ETCHR: Editing To Clarify and Harness Reasoning

2026-05-22

AI Summary

Existing multimodal large language models are limited in fine-grained visual reasoning tasks by their reliance on text-only chain-of-thought reasoning, while “thinking-in-images” approaches often underperform due to fixed toolsets or the generation of noisy images. This work proposes a decoupled, question-conditioned image editor that, for the first time, separates the dedicated editing model from the downstream reasoning model, enabling plug-and-play compatibility. The framework employs a two-stage training strategy: Reasoning Imitation, where supervised fine-tuning learns human-like editing trajectories, followed by Reasoning Enhancement, which leverages reward signals from vision-language models via reinforcement learning to bridge the gap between linguistic intent and visual generation fidelity. Evaluated across five benchmark tasks, the method yields substantial performance gains, improving average Pass@1 by 4.82, 5.47, and 4.61 percentage points on Qwen3-VL-8B, Gemini-3.1-Flash-Lite, and Kimi K2.5, respectively.

Abstract

Multimodal Large Language Models have advanced visual reasoning, yet a purely textual chain of thought remains a bottleneck for questions that require fine-grained focus or view transformations. The ''think with images'' paradigm narrows this gap, but existing approaches are either constrained by fixed predefined toolkits or produce noisy intermediate images from unified multimodal methods. We pursue a third option: using a dedicated image editing model and decouple it with an understanding model. However, off-the-shelf image editors fail as reasoning assistants with two complementary gaps: a language-side gap, where editors trained as passive instruction-followers cannot map an abstract question to an appropriate visual transformation, and a generation-side gap, where edit correctness degrades as reasoning depth grows. Guided by this analysis, we introduce ETCHR (Editing To Clarify and Harness Reasoning), a question-conditioned, reasoning-aware image editor decoupled from the downstream understanding model and trained with a two-stage recipe targeted at the two gaps: Reasoning Imitation via supervised fine-tuning on edit trajectories, followed by Reasoning Enhancement with VLM-derived rewards for edit correctness and downstream reasoning accuracy. Since the editor is decoupled, ETCHR plugs into different open- and closed-source MLLMs in a training-free manner. Across five task families (fine-grained perception, chart understanding, logic reasoning, jigsaw restoration, and 3D understanding), ETCHR raises average Pass@1 from 55.95 to 60.77 (+4.82) with Qwen3-VL-8B, from 65.08 to 70.55 (+5.47) with Gemini-3.1-Flash-Lite, and from 76.55 to 81.16 (+4.61) with the 1T-parameter MoE model Kimi K2.5.

Problem

• visual reasoning
• multimodal large language models
• image editing
• chain of thought
• reasoning bottleneck

Innovation

• reasoning-aware editing
• decoupled image editor
• multimodal reasoning
• two-stage training
• visual chain of thought