Taming I2V models for Image HOI Editing: A Cognitive Benchmark and Agentic Self-Correcting Framework

Current image editing methods excels at static attributes but fails at complex Human-Object Interactions (HOI), a critical challenge unaddressed by existing benchmarks that conflate HOI with static attributes, relying on global metrics incapable of simultaneously assessing dynamic interaction validity and entangled human-object pair preservation. Thus, we first introduce HOI-Edit, a comprehensive benchmark with three progressive cognitive levels, which features an automated metric HOI-Eval that first reliably evaluates instance-level interaction by letting VLM Q&A after thinking with images containing grounded Human-Object pair. Considering the task's essence of remodeling dynamic relationships, we benchmark Image-to-Video (I2V) models, finding them inherently suited for dynamic editing due to their temporal generation capabilities. Crucially, beyond superior performance, this capability provides a "replay of the failure process", offering unique diagnosability into why errors occur. We thus propose SCPE (Self-Correcting Process Editing), a novel, agentic self-correcting framework that constrains the generation of I2V models through iteratively refined prompts, enabling the generated videos to more accurately present the target HOI. Extracted frames from these videos are the final editing results. On HOI-Edit, SCPE achieves performance competitive with state-of-the-art (SOTA) editing models like Nano Banana on interaction.Code is available at https://github.com/oceanflowlab/HOI-Edit.

Editing images is easy when the request changes colors or objects, but much harder when it asks a person to interact with something, such as picking up an apple, riding a skateboard, or opening a box before cleaning it. These edits require the model to preserve the original person and object while understanding where they are, how the action should unfold, and what physical changes should follow. This paper introduces HOI-Edit, a benchmark for testing whether image editing models can handle human-object interactions. It also proposes HOI-Eval, an evaluation method that checks the specific person-object pair instead of judging only the whole image. The paper finds that image-to-video models are useful because their generated motion reveals how an interaction succeeds or fails. Based on this insight, the authors build SCPE, a self-correcting framework that watches the generated video, diagnoses mistakes, improves the prompt, and selects the best final frame as the edited image. Experiments show that SCPE improves interaction editing, spatial understanding, and causal reasoning, while remaining competitive with strong commercial image editors. This work suggests that better image editing may require models to simulate actions as processes, not just repaint pixels.