app.py

#!/usr/bin/env python

import os
import pathlib
import tempfile

import gradio as gr
import torch
from huggingface_hub import snapshot_download
from modelscope.outputs import OutputKeys
from modelscope.pipelines import pipeline

DESCRIPTION = "# ModelScope-Vid2Vid-XL"

if torch.cuda.is_available():
    model_cache_dir = os.getenv("MODEL_CACHE_DIR", "./models")
    model_dir = pathlib.Path(model_cache_dir) / "MS-Vid2Vid-XL"
    snapshot_download(repo_id="damo-vilab/MS-Vid2Vid-XL", repo_type="model", local_dir=model_dir)
    pipe = pipeline(task="video-to-video", model=model_dir.as_posix(), model_revision="v1.1.0", device="cuda:0")
else:
    pipe = None


def video_to_video(video_path: str, text: str) -> str:
    p_input = {"video_path": video_path, "text": text}
    output_file = tempfile.NamedTemporaryFile(suffix=".mp4", delete=False)
    pipe(p_input, output_video=output_file.name)[OutputKeys.OUTPUT_VIDEO]
    return output_file.name


with gr.Blocks(css="style.css") as demo:
    gr.Markdown(DESCRIPTION)
    input_video = gr.Video(label="Input video", type="filepath")
    text_description = gr.Textbox(label="Text description")
    run_button = gr.Button()
    output_video = gr.Video(label="Output video")
    run_button.click(
        fn=video_to_video,
        inputs=[input_video, text_description],
        outputs=output_video,
        api_name="run",
    )
demo.queue(max_size=20).launch()