app.py

import gradio as gr
import numpy as np
import random
import torch
from diffusers import DiffusionPipeline, StableVideoDiffusionPipeline
from PIL import Image
import imageio

device = "cuda" if torch.cuda.is_available() else "cpu"
MAX_SEED = np.iinfo(np.int32).max

# Load SDXL for image generation
sdxl_model_id = "stabilityai/sdxl-turbo"
image_pipe = DiffusionPipeline.from_pretrained(
    sdxl_model_id,
    torch_dtype=torch.float16 if device == "cuda" else torch.float32
).to(device)

# Load Stable Video Diffusion for video generation
svd_model_id = "stabilityai/stable-video-diffusion-img2vid"
video_pipe = StableVideoDiffusionPipeline.from_pretrained(
    svd_model_id,
    torch_dtype=torch.float16 if device == "cuda" else torch.float32,
    variant="fp16" if device == "cuda" else None
)
if device == "cuda":
    video_pipe.enable_model_cpu_offload()

def generate_video_from_text(prompt, seed=0, randomize_seed=True):
    if randomize_seed:
        seed = random.randint(0, MAX_SEED)
    generator = torch.Generator(device=device).manual_seed(seed)

    # Generate image
    image = image_pipe(
        prompt=prompt,
        generator=generator,
        guidance_scale=0.0,
        num_inference_steps=2,
        width=1024,
        height=1024
    ).images[0]

    # Resize for SVD
    image = image.resize((512, 512))

    # Generate video
    video_frames = video_pipe(image).frames[0]
    video_path = f"/tmp/generated_{seed}.mp4"
    imageio.mimsave(video_path, video_frames, fps=7)

    return video_path, image, seed

# Use Interface instead of Blocks
demo = gr.Interface(
    fn=generate_video_from_text,
    inputs=[
        gr.Textbox(label="Prompt", placeholder="Describe your scene..."),
        gr.Number(label="Seed", value=0),
        gr.Checkbox(label="Randomize Seed", value=True)
    ],
    outputs=[
        gr.Video(label="Generated Video"),
        gr.Image(label="Generated Image"),
        gr.Number(label="Seed Used")
    ]
)

# Expose endpoint
demo.api_name = "predict"
demo.launch()