Show, Don’t Tell: Morphing Latent Reasoning into Image Generation

Harold Haodong Chen, Xinxiang Yin, Wen-Jie Shu, Hongfei Zhang, Zixin Zhang, Chenfei Liao, Litao Guo, Qifeng Chen, Ying-Cong Chen
Proceedings of the 43rd International Conference on Machine Learning, PMLR 306:17940-17959, 2026.

Abstract

Text-to-image (T2I) generation has achieved remarkable progress, yet existing methods often lack the ability to dynamically reason and refine during generation–a hallmark of human creativity. Current reasoning-augmented paradigms mostly rely on explicit thought processes, where intermediate reasoning is decoded into discrete text at fixed steps with frequent image decoding and re-encoding, leading to inefficiencies, information loss, and cognitive mismatches. To bridge this gap, we introduce LatentMorph, a novel framework that seamlessly integrates implicit latent reasoning into the T2I generation process. At its core, LatentMorph introduces four lightweight components: (i) a condenser for summarizing intermediate generation states into compact visual memory, (ii) a translator for converting latent thoughts into actionable guidance, (iii) a shaper for dynamically steering next image token predictions, and (iv) an RL-trained invoker for adaptively determining when to invoke reasoning. By performing reasoning entirely in continuous latent spaces, LatentMorph avoids the bottlenecks of explicit reasoning and enables more adaptive self-refinement. Extensive experiments demonstrate that LatentMorph (I) enhances the base model Janus-Pro by 16% on GenEval and 25% on T2I-CompBench; (II) outperforms explicit paradigms (e.g., TwiG) by 15% and 11% on abstract reasoning tasks like WISE and IPV-Txt, (III) while reducing inference time by 44% and token consumption by 51%; and (IV) exhibits 71% cognitive alignment with human intuition on reasoning invocation.

Cite this Paper

BibTeX
@InProceedings{pmlr-v306-chen26fv, title = {Show, Don’t Tell: Morphing Latent Reasoning into Image Generation}, author = {Chen, Harold Haodong and Yin, Xinxiang and Shu, Wen-Jie and Zhang, Hongfei and Zhang, Zixin and Liao, Chenfei and Guo, Litao and Chen, Qifeng and Chen, Ying-Cong}, booktitle = {Proceedings of the 43rd International Conference on Machine Learning}, pages = {17940--17959}, 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/chen26fv/chen26fv.pdf}, url = {https://proceedings.mlr.press/v306/chen26fv.html}, abstract = {Text-to-image (T2I) generation has achieved remarkable progress, yet existing methods often lack the ability to dynamically reason and refine during generation–a hallmark of human creativity. Current reasoning-augmented paradigms mostly rely on explicit thought processes, where intermediate reasoning is decoded into discrete text at fixed steps with frequent image decoding and re-encoding, leading to inefficiencies, information loss, and cognitive mismatches. To bridge this gap, we introduce LatentMorph, a novel framework that seamlessly integrates implicit latent reasoning into the T2I generation process. At its core, LatentMorph introduces four lightweight components: (i) a condenser for summarizing intermediate generation states into compact visual memory, (ii) a translator for converting latent thoughts into actionable guidance, (iii) a shaper for dynamically steering next image token predictions, and (iv) an RL-trained invoker for adaptively determining when to invoke reasoning. By performing reasoning entirely in continuous latent spaces, LatentMorph avoids the bottlenecks of explicit reasoning and enables more adaptive self-refinement. Extensive experiments demonstrate that LatentMorph (I) enhances the base model Janus-Pro by 16% on GenEval and 25% on T2I-CompBench; (II) outperforms explicit paradigms (e.g., TwiG) by 15% and 11% on abstract reasoning tasks like WISE and IPV-Txt, (III) while reducing inference time by 44% and token consumption by 51%; and (IV) exhibits 71% cognitive alignment with human intuition on reasoning invocation.} }
Endnote
%0 Conference Paper %T Show, Don’t Tell: Morphing Latent Reasoning into Image Generation %A Harold Haodong Chen %A Xinxiang Yin %A Wen-Jie Shu %A Hongfei Zhang %A Zixin Zhang %A Chenfei Liao %A Litao Guo %A Qifeng Chen %A Ying-Cong Chen %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-chen26fv %I PMLR %P 17940--17959 %U https://proceedings.mlr.press/v306/chen26fv.html %V 306 %X Text-to-image (T2I) generation has achieved remarkable progress, yet existing methods often lack the ability to dynamically reason and refine during generation–a hallmark of human creativity. Current reasoning-augmented paradigms mostly rely on explicit thought processes, where intermediate reasoning is decoded into discrete text at fixed steps with frequent image decoding and re-encoding, leading to inefficiencies, information loss, and cognitive mismatches. To bridge this gap, we introduce LatentMorph, a novel framework that seamlessly integrates implicit latent reasoning into the T2I generation process. At its core, LatentMorph introduces four lightweight components: (i) a condenser for summarizing intermediate generation states into compact visual memory, (ii) a translator for converting latent thoughts into actionable guidance, (iii) a shaper for dynamically steering next image token predictions, and (iv) an RL-trained invoker for adaptively determining when to invoke reasoning. By performing reasoning entirely in continuous latent spaces, LatentMorph avoids the bottlenecks of explicit reasoning and enables more adaptive self-refinement. Extensive experiments demonstrate that LatentMorph (I) enhances the base model Janus-Pro by 16% on GenEval and 25% on T2I-CompBench; (II) outperforms explicit paradigms (e.g., TwiG) by 15% and 11% on abstract reasoning tasks like WISE and IPV-Txt, (III) while reducing inference time by 44% and token consumption by 51%; and (IV) exhibits 71% cognitive alignment with human intuition on reasoning invocation.
APA
Chen, H.H., Yin, X., Shu, W., Zhang, H., Zhang, Z., Liao, C., Guo, L., Chen, Q. & Chen, Y.. (2026). Show, Don’t Tell: Morphing Latent Reasoning into Image Generation. Proceedings of the 43rd International Conference on Machine Learning, in Proceedings of Machine Learning Research 306:17940-17959 Available from https://proceedings.mlr.press/v306/chen26fv.html.

Related Material