AI Paper Atlas

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

PaCo-RL: Advancing Reinforcement Learning for Consistent Image Generation with Pairwise Reward Modeling

Bowen Ping, Chengyou Jia, Minnan Luo, Changliang Xia, Xin Shen, Zhuohang Dang, Hangwei Qian

CVPR 2026 · Image generation › Text-to-image diffusion · approach: Preference optimization for diffusion · 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
add resolution-decoupled training (0.5x) (ablation)PaCo-GRPO FLUX.1-dev · T2IS ablationPrompt following: 0.698 -> 0.728Improved
add resolution-decoupled training (0.5x) (ablation)PaCo-GRPO FLUX.1-dev · T2IS ablationTraining time: 12.0h -> 6.0hImproved
add resolution-decoupled training (ablation)PaCo-GRPO FLUX.1-dev · T2IS ablationVisual consistency: 0.452 -> 0.493Improved
add log-tamed reward aggregation (ablation)PaCo-GRPO FLUX.1-dev · T2IS ablationAesthetics: 0.471 -> 0.555Improved
add log-tamed reward aggregation (ablation)PaCo-GRPO FLUX.1-dev · T2IS ablationVisual consistency: 0.557 -> 0.493Hurt
add log-tamed reward aggregation (ablation)PaCo-GRPO FLUX.1-dev · reward ratio during trainingConsistency/Prompt Alignment ratio: above 2.5 -> below 1.8Improved
train at 256x256 instead of 512x512 (ablation)PaCo-GRPO FLUX.1-dev · T2IS trainingrewardHurt
train at 512x512 instead of 1024x1024 (ablation)PaCo-GRPO FLUX.1-dev · T2IS trainingrewardNo clear effect
fine-tune Qwen2.5-VL-7B on PaCo-Dataset into PaCo-Reward-7BQwen2.5-VL-7B · ConsistencyRankAccuracy: 0.344 -> 0.449Improved
fine-tune Qwen2.5-VL-7B on PaCo-DatasetQwen2.5-VL-7B · EditReward-BenchOverall accuracy: 0.432 -> 0.751Improved
train with reasoning-augmented labels instead of binary labels onlyPaCo-Reward-7B · ConsistencyRankAccuracy: 0.441 -> 0.449Improved
add PaCo-Reward-7B RL trainingFLUX.1-Kontext · GEdit-Bench EN-IOverall: 6.143 -> 6.636Improved

Earlier methods this paper compares with

Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.

Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.

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