MSRAMIE: Multimodal Structured Reasoning Agent for Multi-instruction Image Editing

TL;DR

MSRAMIE is a training-free multimodal reasoning framework that enhances multi-instruction image editing by decomposing complex tasks into iterative steps, improving instruction following and output quality without additional training.

Contribution

It introduces a novel reasoning topology with Tree-of-States and Graph-of-References, enabling systematic, interpretable multi-step image editing without retraining.

Findings

Improves instruction following accuracy by over 15% with complex instructions.

Achieves over 100% success rate in completing all modifications in a single run.

Maintains perceptual quality and visual consistency in edited images.

Abstract

Existing instruction-based image editing models perform well with simple, single-step instructions but degrade in realistic scenarios that involve multiple, lengthy, and interdependent directives. A main cause is the scarcity of training data with complex multi-instruction annotations. However, it is costly to collect such data and retrain these models. To address this challenge, we propose MSRAMIE, a training-free agent framework built on Multimodal Large Language Model (MLLM). MSRAMIE takes existing editing models as plug-in components and handle multi-instruction tasks via structured multimodal reasoning. It orchestrates iterative interactions between an MLLM-based Instructor and an image editing Actor, introducing a novel reasoning topology that comprises the proposed Tree-of-States and Graph-of-References. During inference, complex instructions are decomposed into multiple editing steps which enable state transitions, cross-step information aggregation, and original input recall, which enables systematic exploration of the image editing space and flexible progressive output refinement. The visualizable inference topology further provides interpretable and controllable decision pathways. Experiments show that as the instruction complexity increases, MSRAMIE can improve instruction following over 15% and increases the probability of finishing all modifications in a single run over 100%, while preserving perceptual quality and maintaining visual consistency.