meituan-longcat/LongCat-Video — reverse-engineered prompt

Build me a local demo app for LongCat Video that I can run on a GPU machine without needing to touch command lines after setup.

I want a simple web page where I can choose what I’m trying to make, text to video, image to video, continue an existing video, long video, or an audio driven avatar video. For each mode, show only the inputs I need, like a prompt box, image upload, video upload, audio upload, seed, resolution, and output length. When I click generate, show progress, keep the app from crashing on errors, and save the finished video so I can preview and download it.

Please wire it to the LongCat Video and LongCat Video Avatar weights from Hugging Face, include clear setup steps for the Python environment, CUDA requirements, model downloads, and how to launch the app. Use the existing demo scripts where possible, and look up current docs online if needed.