PKU-YuanGroup/ConsisID ? reverse-engineered prompt

Build me a simple app for ConsisID that lets a user upload a face photo and type a text prompt, then generates a short video that keeps the person’s identity consistent across the frames.

I want the whole thing to be easy to run locally, with a clean demo page, a clear inference script, and sensible defaults so I can try it without digging through the code. If the model weights or setup details need to be pulled from current docs online, please do that and wire everything up so the app works end to end.

Please also make sure the repo includes the basic training and evaluation entry points already expected here, and keep the code organized so it is easy to use for both demo generation and future training runs.