
FaceVid-1K
Large-scale high-quality multiracial human face video dataset for evaluating and improving image-to-video and text-to-video face generation methods.
Project pageprojects
Selected work with reusable assets, project pages, demos, or engineering ownership. Paper-only work is kept in the publications list.

Large-scale high-quality multiracial human face video dataset for evaluating and improving image-to-video and text-to-video face generation methods.
Project page
Diagnostic benchmark for faithful 4D video editing, evaluating preservation, reveal completion, and scene expansion with region-aware metrics.
Project pageReusable per-item predictions, derived labels, and cached evaluation pipeline for auditing Video LLM scaling behavior and adaptive inference policies.
CodeRealtime avatar dialogue system with custom avatars, decoupled head/lip motion, Mandarin lip optimization, and super-resolution enhancement.

Customized 3D cartoon assets generated from user photos.

Multi-agent framework for unconstrained story envisioning through closed-loop cognitive orchestration.
Project pagei dont want to be jiucai anymore.