PhysEdit: Physically-Consistent Region-Aware Image Editing via Adaptive Spatio-Temporal Reasoning

2026-05-01

AI Summary

Existing reasoning-based image editing methods rely on fixed strategies, limiting their ability to adapt to variations in spatial scope and reasoning depth across different instructions. This work proposes PhysEdit, a novel framework that introduces conditional computation into the image editing reasoning process for the first time. PhysEdit dynamically adjusts both the number of reasoning steps and the regions of focus through two plug-and-play inference-time modules: Complexity-Adaptive Reasoning Depth (CARD) and Spatial Reasoning Mask (SRM). By leveraging cross-attention to extract instruction-conditioned spatial priors, the method achieves efficient and physically consistent edits. On the ImgEdit benchmark, PhysEdit accelerates overall inference by 1.18× (64.3s vs. 76.1s), improves instruction adherence by 0.7%, and maintains identity preservation on par with the baseline, with appearance-related edits achieving up to a 1.52× speedup.

Abstract

Image editing instructions are heterogeneous: a color swap, an object insertion, and a physical-action edit all demand different spatial coverage and different reasoning depth, yet existing reasoning-based editors apply a single fixed inference recipe to every instruction. We argue that adaptivity along both the spatial and temporal axes is the missing degree of freedom, and we present PhysEdit, an editing framework built around this principle. PhysEdit introduces two inference-time modules that compose without retraining the backbone. At its core, (1) Complexity-Adaptive Reasoning Depth (CARD) predicts edit complexity directly from the instruction and reference image and allocates the reasoning step count N_r and reasoning-token length r per sample -- turning a previously fixed inference schedule into a conditional-computation problem. CARD is supported by (2) a Spatial Reasoning Mask (SRM) that extracts an instruction-conditioned spatial prior from cross-attention to confine reasoning to regions that semantically require it. On the full 737-case ImgEdit Basic-Edit Suite, PhysEdit delivers a 1.18x wall-clock speedup (64.3s vs. 76.1s per sample) over a strong reasoning baseline while slightly improving instruction adherence (CLIP-T 0.2283 vs. 0.2266, +0.7%) and matching identity preservation within noise (CLIP-I 0.8246 vs. 0.8280). The speedup is category-dependent and reaches 1.52x on appearance-level edits, validating CARD's adaptive allocation as the principal source of efficiency gain. A 30-sample pilot with full ablations isolates the contribution of each module.

Problem

• image editing
• adaptive reasoning
• spatio-temporal reasoning
• physically-consistent editing
• instruction heterogeneity

Innovation

• adaptive reasoning
• spatio-temporal reasoning
• region-aware editing
• conditional computation
• physically-consistent editing