app.py

import gradio as gr
import torch
from diffusers import DiffusionPipeline
from diffusers.utils import load_image, export_to_video
import tempfile
import os

# Load the model (we'll initialize it when first used to save resources)
pipe = None

def generate_video(image, prompt, seed=42):
    global pipe
    
    # Initialize the model if not already loaded
    if pipe is None:
        pipe = DiffusionPipeline.from_pretrained(
            "Wan-AI/Wan2.1-VACE-14B", 
            torch_dtype=torch.float16
        )
        pipe.to("cuda")
    
    # Set the seed for reproducibility
    torch.manual_seed(seed)
    
    # Generate the video frames
    output = pipe(image=image, prompt=prompt).frames[0]
    
    # Save to temporary file
    temp_dir = tempfile.mkdtemp()
    output_path = os.path.join(temp_dir, "output.mp4")
    export_to_video(output, output_path)
    
    return output_path

# Create Gradio interface
with gr.Blocks(title="Wan2.1 Video Generation") as demo:
    gr.Markdown("# Wan2.1-VACE-14B Video Generation")
    gr.Markdown("Generate videos from images and prompts using Wan2.1 model")
    
    with gr.Row():
        with gr.Column():
            input_image = gr.Image(label="Input Image", type="pil")
            prompt = gr.Textbox(label="Prompt", placeholder="Describe the video you want to generate...")
            seed = gr.Number(label="Seed", value=42, precision=0)
            generate_btn = gr.Button("Generate Video")
        
        with gr.Column():
            output_video = gr.Video(label="Generated Video")
    
    # Example inputs
    gr.Examples(
        examples=[
            [
                "https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/diffusers/guitar-man.png",
                "A man with short gray hair plays a red electric guitar.",
                42
            ],
            [
                "https://upload.wikimedia.org/wikipedia/commons/thumb/4/41/Sunflower_from_Silesia2.jpg/1200px-Sunflower_from_Silesia2.jpg",
                "A sunflower slowly blooming in the sunlight",
                123
            ]
        ],
        inputs=[input_image, prompt, seed],
        outputs=output_video,
        fn=generate_video,
        cache_examples=True
    )
    
    generate_btn.click(
        fn=generate_video,
        inputs=[input_image, prompt, seed],
        outputs=output_video
    )

demo.launch()