Update app.py

app.py

@@ -1,154 +1,495 @@
 import gradio as gr
 import numpy as np
 import random
-
-# import spaces #[uncomment to use ZeroGPU]
-from diffusers import DiffusionPipeline
 import torch
+import gc
+from typing import Optional, Tuple
+import spaces  # For ZeroGPU support
+
+# Image generation models
+from diffusers import (
+    DiffusionPipeline, StableDiffusionPipeline, 
+    StableDiffusionXLPipeline, AutoPipelineForText2Image,
+    AnimateDiffPipeline, DiffusionPipeline as VideoPipeline
+)
 
 device = "cuda" if torch.cuda.is_available() else "cpu"
-model_repo_id = "stabilityai/sdxl-turbo"  # Replace to the model you would like to use
+MAX_SEED = np.iinfo(np.int32).max
+MAX_IMAGE_SIZE = 1024
 
-if torch.cuda.is_available():
-    torch_dtype = torch.float16
-else:
-    torch_dtype = torch.float32
+# Model configurations optimized for different hardware
+MODEL_CONFIGS = {
+    "Image Models": {
+        "SDXL-Turbo (Fast)": {
+            "repo_id": "stabilityai/sdxl-turbo",
+            "pipeline_class": "auto",
+            "cpu_friendly": True,
+            "vram_usage": "Low",
+            "default_steps": 2,
+            "default_guidance": 0.0,
+            "torch_dtype": torch.float16 if torch.cuda.is_available() else torch.float32
+        },
+        "SD 1.5 (CPU Optimized)": {
+            "repo_id": "runwayml/stable-diffusion-v1-5",
+            "pipeline_class": "sd15",
+            "cpu_friendly": True,
+            "vram_usage": "Low",
+            "default_steps": 20,
+            "default_guidance": 7.5,
+            "torch_dtype": torch.float32
+        },
+        "SD 2.1 (Balanced)": {
+            "repo_id": "stabilityai/stable-diffusion-2-1",
+            "pipeline_class": "sd21",
+            "cpu_friendly": False,
+            "vram_usage": "Medium",
+            "default_steps": 25,
+            "default_guidance": 7.5,
+            "torch_dtype": torch.float16 if torch.cuda.is_available() else torch.float32
+        },
+        "SDXL Base (High Quality)": {
+            "repo_id": "stabilityai/stable-diffusion-xl-base-1.0",
+            "pipeline_class": "sdxl",
+            "cpu_friendly": False,
+            "vram_usage": "High",
+            "default_steps": 30,
+            "default_guidance": 7.5,
+            "torch_dtype": torch.float16 if torch.cuda.is_available() else torch.float32
+        }
+    },
+    "Video Models": {
+        "AnimateDiff (Motion)": {
+            "repo_id": "guoyww/animatediff-motion-adapter-v1-5-2",
+            "pipeline_class": "animatediff",
+            "cpu_friendly": False,
+            "vram_usage": "High",
+            "default_steps": 25,
+            "default_guidance": 7.5,
+            "torch_dtype": torch.float16 if torch.cuda.is_available() else torch.float32
+        },
+        "Zeroscope v2 (Text-to-Video)": {
+            "repo_id": "cerspense/zeroscope_v2_576w",
+            "pipeline_class": "video",
+            "cpu_friendly": False,
+            "vram_usage": "Very High",
+            "default_steps": 40,
+            "default_guidance": 9.0,
+            "torch_dtype": torch.float16 if torch.cuda.is_available() else torch.float32
+        }
+    }
+}
 
-pipe = DiffusionPipeline.from_pretrained(model_repo_id, torch_dtype=torch_dtype)
-pipe = pipe.to(device)
+# Global pipeline cache
+current_pipeline = None
+current_model_name = None
 
-MAX_SEED = np.iinfo(np.int32).max
-MAX_IMAGE_SIZE = 1024
+def clear_pipeline():
+    """Clear current pipeline to free memory"""
+    global current_pipeline
+    if current_pipeline is not None:
+        del current_pipeline
+        current_pipeline = None
+        gc.collect()
+        if torch.cuda.is_available():
+            torch.cuda.empty_cache()
 
