app.py

import torch  
import gradio as gr  
import sys  
import traceback  

# Improved import with fallback handling  
try:  
    from diffusers import AnimateDiffPipeline, MotionAdapter, EulerDiscreteScheduler  
    from diffusers.utils import export_to_gif  
    from huggingface_hub import hf_hub_download  
    from safetensors.torch import load_file  
except ImportError as e:  
    print(f"Import Error: {e}")  
    sys.exit(1)  

# Comprehensive Base Models  
BASE_MODELS = {  
    "Realistic": [  
        "emilianJR/epiCRealism",  
        "SG161222/Realistic_Vision_V5.1_noVAE",  
        "Lykon/dreamshaper-8",  
        "digiplay/AbsoluteReality_v1.8.1",  
    ],  
    "Anime & Cartoon": [  
        "cagliostroaic/ToonYou",  
        "Sangyun/IMP",  
        "Lykon/Mistoon_Anime",  
        "digiplay/DynaVision_v1.0",  
    ]  
}  

def detect_device():  
    """  
    Robust device detection with detailed logging  
    """  
    try:  
        if torch.cuda.is_available():  
            print(f"CUDA Available. Using GPU: {torch.cuda.get_device_name(0)}")  
            return "cuda"  
        elif torch.backends.mps.is_available():  
            print("Using MPS (Apple Silicon)")  
            return "mps"  
        else:  
            print("No GPU detected. Falling back to CPU.")  
            return "cpu"  
    except Exception as e:  
        print(f"Device detection error: {e}")  
        return "cpu"  

def generate_video(prompt, base_model, steps=4, motion_strength=0.7, guidance_scale=1.0):  
    try:  
        device = detect_device()  
        dtype = torch.float16 if device == "cuda" else torch.float32  

        # Official AnimateDiff-Lightning Repository  
        repo = "ByteDance/AnimateDiff-Lightning"  
        ckpt = f"animatediff_lightning_{steps}step_diffusers.safetensors"  

        # Motion Adapter Setup with Error Handling  
        try:  
            adapter = MotionAdapter().to(device, dtype)  
            adapter.load_state_dict(  
                load_file(  
                    hf_hub_download(repo, ckpt),   
                    device=device  
                )  
            )  
        except Exception as adapter_error:  
            print(f"Motion Adapter Loading Error: {adapter_error}")  
            return None  

        # Flexible Model Loading  
        try:  
            pipe = AnimateDiffPipeline.from_pretrained(  
                base_model,   
                motion_adapter=adapter,   
                torch_dtype=dtype  
            ).to(device)  
        except Exception as model_error:  
            print(f"Model loading error with {base_model}: {model_error}")  
            base_model = "emilianJR/epiCRealism"  
            pipe = AnimateDiffPipeline.from_pretrained(  
                base_model,   
                motion_adapter=adapter,   
                torch_dtype=dtype  
            ).to(device)  

        pipe.scheduler = EulerDiscreteScheduler.from_config(  
            pipe.scheduler.config,   
            timestep_spacing="trailing",   
            beta_schedule="linear"  
        )  

        # Generation with Enhanced Error Handling  
        try:  
            output = pipe(  
                prompt=prompt,   
                guidance_scale=guidance_scale,  
                num_inference_steps=steps  
            )  
            
            gif_path = "animation.gif"  
            export_to_gif(output.frames[0], gif_path)  
            return gif_path  
        
        except Exception as gen_error:  
            print(f"Video Generation Error: {gen_error}")  
            return None  

    except Exception as e:  
        print(f"Unexpected error in video generation: {e}")  
        traceback.print_exc()  
        return None  

def create_interface():  
    with gr.Blocks() as demo:  
        gr.Markdown("## AnimateDiff-Lightning Video Generator")  
        
        with gr.Row():  
            prompt = gr.Textbox(  
                label="Prompt",   
                placeholder="Describe the video you want to generate...",  
                lines=3  
            )  
            base_model = gr.Dropdown(  
                choices=[  
                    *BASE_MODELS["Realistic"],   
                    *BASE_MODELS["Anime & Cartoon"]  
                ],   
                label="Base Model",  
                value="emilianJR/epiCRealism"  
            )  
        
        with gr.Row():  
            steps = gr.Slider(  
                minimum=1,   
                maximum=8,   
                step=1,   
                value=4,  
                label="Inference Steps"  
            )  
            guidance_scale = gr.Slider(  
                minimum=0.1,   
                maximum=20,   
                step=0.1,   
                value=1.0,   
                label="Guidance Scale"  
            )  
            motion_strength = gr.Slider(  
                minimum=0,   
                maximum=1,   
                step=0.1,   
                value=0.7,  
                label="Motion Strength"  
            )  
        
        generate_btn = gr.Button("Generate Video", variant="primary")  
        output = gr.Image(label="Generated Video")  
        
        generate_btn.click(  
            generate_video,   
            inputs=[prompt, base_model, steps, motion_strength, guidance_scale],   
            outputs=output  
        )  
    
    return demo  

def main():  
    print("Initializing AnimateDiff-Lightning Gradio Interface...")  
    
    try:  
        demo = create_interface()  
        
        # Public sharing with detailed config  
        demo.launch(  
            share=True,           # Create public link  
            debug=True,           # Detailed error reporting  
            show_error=True,      # Display errors in UI  
            server_name="0.0.0.0" # Accessible from any IP  
        )  
    
    except Exception as e:  
        print(f"Gradio Launch Error: {e}")  
        traceback.print_exc()  

if __name__ == "__main__":  
    main()