, disfigured:1.3), poorly drawn, bad anatomy, wrong anatomy, extra limb, missing limb, floating limbs, (mutated hands and fingers:1.4), disconnected limbs, mutation, mutated, ugly, disgusting, blurry, amputation, NSFW",
                 visible=True,
             )
         seed = gr.Slider(
@@ -149,38 +146,33 @@ with gr.Blocks(css=css) as demo:
         with gr.Row(visible=True):
             width = gr.Slider(
                 label="Width",
-                minimum=512,
+                minimum=256,
                 maximum=MAX_IMAGE_SIZE,
-                step=64,
+                step=32,
                 value=1024,
             )
             height = gr.Slider(
                 label="Height",
-                minimum=512,
+                minimum=256,
                 maximum=MAX_IMAGE_SIZE,
-                step=64,
+                step=32,
                 value=1024,
             )
         with gr.Row():
             guidance_scale = gr.Slider(
                 label="Guidance Scale",
                 minimum=0.1,
-                maximum=6,
+                maximum=20,
                 step=0.1,
                 value=3.0,
             )
-            num_inference_steps = gr.Slider(
-                label="Number of inference steps",
-                minimum=1,
-                maximum=15,
-                step=1,
-                value=4,
-            )
 
     gr.Examples(
         examples=examples,
         inputs=prompt,
-        cache_examples=False
+        outputs=[result, seed],
+        fn=generate,
+        cache_examples=CACHE_EXAMPLES,
     )
 
     use_negative_prompt.change(
@@ -205,13 +197,11 @@ with gr.Blocks(css=css) as demo:
             width,
             height,
             guidance_scale,
-            num_inference_steps,
             randomize_seed,
-            num_images
         ],
         outputs=[result, seed],
         api_name="run",
     )
 
 if __name__ == "__main__":
-    demo.queue().launch()
+    demo.queue(max_size=20).launch()