AI Paper Atlas

Research map / Image generation / Text-to-image diffusion

RubricRL: Simple Generalizable Rewards for Text-to-Image Generation

Xuelu Feng, Yunsheng Li, Ziyu Wan, Zixuan Gao, Junsong Yuan, Dongdong Chen, Chunming Qiao

ECCV 2026 · Image generation › Text-to-image diffusion · approach: T2I evaluation and controllable adapters · 9 citations (OpenAlex)

Read the paperCheck your idea against 137,217 papers

AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.

Measured effects reported in the paper

What was changedWhereMeasuredEffect
replace scalar/composite rewards with rubric rewardPhi3-3.8B AR GRPO · GenEvaloverall: 0.8080 (X-Omni) -> 0.8468; SFT 0.7624Improved
apply RubricRL via DiffusionNFTQwen-Image 4B · GenEvaloverall: 0.8406 -> 0.9091Improved
apply RubricRL via DiffusionNFTFLUX.1 Dev · GenEvaloverall: 0.8121 -> 0.8969Improved
apply RubricRLQwen2.5-0.5B · GenEval countingcounting: 0.2500 -> 0.2469Hurt
apply RubricRLQwen2.5-0.5B · DPG-Bench OtherOther: 84.27 -> 68.00Hurt
apply RubricRLPhi3-3.8B / Qwen2.5-0.5B · DPG-Benchoverall: 81.25 -> 86.07; 78.02 -> 81.43Improved
add hybrid rollout sampling (top-K + random) instead of none (ablation)RubricRL · GenEvaloverall: 0.8349 -> 0.8468Improved
use FFKC-1D or DAPO sampling instead of vanilla (ablation)RubricRL · GenEvaloverall: 0.8349 -> 0.8353 / 0.8333No clear effect
normalize advantage over retained N rollouts instead of all N' (ablation)RubricRL · GenEvaloverall: 0.8337 -> 0.8468Improved
use binary instead of continuous rubric scores (ablation)RubricRL · GenEvaloverall: 0.8170 -> 0.8468Improved
increase number of rubrics M from 4 to 10 (ablation)RubricRL · GenEvaloverall: 0.7866 -> 0.8468Improved
increase M from 10 to 12 (ablation)RubricRL · GenEvaloverall: 0.8468 -> 0.8427No clear effect

Closest papers by meaning

Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy