Making Image Editing Easier via Adaptive Task Reformulation with Agentic Executions
For practitioners using existing image editing models, this work offers a method to improve reliability without model modification, though it is incremental as it repurposes existing agents for task reformulation.
The paper identifies that many failures in instruction-guided image editing are due to poorly formulated tasks (e.g., small targets, implicit relations) rather than model capacity, and proposes an adaptive task reformulation framework using a MLLM agent. Experiments across multiple benchmarks and backbones show consistent improvements, with large gains on challenging cases.
Instruction guided image editing has advanced substantially with recent generative models, yet it still fails to produce reliable results across many seemingly simple cases. We observe that a large portion of these failures stem not from insufficient model capacity, but from poorly formulated editing tasks, such as those involving small targets, implicit spatial relations, or under-specified instructions. In this work, we frame image editing failures as a task formulation problem and propose an adaptive task reformulation framework that improves editing performance without modifying the underlying model. Our key idea is to transform the original image-instruction pair into a sequence of operations that are dynamically determined and executed by a MLLM agent through analysis, routing, reformulation, and feedback-driven refinement. Experiments on multiple benchmarks, including ImgEdit, PICA, and RePlan, across diverse editing backbones such as Qwen Image Edit and Nano Banana, show consistent improvements, with especially large gains on challenging cases. These results suggest that task reformulation is a critical but underexplored factor, and that substantial gains can be achieved by better matching editing tasks to the effective operating regime of existing models.