TencentARC/VideoPainter ? reverse-engineered prompt

Build me a working version of this VideoPainter project so I can inpaint and edit videos with a mask and text prompt, and also run the demo, inference, training, and evaluation scripts from the repo.

I want a simple workflow where I can upload a video, provide masks for the parts I want removed or changed, enter a prompt for the replacement content, and get back a cleaned up edited video. If there is already a Gradio style app in the repo, make sure it runs smoothly and is easy to use. Also make sure the training and evaluation commands actually work with the provided data layout, and the checkpoint loading is wired up correctly.

Use the existing README and code in the repo as the source of truth, and look up any current docs online if you need to. Please fix anything missing so the project can run end to end without me having to figure out the plumbing.