+def get_pipeline_class(pipeline_type: str):
+    """Get the appropriate pipeline class"""
+    if pipeline_type == "auto":
+        return AutoPipelineForText2Image
+    elif pipeline_type == "sd15":
+        return StableDiffusionPipeline
+    elif pipeline_type == "sd21":
+        return StableDiffusionPipeline
+    elif pipeline_type == "sdxl":
+        return StableDiffusionXLPipeline
+    elif pipeline_type == "animatediff":
+        return AnimateDiffPipeline
+    elif pipeline_type == "video":
+        return VideoPipeline
+    else:
+        return DiffusionPipeline
 
-# @spaces.GPU #[uncomment to use ZeroGPU]
-def infer(
-    prompt,
-    negative_prompt,
-    seed,
-    randomize_seed,
-    width,
-    height,
-    guidance_scale,
-    num_inference_steps,
-    progress=gr.Progress(track_tqdm=True),
-):
-    if randomize_seed:
-        seed = random.randint(0, MAX_SEED)
+def load_model(model_name: str, model_type: str = "Image Models"):
+    """Load a model with memory optimization"""
+    global current_pipeline, current_model_name
+    
+    if current_model_name == model_name and current_pipeline is not None:
+        return current_pipeline
+    
+    # Clear previous model
+    clear_pipeline()
+    
+    config = MODEL_CONFIGS[model_type][model_name]
+    pipeline_class = get_pipeline_class(config["pipeline_class"])
+    
+    try:
+        # Load with optimizations
+        pipe = pipeline_class.from_pretrained(
+            config["repo_id"],
+            torch_dtype=config["torch_dtype"],
+            use_safetensors=True,
+            variant="fp16" if torch.cuda.is_available() and config["torch_dtype"] == torch.float16 else None
+        )
+        
+        # Apply optimizations based on hardware
+        if torch.cuda.is_available():
+            pipe = pipe.to(device)
+            # Enable memory efficient attention
+            if hasattr(pipe, 'enable_attention_slicing'):
+                pipe.enable_attention_slicing()
+            if hasattr(pipe, 'enable_xformers_memory_efficient_attention'):
+                try:
+                    pipe.enable_xformers_memory_efficient_attention()
+                except:
+                    pass
+        else:
+            pipe = pipe.to(device)
+            # CPU optimizations
+            if hasattr(pipe, 'enable_attention_slicing'):
+                pipe.enable_attention_slicing()
+        
+        current_pipeline = pipe
+        current_model_name = model_name
+        return pipe
+        
+    except Exception as e:
+        return f"Error loading model: {str(e)}"
 
-    generator = torch.Generator().manual_seed(seed)
+@spaces.GPU(duration=60)  # ZeroGPU support
+def generate_image(
+    model_name: str,
+    prompt: str,
+    negative_prompt: str,
+    seed: int,
+    randomize_seed: bool,
+    width: int,
+    height: int,
+    guidance_scale: float,
+    num_inference_steps: int,
+    progress=gr.Progress(track_tqdm=True),
+) -> Tuple[Optional[np.ndarray], int, str]:
+    
+    if not prompt.strip():
+        return None, seed, "Please enter a prompt"
+    
+    try:
+        # Load model
+        pipe = load_model(model_name, "Image Models")
+        if isinstance(pipe, str):  # Error message
+            return None, seed, pipe
+        
+        # Handle seed
+        if randomize_seed:
+            seed = random.randint(0, MAX_SEED)
+        
+        generator = torch.Generator(device=device).manual_seed(seed)
+        
+        # Adjust parameters for CPU
+        if device == "cpu":
+            width = min(width, 512)
+            height = min(height, 512)
+            num_inference_steps = min(num_inference_steps, 20)
+        
+        # Generate image
+        with torch.autocast(device):
+            result = pipe(
+                prompt=prompt,
+                negative_prompt=negative_prompt if negative_prompt.strip() else None,
+                guidance_scale=guidance_scale,
+                num_inference_steps=num_inference_steps,
+                width=width,
+                height=height,
+                generator=generator,
+            )
+        
+        image = result.images[0]
+        return image, seed, "✅ Image generated successfully!"
+        
+    except Exception as e:
+        error_msg = f"❌ Generation failed: {str(e)}"
+        if "out of memory" in str(e).lower():
+            error_msg += "\n💡 Try: Lower resolution, fewer steps, or use a CPU-friendly model"
+        return None, seed, error_msg
 
