Update app.py

app.py

@@ -1,184 +1,217 @@
-import gradio as gr
-import torch
-from diffusers import StableDiffusionPipeline
 
-from diffusion_webui.utils.model_list import stable_model_list
-from diffusion_webui.utils.scheduler_list import (
-    SCHEDULER_MAPPING,
-    get_scheduler,
-)
-import spaces
 
+import os
+import random
+import uuid
+import json
 
-class StableDiffusionText2ImageGenerator:
-    def __init__(self):
-        self.pipe = None
-
-    def load_model(
-        self,
-        model_path,
-        scheduler,
-    ):
-        if self.pipe is None:
-            self.pipe = StableDiffusionPipeline.from_pretrained(
-                model_path, safety_checker=None, torch_dtype=torch.float16
+import gradio as gr
+import numpy as np
+from PIL import Image
+import spaces
+import torch
+from diffusers import StableDiffusionXLPipeline, DPMSolverSinglestepScheduler
+
+# 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"))
+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
+
+def randomize_seed_fn(seed: int, randomize_seed: bool) -> int:
+    if randomize_seed:
+        seed = random.randint(0, MAX_SEED)
+    return seed
+
+@spaces.GPU()
+def generate(
+    prompt: str,
+    negative_prompt: str = "",
+    use_negative_prompt: bool = False,
+    seed: int = 1,
+    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
+    progress=gr.Progress(track_tqdm=True),
+):
+    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
+
+examples = [
+    "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"
+]
+
+css = '''
+.gradio-container{max-width: 700px !important}
+h1{text-align:center}
+footer {
+    visibility: hidden
+}
+'''
+
+with gr.Blocks(css=css) as demo:
+    gr.Markdown("""# Black Hole SDXL-Lightning""")
+    with gr.Group():
+        with gr.Row():
+            prompt = gr.Text(
+                label="Prompt",
+                show_label=False,
+                max_lines=1,
+                placeholder="Enter your prompt",
+                container=False,
             )
-
-        self.pipe = get_scheduler(pipe=self.pipe, scheduler=scheduler)
-        self.pipe.to("cuda")
-        self.pipe.enable_xformers_memory_efficient_attention()
-
-        return self.pipe
-
-    @spaces.GPU()
-    def generate_image(
-        self,
-        model_path: str,
-        prompt: str,
-        negative_prompt: str,
-        num_images_per_prompt: int,
-        scheduler: str,
-        guidance_scale: int,
-        num_inference_step: int,
-        height: int,
-        width: int,
-        seed_generator=0,
-    ):
-        pipe = self.load_model(
-            model_path=model_path,
-            scheduler=scheduler,
+            run_button = gr.Button("Run", scale=0)
+        result = gr.Gallery(elem_id="gallery", label="Result", 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)
+            negative_prompt = gr.Text(
+                label="Negative prompt",
+                max_lines=5,
+                lines=4,
+                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(
+            label="Seed",
+            minimum=0,
+            maximum=MAX_SEED,
+            step=1,
+            value=0,
         )
-        if seed_generator == 0:
-            random_seed = torch.randint(0, 1000000, (1,))
-            generator = torch.manual_seed(random_seed)
-        else:
-            generator = torch.manual_seed(seed_generator)
-
-        images = pipe(
-            prompt=prompt,
-            height=height,
-            width=width,
-            negative_prompt=negative_prompt,
-            num_images_per_prompt=num_images_per_prompt,
-            num_inference_steps=num_inference_step,
-            guidance_scale=guidance_scale,
-            generator=generator,
-        ).images
-
-        return images
-
-    def app():
-        with gr.Blocks():
-            with gr.Row():
-                with gr.Column():
-                    text2image_prompt = gr.Textbox(
-                        lines=1,
-                        placeholder="Prompt",
-                        show_label=False,
-                    )
-
-                    text2image_negative_prompt = gr.Textbox(
-                        lines=1,
-                        placeholder="Negative Prompt",
-                        show_label=False,
-                    )
-                    with gr.Row():
-                        with gr.Column():
-                            text2image_model_path = gr.Dropdown(
-                                choices=stable_model_list,
-                                value=stable_model_list[0],
-                                label="Text-Image Model Id",
-                            )
-
-                            text2image_guidance_scale = gr.Slider(
-                                minimum=0.1,
-                                maximum=15,
-                                step=0.1,
-                                value=7.5,
