AnyID

Ultra-Fidelity Universal Identity-Preserving Video Generation from Any Visual References

CVPR 2026

Jiahao Wang¹ Hualian Sheng² Sijia Cai^2,† Yuxiao Yang³ Weizhan Zhang^1,* Caixia Yan¹ Bing Deng² Jieping Ye²

^†Project Lead ^*Corresponding Author

¹Xi'an Jiaotong University ²Alibaba Cloud ³Tsinghua University

Paper Code (coming soon)

Model (coming soon)

Create Ultra-Real Actor

Ultra-Real Human Generation from Free-Form References

Given multiple free-form visual references (e.g., faces, portraits, videos) of a desired person, with one chosen as the primary reference, AnyID captures a dynamic holistic representation of the identity and generates hyper-realistic videos of the person across different contexts (e.g., camera movement, hairstyle, clothes, expression, behavior, background) controlled by the prompt. For aspects not mentioned in the prompt, AnyID will automatically align with the primary reference, thereby achieving the editing effect.

Primary Reference