-    image = pipe(
-        prompt=prompt,
-        negative_prompt=negative_prompt,
-        guidance_scale=guidance_scale,
-        num_inference_steps=num_inference_steps,
-        width=width,
-        height=height,
-        generator=generator,
-    ).images[0]
+@spaces.GPU(duration=120)  # Longer duration for video
+def generate_video(
+    model_name: str,
+    prompt: str,
+    negative_prompt: str,
+    seed: int,
+    randomize_seed: bool,
+    num_frames: int,
+    guidance_scale: float,
+    num_inference_steps: int,
+    progress=gr.Progress(track_tqdm=True),
+) -> Tuple[Optional[str], int, str]:
+    
+    if not prompt.strip():
+        return None, seed, "Please enter a prompt"
+    
+    if device == "cpu":
+        return None, seed, "❌ Video generation requires GPU"
+    
+    try:
+        # Load model
+        pipe = load_model(model_name, "Video Models")
+        if isinstance(pipe, str):  # Error message
+            return None, seed, pipe
+        
+        # Handle seed
+        if randomize_seed:
+            seed = random.randint(0, MAX_SEED)
+        
+        generator = torch.Generator(device=device).manual_seed(seed)
+        
+        # Generate video
+        with torch.autocast(device):
+            if "animatediff" in model_name.lower():
+                result = pipe(
+                    prompt=prompt,
+                    negative_prompt=negative_prompt if negative_prompt.strip() else None,
+                    num_frames=num_frames,
+                    guidance_scale=guidance_scale,
+                    num_inference_steps=num_inference_steps,
+                    generator=generator,
+                )
+                # Save as GIF
+                video_path = "output_video.gif"
+                result.export_to_gif(video_path)
+            else:
+                result = pipe(
+                    prompt=prompt,
+                    negative_prompt=negative_prompt if negative_prompt.strip() else None,
+                    num_frames=num_frames,
+                    guidance_scale=guidance_scale,
+                    num_inference_steps=num_inference_steps,
+                    generator=generator,
+                )
+                # Save as MP4
+                video_path = "output_video.mp4"
+                result.export_to_video(video_path)
+        
+        return video_path, seed, "✅ Video generated successfully!"
+        
+    except Exception as e:
+        error_msg = f"❌ Video generation failed: {str(e)}"
+        if "out of memory" in str(e).lower():
+            error_msg += "\n💡 Try: Fewer frames, lower steps, or switch to image generation"
+        return None, seed, error_msg
 
-    return image, seed
+def get_model_defaults(model_name: str, model_type: str):
+    """Get default values for selected model"""
+    if model_type in MODEL_CONFIGS and model_name in MODEL_CONFIGS[model_type]:
+        config = MODEL_CONFIGS[model_type][model_name]
+        return config["default_steps"], config["default_guidance"]
+    return 20, 7.5
 
+# Example prompts
+image_examples = [
+    "A majestic dragon flying over a mystical forest, detailed, 8k",
+    "Cyberpunk cityscape at night, neon lights, futuristic",
+    "Portrait of a wise old wizard with glowing eyes",
+    "Serene mountain lake at sunset, photorealistic"
+]
 
-examples = [
-    "Astronaut in a jungle, cold color palette, muted colors, detailed, 8k",
-    "An astronaut riding a green horse",
-    "A delicious ceviche cheesecake slice",
+video_examples = [
+    "A cat walking through a magical garden",
+    "Ocean waves crashing on a beach at sunset",
+    "A butterfly flying around colorful flowers",
+    "Clouds moving across a blue sky"
 ]
 
+# CSS for better styling
 css = """
 #col-container {
     margin: 0 auto;
-    max-width: 640px;
+    max-width: 900px;
+}
+.model-info {
+    padding: 10px;
+    margin: 5px 0;
+    border-radius: 5px;
+    background-color: #f0f0f0;
+}
+.status-success {
+    color: #28a745;
+}
+.status-error {
+    color: #dc3545;
 }
 """
 
