Inline Critic Steers Image Editing
For practitioners of instruction-based image editing, this work provides a method to refine generation in real-time without requiring full model retraining or post-hoc correction, achieving significant performance gains.
The paper introduces Inline Critic, a learnable token that critiques and steers a frozen image-editing model's hidden states during a forward pass, achieving state-of-the-art results on GEdit-Bench (7.89), a +9.4 gain on RISEBench, and the strongest open-source result on KRIS-Bench (81.92, surpassing GPT-4o).
Instruction-based image editing exhibits heterogeneous difficulty not only across cases but also across regions of an image, motivating refinement approaches that allocate correction to where the model struggles. Existing refinement signals arrive late, after a fully generated image or a completed denoising step. We ask whether such a signal can act within an ongoing forward pass. To investigate this, we probe a frozen image-editing model and find that although generation capability emerges only in the last few layers, the error pattern is already set in early layers (rank correlation \r{ho} = 0.83 with the final-layer error map). Based on this, we introduce Inline Critic, a learnable token that critiques a frozen model's predictions at its intermediate layers and steers its hidden states to refine generation during the forward pass. A three-stage recipe is proposed to stabilize the training from learning how to critique to steering generation. As a result, we achieve state of the art on GEdit-Bench (7.89), a +9.4 gain on RISEBench over the same backbone, and the strongest open-source result on KRIS-Bench (81.92, surpassing GPT-4o). We further provide analyses showing that the critic genuinely shapes the model's attention and prediction updates at subsequent layers.