app.py

import gradio as gr
import torch
import numpy as np
from diffusers import I2VGenXLPipeline
from PIL import Image
from moviepy.editor import ImageSequenceClip
import io

def generate_video(image, prompt, negative_prompt, video_length):
    generator = torch.manual_seed(8888)

    # Set the device to CPU or a non-NVIDIA GPU
    device = torch.device("mps" if torch.backends.mps.is_available() else "cpu")
    print(f"Using device: {device}")

    # Load the pipeline
    pipeline = I2VGenXLPipeline.from_pretrained("ali-vilab/i2vgen-xl", torch_dtype=torch.float32)
    pipeline.to(device)  # Move the model to the selected device

    # Generate frames with progress tracking
    frames = []
    total_frames = video_length * 30  # Assuming 30 frames per second

    for i in range(total_frames):
        frame = pipeline(
            prompt=prompt,
            image=image,
            num_inference_steps=5,
            negative_prompt=negative_prompt,
            guidance_scale=9.0,
            generator=generator,
            num_frames=1
        ).frames[0]
        frames.append(np.array(frame))

        # Update progress
        yield (i + 1) / total_frames  # Yield progress

    # Create a video clip from the frames
    output_file = "output_video.mp4"
    clip = ImageSequenceClip(frames, fps=30)  # Set the frames per second
    clip.write_videofile(output_file, codec='libx264', audio=False)

    return output_file

# Gradio interface
def interface(image, prompt, negative_prompt, video_length):
    # Convert the uploaded image to a PIL Image
    image = Image.open(io.BytesIO(image.read()))
    
    # Generate video and track progress
    return generate_video(image, prompt, negative_prompt, video_length)

# Create Gradio Blocks
with gr.Blocks() as demo:
    gr.Markdown("# AI-Powered Video Generation")
    
    with gr.Row():
        image_input = gr.Image(type="file", label="Upload Image")
        prompt_input = gr.Textbox(label="Enter the Prompt")
        negative_prompt_input = gr.Textbox(label="Enter the Negative Prompt")
        video_length_input = gr.Number(label="Video Length (seconds)", value=10, precision=0)

    generate_button = gr.Button("Generate Video")
    output_video = gr.Video(label="Output Video")

    # Define the button action
    generate_button.click(
        interface,
        inputs=[image_input, prompt_input, negative_prompt_input, video_length_input],
        outputs=output_video,
        show_progress=True  # Show progress bar
    )

# Launch the Gradio app
demo.launch()