app.py

import gradio as gr
import moviepy.editor as mp
from diffusers import DiffusionPipeline
import spaces

# Load diffusion pipelines
image_pipeline = DiffusionPipeline.from_pretrained("ByteDance/SDXL-Lightning")
video_pipeline = DiffusionPipeline.from_pretrained("stabilityai/stable-video-diffusion-img2vid-xt-1-1")

@spaces.GPU(duration=120)
def generate_images(prompt, num_images):
  """Generates images using the image pipeline."""
  images = []
  for _ in range(num_images):
    generated_image = image_pipeline.pipe(prompt=prompt).images[0]
    images.append(generated_image)
  return images

@spaces.GPU(duration=120)
def generate_videos(images):
  """Generates videos from a list of images using the video pipeline."""
  videos = []
  for image in images:
    generated_video = video_pipeline.pipe(images=image).images[0]
    videos.append(generated_video)
  return videos

def combine_videos(video_clips):
  """Combines video clips into a single video using moviepy."""
  final_clip = mp.concatenate_videoclips(video_clips)
  return final_clip

def generate(prompt, num_images):
  """Generates and combines images and videos."""
  images = generate_images(prompt, num_images)
  video_clips = generate_videos(images)
  combined_video = combine_videos(video_clips)
  return combined_video

# Gradio interface with improved formatting and video output
interface = gr.Interface(
    fn=generate,
    inputs="text",
    outputs="video",
    interpretation="markdown",
    title="AI Video Generation",
    description="Enter a prompt to generate a video using diffusion models.",
    css="""
      .output-video {
        width: 100%; /* Adjust width as needed */
        height: 400px; /* Adjust height as desired */
      }
    """,
)

# Launch the interface
interface.launch()