Diagnosing and Correcting Concept Omission in Multimodal Diffusion Transformers

Multimodal Diffusion Transformers (MM-DiTs) have achieved remarkable progress in text-to-image generation, yet they frequently suffer from concept omission, where specified objects or attributes fail to emerge in the generated image. By performing linear probing on text tokens, we demonstrate that text embeddings can distinguish a characteristic `omission signal' representing the absence of target concepts. Leveraging this insight, we propose Omission Signal Intervention (OSI), which amplifies the omission signal to actively catalyze the generation of missing concepts. Comprehensive experiments on FLUX.1-Dev and SD3.5-Medium demonstrate that OSI significantly alleviates concept omission even in extreme scenarios.

Text-to-image generation models sometimes omit certain concepts specified in a text prompt, but when and why this happens has been poorly understood. We discovered that specific internal signals inside the model encode whether a concept is being omitted, and proposed Omission Signal Intervention (OSI), a method that directly manipulates these signals to correct omissions. This improves the reliability of text-to-image generation, enabling models to more faithfully reflect user intent.