Omni-Woodpecker: Hallucination Correction via Evidence-Prior Rebalancing for Omni LLMs

Omni-Woodpecker is a training-free framework that addresses cross-modal hallucinations in Omni LLMs by identifying conflicts from complementary modality-specific views, using orthogonal decomposition to separate grounding evidence from competing priors, and adaptively rebalancing fused representations before decoding. Experiments show the method is effective and generalizes well across architectures.