DynASyn: Multi-Subject Personalization Enabling Dynamic Action Synthesis

Yongjin Choi, Chanhun Park, Seung Jun Baek · 2025

Read DynASyn: Multi-Subject Personalization Enabling Dynamic Action Synthesis by Yongjin Choi, Chanhun Park, Seung Jun Baek (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-2503-17728.txt.

Description

Image generators can be personalized to place a known person or object into new scenes, but getting several subjects to act and interact in fresh ways is hard. Yongjin Choi, Chanhun Park, and Seung Jun Baek propose DynASyn, a method that personalizes multiple subjects from a single reference photo. The challenge is that models overfit the photo and freeze the subjects' poses.

Their method aligns learned concept priors with each subject's look and actions, guiding the model's attention so identity survives while behavior changes. Extra prompt and image augmentation widens the range of actions. Tests show realistic images of subjects in new settings and dynamic interactions, beating baseline methods on both measured scores and visual quality.

Catalog record

CREATORYongjin Choi, Chanhun Park, Seung Jun Baek
YEAR2025
COLLECTIONPAPERS
FORMATTXT
FILE SIZE37 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2503-17728

Browse the full library · About the Tower of Babel