-with gr.Blocks(css=css) as demo:
+# Main Gradio interface
+with gr.Blocks(css=css, title="Multi-Model AI Generator") as demo:
     with gr.Column(elem_id="col-container"):
-        gr.Markdown(" # Text-to-Image Gradio Template")
-
-        with gr.Row():
-            prompt = gr.Text(
-                label="Prompt",
-                show_label=False,
-                max_lines=1,
-                placeholder="Enter your prompt",
-                container=False,
-            )
-
-            run_button = gr.Button("Run", scale=0, variant="primary")
-
-        result = gr.Image(label="Result", show_label=False)
-
-        with gr.Accordion("Advanced Settings", open=False):
-            negative_prompt = gr.Text(
-                label="Negative prompt",
-                max_lines=1,
-                placeholder="Enter a negative prompt",
-                visible=False,
-            )
-
-            seed = gr.Slider(
-                label="Seed",
-                minimum=0,
-                maximum=MAX_SEED,
-                step=1,
-                value=0,
-            )
-
-            randomize_seed = gr.Checkbox(label="Randomize seed", value=True)
-
-            with gr.Row():
-                width = gr.Slider(
-                    label="Width",
-                    minimum=256,
-                    maximum=MAX_IMAGE_SIZE,
-                    step=32,
-                    value=1024,  # Replace with defaults that work for your model
-                )
-
-                height = gr.Slider(
-                    label="Height",
-                    minimum=256,
-                    maximum=MAX_IMAGE_SIZE,
-                    step=32,
-                    value=1024,  # Replace with defaults that work for your model
-                )
-
-            with gr.Row():
-                guidance_scale = gr.Slider(
-                    label="Guidance scale",
-                    minimum=0.0,
-                    maximum=10.0,
-                    step=0.1,
-                    value=0.0,  # Replace with defaults that work for your model
-                )
-
-                num_inference_steps = gr.Slider(
-                    label="Number of inference steps",
-                    minimum=1,
-                    maximum=50,
-                    step=1,
-                    value=2,  # Replace with defaults that work for your model
-                )
-
-        gr.Examples(examples=examples, inputs=[prompt])
-    gr.on(
-        triggers=[run_button.click, prompt.submit],
-        fn=infer,
-        inputs=[
-            prompt,
-            negative_prompt,
-            seed,
-            randomize_seed,
-            width,
-            height,
-            guidance_scale,
-            num_inference_steps,
-        ],
-        outputs=[result, seed],
-    )
+        gr.Markdown("# 🎨 Multi-Model AI Generator")
+        gr.Markdown("Generate images and videos using various AI models optimized for different hardware configurations.")
+        
+        # Hardware info
+        hardware_info = f"🖥️ **Device**: {device.upper()}"
+        if torch.cuda.is_available():
+            gpu_name = torch.cuda.get_device_name(0)
+            vram_gb = torch.cuda.get_device_properties(0).total_memory / 1e9
+            hardware_info += f" ({gpu_name}, {vram_gb:.1f}GB VRAM)"
+        gr.Markdown(hardware_info)
+        
+        with gr.Tabs():
+            # IMAGE GENERATION TAB
+            with gr.TabItem("🖼️ Image Generation"):
+                with gr.Row():
+                    with gr.Column(scale=3):
+                        img_prompt = gr.Text(
+                            label="Prompt",
+                            placeholder="Describe the image you want to generate...",
+                            lines=2
+                        )
+                    with gr.Column(scale=1):
+                        img_generate_btn = gr.Button("🎨 Generate Image", variant="primary", size="lg")
+                
+                with gr.Row():
+                    with gr.Column(scale=2):
+                        img_model_dropdown = gr.Dropdown(
+                            choices=list(MODEL_CONFIGS["Image Models"].keys()),
+                            value="SDXL-Turbo (Fast)",
+                            label="Model",
+                            info="Choose based on your hardware capabilities"
+                        )
+                        
+                        # Model info display
+                        img_model_info = gr.Markdown("", elem_classes="model-info")
+                        
+                    with gr.Column(scale=3):
+                        img_result = gr.Image(label="Generated Image", height=400)
+                
+                with gr.Accordion("⚙️ Advanced Settings", open=False):
+                    with gr.Row():
+                        img_negative_prompt = gr.Text(
+                            label="Negative Prompt",
+                            placeholder="What you don't want in the image...",
+                            lines=1
+                        )
+                    
+                    with gr.Row():
+                        img_seed = gr.Slider(0, MAX_SEED, value=0, label="Seed")
+                        img_randomize_seed = gr.Checkbox(label="Random Seed", value=True)
+                    
+                    with gr.Row():
+                        img_width = gr.Slider(256, MAX_IMAGE_SIZE, value=512, step=64, label="Width")
