Author name: Ziwei Huang, Ying Shu, Hao Fang, Quanyu Long, Wenya Wang, Qiushi Guo, Tiezheng Ge, Leilei Gan

arXiv:2510.18263v2 Announce Type: replace
Abstract: Subject-driven image generation models face a fundamental trade-off between identity preservation (fidelity) and prompt adherence (editability). While online reinforcement learning (RL), specifically…