app.py

import gradio as gr
from diffusers import DiffusionPipeline
from diffusers.utils import export_to_video
import torch
import os
from PIL import Image
import spaces

# Load the pre-trained pipeline
pipeline = DiffusionPipeline.from_pretrained("stabilityai/stable-video-diffusion-img2vid-xt")

# Define the Gradio interface
interface = gr.Interface(
    fn=lambda img: generate_video(img),
    inputs=gr.Image(type="pil"),
    outputs=gr.Video(),
    title="Stable Video Diffusion",
    description="Upload an image to generate a video",
    theme="soft"
)

@spaces.GPU(duration=480)
def generate_video(image):
  """
  Generates a video from an input image using the pipeline.

  Args:
      image: A PIL Image object representing the input image.

  Returns:
      The path of a video file.
  """
  video_frames = pipeline(image=image, num_inference_steps=20).images

  # Frames to Video
  os.makedirs("outputs", exist_ok=True)
  base_count = len(glob(os.path.join("outputs", "*.mp4")))
  video_path = os.path.join("outputs", f"{base_count:06d}.mp4")
  export_to_video(video_frames, video_path, fps=6)

  return video_path

# Launch the Gradio app
interface.launch()