ThinkRL-Edit: Thinking in Reinforcement Learning for Reasoning-Centric Image Editing

TL;DR

ThinkRL-Edit introduces a reasoning-centric reinforcement learning framework for image editing that enhances visual reasoning, explores multiple semantic hypotheses, and yields more accurate and coherent edits compared to prior methods.

Contribution

The paper proposes a novel RL framework that decouples reasoning from synthesis, uses Chain-of-Thought sampling, and introduces unbiased reward strategies for improved reasoning-centric image editing.

Findings

Outperforms prior methods on reasoning-centric image editing tasks.

Produces instruction-faithful, visually coherent, and semantically grounded edits.

Enhances reasoning exploration and reward precision in image editing.

Abstract

Instruction-driven image editing with unified multimodal generative models has advanced rapidly, yet their underlying visual reasoning remains limited, leading to suboptimal performance on reasoning-centric edits. Reinforcement learning (RL) has been investigated for improving the quality of image editing, but it faces three key challenges: (1) limited reasoning exploration confined to denoising stochasticity, (2) biased reward fusion, and (3) unstable VLM-based instruction rewards. In this work, we propose ThinkRL-Edit, a reasoning-centric RL framework that decouples visual reasoning from image synthesis and expands reasoning exploration beyond denoising. To the end, we introduce Chain-of-Thought (CoT)-based reasoning sampling with planning and reflection stages prior to generation in online sampling, compelling the model to explore multiple semantic hypotheses and validate their plausibility before committing to a visual outcome. To avoid the failures of weighted aggregation, we propose an unbiased chain preference grouping strategy across multiple reward dimensions. Moreover, we replace interval-based VLM scores with a binary checklist, yielding more precise, lower-variance, and interpretable rewards for complex reasoning. Experiments show our method significantly outperforms prior work on reasoning-centric image editing, producing instruction-faithful, visually coherent, and semantically grounded edits.