+                        img_height = gr.Slider(256, MAX_IMAGE_SIZE, value=512, step=64, label="Height")
+                    
+                    with gr.Row():
+                        img_guidance = gr.Slider(0.0, 20.0, value=7.5, step=0.5, label="Guidance Scale")
+                        img_steps = gr.Slider(1, 50, value=20, step=1, label="Inference Steps")
+                
+                img_status = gr.Markdown("Ready to generate!", elem_classes="status-success")
+                gr.Examples(examples=image_examples, inputs=[img_prompt])
+            
+            # VIDEO GENERATION TAB
+            with gr.TabItem("🎬 Video Generation"):
+                with gr.Row():
+                    with gr.Column(scale=3):
+                        vid_prompt = gr.Text(
+                            label="Prompt",
+                            placeholder="Describe the video you want to generate...",
+                            lines=2
+                        )
+                    with gr.Column(scale=1):
+                        vid_generate_btn = gr.Button("🎬 Generate Video", variant="primary", size="lg")
+                
+                with gr.Row():
+                    with gr.Column(scale=2):
+                        vid_model_dropdown = gr.Dropdown(
+                            choices=list(MODEL_CONFIGS["Video Models"].keys()),
+                            value="AnimateDiff (Motion)",
+                            label="Model",
+                            info="Video generation requires GPU"
+                        )
+                        
+                        vid_model_info = gr.Markdown("", elem_classes="model-info")
+                        
+                    with gr.Column(scale=3):
+                        vid_result = gr.Video(label="Generated Video", height=400)
+                
+                with gr.Accordion("⚙️ Video Settings", open=False):
+                    with gr.Row():
+                        vid_negative_prompt = gr.Text(
+                            label="Negative Prompt",
+                            placeholder="What you don't want in the video...",
+                            lines=1
+                        )
+                    
+                    with gr.Row():
+                        vid_seed = gr.Slider(0, MAX_SEED, value=0, label="Seed")
+                        vid_randomize_seed = gr.Checkbox(label="Random Seed", value=True)
+                    
+                    with gr.Row():
+                        vid_frames = gr.Slider(8, 64, value=16, step=8, label="Number of Frames")
+                        vid_guidance = gr.Slider(1.0, 20.0, value=7.5, step=0.5, label="Guidance Scale")
+                        vid_steps = gr.Slider(10, 50, value=25, step=1, label="Inference Steps")
+                
+                vid_status = gr.Markdown("Ready to generate!", elem_classes="status-success")
+                gr.Examples(examples=video_examples, inputs=[vid_prompt])
+        
+        # Model info update functions
+        def update_img_model_info(model_name):
+            config = MODEL_CONFIGS["Image Models"][model_name]
+            info = f"""
+            **VRAM Usage**: {config['vram_usage']} | **CPU Friendly**: {'✅' if config['cpu_friendly'] else '❌'}
+            
+            **Recommended Settings**: {config['default_steps']} steps, {config['default_guidance']} guidance
+            """
+            steps, guidance = get_model_defaults(model_name, "Image Models")
+            return info, steps, guidance
+        
+        def update_vid_model_info(model_name):
+            config = MODEL_CONFIGS["Video Models"][model_name]
+            info = f"""
+            **VRAM Usage**: {config['vram_usage']} | **CPU Friendly**: {'✅' if config['cpu_friendly'] else '❌'}
+            
+            **Recommended Settings**: {config['default_steps']} steps, {config['default_guidance']} guidance
+            """
+            steps, guidance = get_model_defaults(model_name, "Video Models")
+            return info, steps, guidance
+        
+        # Event handlers
+        img_model_dropdown.change(
+            update_img_model_info,
+            inputs=[img_model_dropdown],
+            outputs=[img_model_info, img_steps, img_guidance]
+        )
+        
+        vid_model_dropdown.change(
+            update_vid_model_info,
+            inputs=[vid_model_dropdown],
+            outputs=[vid_model_info, vid_steps, vid_guidance]
+        )
+        
+        # Generation event handlers
+        img_generate_btn.click(
+            generate_image,
+            inputs=[
+                img_model_dropdown, img_prompt, img_negative_prompt,
+                img_seed, img_randomize_seed, img_width, img_height,
+                img_guidance, img_steps
+            ],
+            outputs=[img_result, img_seed, img_status]
+        )
+        
+        vid_generate_btn.click(
+            generate_video,
+            inputs=[
+                vid_model_dropdown, vid_prompt, vid_negative_prompt,
+                vid_seed, vid_randomize_seed, vid_frames,
+                vid_guidance, vid_steps
+            ],
+            outputs=[vid_result, vid_seed, vid_status]
+        )
 
 if __name__ == "__main__":
-    demo.launch()
+    demo.launch(
+        share=True,
+        server_name="0.0.0.0",
+        server_port=7860,
+        show_error=True
+    )