Diagnosing and Correcting Concept Omission in Multimodal Diffusion Transformers

Kanghyun Baek, Jaihyun Lew, Chaehun Shin, Jungbeom Lee, Sungroh Yoon
Proceedings of the 43rd International Conference on Machine Learning, PMLR 306:5107-5124, 2026.

Abstract

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.

Cite this Paper

BibTeX
@InProceedings{pmlr-v306-baek26c, title = {Diagnosing and Correcting Concept Omission in Multimodal Diffusion Transformers}, author = {Baek, Kanghyun and Lew, Jaihyun and Shin, Chaehun and Lee, Jungbeom and Yoon, Sungroh}, booktitle = {Proceedings of the 43rd International Conference on Machine Learning}, pages = {5107--5124}, year = {2026}, editor = {Zhang, Tong and Dudik, Miroslav and Jaggi, Martin and Agarwal, Alekh and Li, Sharon and Schuurmans, Dale and Zhu, Jerry and Berkenkamp, Felix and Dong, Hanze and Bietti, Alberto}, volume = {306}, series = {Proceedings of Machine Learning Research}, month = {06--11 Jul}, publisher = {PMLR}, pdf = {https://raw.githubusercontent.com/mlresearch/v306/main/assets/baek26c/baek26c.pdf}, url = {https://proceedings.mlr.press/v306/baek26c.html}, abstract = {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.} }
Endnote
%0 Conference Paper %T Diagnosing and Correcting Concept Omission in Multimodal Diffusion Transformers %A Kanghyun Baek %A Jaihyun Lew %A Chaehun Shin %A Jungbeom Lee %A Sungroh Yoon %B Proceedings of the 43rd International Conference on Machine Learning %C Proceedings of Machine Learning Research %D 2026 %E Tong Zhang %E Miroslav Dudik %E Martin Jaggi %E Alekh Agarwal %E Sharon Li %E Dale Schuurmans %E Jerry Zhu %E Felix Berkenkamp %E Hanze Dong %E Alberto Bietti %F pmlr-v306-baek26c %I PMLR %P 5107--5124 %U https://proceedings.mlr.press/v306/baek26c.html %V 306 %X 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.
APA
Baek, K., Lew, J., Shin, C., Lee, J. & Yoon, S.. (2026). Diagnosing and Correcting Concept Omission in Multimodal Diffusion Transformers. Proceedings of the 43rd International Conference on Machine Learning, in Proceedings of Machine Learning Research 306:5107-5124 Available from https://proceedings.mlr.press/v306/baek26c.html.

Related Material