Mitigating Content Shift and Hallucination in GenAI Image Editing via Structural Refinement

2026-05-28

AI Summary

This work addresses the degradation of pixel-level fidelity in generative AI (GenAI) image editors, which often suffer from spatial misalignment, texture distortion, and content hallucination—issues that critically impair downstream tasks requiring strict structural consistency. To tackle this, the paper introduces, for the first time, the problem of “structure-preserving GenAI fusion” and proposes a black-box post-processing framework that seamlessly integrates the original image with its GenAI-enhanced counterpart without relying on model priors. By modeling coarse-grained spatial and photometric correspondences and incorporating a multi-stage fusion strategy guided by structural constraints, the method effectively transfers visual enhancements while suppressing hallucinated content. Compared to existing style transfer and image fusion approaches, the proposed solution significantly improves pixel-level structural consistency while preserving input resolution and aesthetic quality.

Abstract

Generative AI (GenAI) image editors, such as Nano Banana, produce visually compelling results for retouching tasks, enabling non-experts to edit images through text prompts alone. However, the generative nature of these models often introduces spatial misalignment, texture distortion, and content hallucination, all of which are detrimental to downstream workflows that require pixel-level fidelity. We identify a problem setting we call "structure-preserving GenAI fusion" for black-box GenAI image retouching: retain the perceptual enhancements of a GenAI output while enforcing structural faithfulness to the original input image. To address this problem, we propose a post-processing framework that fuses an input image with its GenAI-enhanced counterpart by first establishing coarse spatial and photometric correspondences, then performing a fusion stage that transfers desired enhancements while suppressing hallucinated content. In the absence of direct prior work in this setting, we evaluate our framework against representative methods from photorealistic style transfer and image fusion. Our experiments demonstrate that our method better preserves aesthetic quality while maintaining pixel-level structural consistency and the input resolution.

Problem

• content hallucination
• structural consistency
• GenAI image editing
• spatial misalignment
• texture distortion

Innovation

• structure-preserving fusion
• GenAI image editing
• content hallucination mitigation
• spatial alignment
• photometric correspondence