Zheng-Chong/CatV2TON ? reverse-engineered prompt

Build me a virtual try on app based on this CatV2TON project.

I want to be able to upload a person image or short video, choose a clothing item, and generate a realistic try on result that keeps the person’s pose and looks natural. It should support both image and video try on, and I’d like a simple way to run inference from the app for each one. Please also include a clear results view where I can compare the original and generated output side by side.

If possible, make it easy to use with the pretrained checkpoints from the repo, and add basic controls for things like dataset choice, output folder, seed, and image quality settings that already seem supported here. I’d also like a small evaluation page or script runner so I can check metrics for generated image and video outputs.

Use the existing code in the repo, and look up any current docs online if you need to fill in missing setup details.