Learning Native Reflection in Unified Models with Interleaved Reinforcement Learning
UMM-Reflection trains one unified model with interleaved RL so reflection and image revisions improve together.
UMM-Reflection applies reinforcement learning to complete diagnose-and-revise trajectories inside a single unified multimodal model, so one advantage updates both reflection text and flow-based image edits. Sibling trajectories share an initial image, letting group-relative advantages compare repair strategies without an external verifier at inference. On BAGEL it improves GenEval by 12.05 points over supervised fine-tuning, with transferred gains of +10.97 on WISE, +3.48 on OneIG-Bench, and +4.63 on T2I-CompBench++.
- Applies group-relative RL to full reflection and revision trajectories in one model.
- One trajectory-level advantage updates both reflection tokens and flow-based revisions.
- On BAGEL, GenEval rises 12.05 points over supervised fine-tuning.
- Gains transfer to WISE, OneIG-Bench, and T2I-CompBench++ without extra training.
- AI models
- BAGELUMM-Reflection
From the original article
1 min read at huggingface.co
Unified multimodal models can both look at and render images, so in principle they can repair their own generations: diagnose what an image gets wrong, revise it, observe the result, and diagnose again. Whether a revision helps is known only after it is rendered, so the reflection text and the image generation must be learned jointly, over the whole loop.
By Yijia Fan, Ziqi Huang, Zhongang Cai · huggingface.co
A short excerpt; the article and its images belong to the publisher. Summary, key facts and assessment above are AI-written from the article and can be wrong.