Durian: Dual Reference Image-Guided Portrait Animation with Attribute Transfer
Hyunsoo Cha, Byungjun Kim, Hanbyul Joo · 2025
Read Durian: Dual Reference Image-Guided Portrait Animation with Attribute Transfer by Hyunsoo Cha, Byungjun Kim, Hanbyul Joo (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2509-04434.txt.
Description
Hyunsoo Cha and two co-authors animate portraits that borrow traits from other faces in a 2025 research paper. Their system, Durian, moves a target portrait while drawing attributes, such as hairstyle or accessories, from reference images of different people. Training such transfer normally needs paired examples of the same person with and without the trait, which rarely exist at scale. Their self-reconstruction trick splits ordinary portrait videos into pseudo pairs: one frame supplies the attribute, another supplies identity.
A dual reference network reads both frames separately, then fuses them through spatial attention inside a diffusion model. Complementary masking keeps each stream specialized. Mask expansion and augmentation bridge the gap to true cross-identity use. Multi-trait blending and smooth interpolation come free.
Catalog record
| CREATOR | Hyunsoo Cha, Byungjun Kim, Hanbyul Joo |
|---|---|
| YEAR | 2025 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 52 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2509-04434