EditThinker: Unlocking Iterative Reasoning for Any Image Editor

2025-12-05

AI Summary

Existing instruction-driven image editing methods generate aesthetically high-quality images but suffer from low instruction adherence and single-shot success rates due to inherent stochasticity in one-step inference and shallow semantic understanding. Method: We propose “Think-Edit”, the first iterative reasoning framework for image editing that integrates multi-step, multimodal reasoning within a single large vision-language model. It unifies critical evaluation, causal reasoning, and instruction refinement, and jointly aligns the chain-of-thought with editing actions via supervised fine-tuning and reinforcement learning. Contribution/Results: Our “critique–refine–generate” closed-loop mechanism significantly improves instruction following across four benchmarks. It boosts average single-shot editing success rate by 23.6% and user satisfaction by 31.4%, establishing a new paradigm for controllable, cognition-aware image editing.

Technology Category

• Computer Vision: Large Vision Models
• Machine Learning: Large Multimodal Models (LMMs)
• Cognitive Modeling & Cognitive Systems: Conceptual Inference and Reasoning

Application Category

• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking
• Economics, Online Markets and Human Computation: LLM based quality controls for crowd work
• User Modeling, Personalization and Recommendation: Fairness-aware retrieval and ranking

Abstract

Instruction-based image editing has emerged as a prominent research area, which, benefiting from image generation foundation models, have achieved high aesthetic quality, making instruction-following capability the primary challenge. Existing approaches improve instruction adherence via supervised or reinforcement learning, yet single-turn success rates remain limited due to inherent stochasticity and a lack of deliberation. In this work, we propose a deliberative editing framework to 'think' while they edit, which simulates the human cognitive loop by iteratively executing a Think-while-Edit cycle: Critiquing results and Refining instructions , followed by Repeating the generation until satisfactory. Specifically, we train a single MLLM, EditThinker, to act as the reasoning engine of this framework, which jointly produce the critique score, reasoning process, and refined instructions. We employ reinforcement learning to align the EditThinker's thinking with its editing, thereby generating more targeted instruction improvements. Extensive experiments on four benchmarks demonstrate that our approach significantly improves the instruction-following capability of any image editing model by a large margin. We will release our data construction framework, datasets, and models to benefit the community.

Problem

• Improving instruction adherence in image editing models
• Addressing single-turn limitations via iterative reasoning cycles
• Enhancing deliberation in instruction-based image generation systems

Innovation

• Iterative Think-while-Edit cycle for deliberation
• Single MLLM jointly produces critique and refined instructions
• Reinforcement learning aligns thinking with editing actions