-                                label="Guidance Scale",
-                            )
-
-                            text2image_num_inference_step = gr.Slider(
-                                minimum=1,
-                                maximum=100,
-                                step=1,
-                                value=50,
-                                label="Num Inference Step",
-                            )
-                            text2image_num_images_per_prompt = gr.Slider(
-                                minimum=1,
-                                maximum=30,
-                                step=1,
-                                value=1,
-                                label="Number Of Images",
-                            )
-                        with gr.Row():
-                            with gr.Column():
-                                text2image_scheduler = gr.Dropdown(
-                                    choices=list(SCHEDULER_MAPPING.keys()),
-                                    value=list(SCHEDULER_MAPPING.keys())[0],
-                                    label="Scheduler",
-                                )
-
-                                text2image_height = gr.Slider(
-                                    minimum=128,
-                                    maximum=1280,
-                                    step=32,
-                                    value=512,
-                                    label="Image Height",
-                                )
-
-                                text2image_width = gr.Slider(
-                                    minimum=128,
-                                    maximum=1280,
-                                    step=32,
-                                    value=512,
-                                    label="Image Width",
-                                )
-                                text2image_seed_generator = gr.Slider(
-                                    label="Seed(0 for random)",
-                                    minimum=0,
-                                    maximum=1000000,
-                                    value=0,
-                                )
-                    text2image_predict = gr.Button(value="Generator")
-
-                with gr.Column():
-                    output_image = gr.Gallery(
-                        label="Generated images",
-                        show_label=False,
-                        elem_id="gallery",
-                    ).style(grid=(1, 2), height=200)
-
-            text2image_predict.click(
-                fn=StableDiffusionText2ImageGenerator().generate_image,
-                inputs=[
-                    text2image_model_path,
-                    text2image_prompt,
-                    text2image_negative_prompt,
-                    text2image_num_images_per_prompt,
-                    text2image_scheduler,
-                    text2image_guidance_scale,
-                    text2image_num_inference_step,
-                    text2image_height,
-                    text2image_width,
-                    text2image_seed_generator,
-                ],
-                outputs=output_image,
+        randomize_seed = gr.Checkbox(label="Randomize seed", value=True)
+        with gr.Row(visible=True):
+            width = gr.Slider(
+                label="Width",
+                minimum=512,
+                maximum=MAX_IMAGE_SIZE,
+                step=64,
+                value=1024,
+            )
+            height = gr.Slider(
+                label="Height",
+                minimum=512,
+                maximum=MAX_IMAGE_SIZE,
+                step=64,
+                value=1024,
             )
-
-import gradio as gr
-
-
-def diffusion_app():
-    app = gr.Blocks()
-    with app:
         with gr.Row():
-            with gr.Column():
-                    StableDiffusionText2ImageGenerator.app()
-
-    app.launch(debug=True)
+            guidance_scale = gr.Slider(
+                label="Guidance Scale",
+                minimum=0.1,
+                maximum=6,
+                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
+    )
+
+    use_negative_prompt.change(
+        fn=lambda x: gr.update(visible=x),
+        inputs=use_negative_prompt,
+        outputs=negative_prompt,
+        api_name=False,
+    )
+
+    gr.on(
+        triggers=[
+            prompt.submit,
+            negative_prompt.submit,
+            run_button.click,
+        ],
+        fn=generate,
+        inputs=[
+            prompt,
+            negative_prompt,
+            use_negative_prompt,
+            seed,
+            width,
+            height,
+            guidance_scale,
+            num_inference_steps,
+            randomize_seed,
+            num_images
+        ],
+        outputs=[result, seed],
+        api_name="run",
+    )
 
 if __name__ == "__main__":
-    diffusion_app()
+    demo.queue().launch()