OmniHarness：通过符号策略学习实现可泛化视觉生成
原标题：OmniHarness: Harnessing Generalizable Visual Generation via Symbolic Policy Learning
AI 摘要
北京航空航天大学、香港中文大学和新加坡国立大学的研究者提出 OmniHarness 框架，通过符号策略学习实现可泛化的视觉生成。该方法将验证过的执行过程抽象为面向视觉生成任务族的符号策略，保留共享流程与适用条件，并在执行中通过中间验证进行纠错与恢复。在 ComfyBench 创意任务上，OmniHarness 达到 95.0% 的解决率，超过最强基线 27.5 个百分点，且冻结的策略快照可即插即用地提升现有视觉智能体系统。
正文节选
1]Beihang University 2]The Chinese University of Hong Kong 3]National University of Singapore \contribution[*]Corresponding authors: Xu Xu (), Jinxiu Liu () \checkdata[Resources] OmniHarness: Harnessing Generalizable Visual Generation via Symbolic Policy Learning Abstract Unified multimodal large language models (MLLMs) and multi-agent systems have advanced visual generation. However, three limitations remain. (1) Existing methods often distill task-specific experience with limited generalizabil