OmniHarness: Harnessing Generalizable Visual Generation via Symbolic Policy Learning

A new method called OmniHarness applies symbolic policy learning to unified multimodal large language models (MLLMs) and multi-agent systems to make visual generation more generalizable. The work identifies three limitations in existing approaches: task-specific experience distillation with limited generalizability, reflection deferred until task completion, and a third limitation the source leaves truncated. OmniHarness is positioned as addressing these constraints in visual generation.

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 generalizability. 2 Reflection is often deferred until task completion. 3 Knowledg