Agentic Retoucher for Text-To-Image Generation

TL;DR

This paper introduces Agentic Retoucher, a hierarchical framework that improves text-to-image generation by localizing and correcting distortions through perception, reasoning, and action modules, leading to higher quality and more reliable images.

Contribution

It presents a novel decision-driven correction framework with perception, reasoning, and action components, and introduces a new dataset for evaluating artifact correction in T2I models.

Findings

Outperforms state-of-the-art methods in perceptual quality.

Achieves better distortion localization and human preference alignment.

Establishes a new paradigm for self-corrective T2I generation.

Abstract

Text-to-image (T2I) diffusion models such as SDXL and FLUX have achieved impressive photorealism, yet small-scale distortions remain pervasive in limbs, face, text and so on. Existing refinement approaches either perform costly iterative re-generation or rely on vision-language models (VLMs) with weak spatial grounding, leading to semantic drift and unreliable local edits. To close this gap, we propose Agentic Retoucher, a hierarchical decision-driven framework that reformulates post-generation correction as a human-like perception-reasoning-action loop. Specifically, we design (1) a perception agent that learns contextual saliency for fine-grained distortion localization under text-image consistency cues, (2) a reasoning agent that performs human-aligned inferential diagnosis via progressive preference alignment, and (3) an action agent that adaptively plans localized inpainting guided by user preference. This design integrates perceptual evidence, linguistic reasoning, and controllable correction into a unified, self-corrective decision process. To enable fine-grained supervision and quantitative evaluation, we further construct GenBlemish-27K, a dataset of 6K T2I images with 27K annotated artifact regions across 12 categories. Extensive experiments demonstrate that Agentic Retoucher consistently outperforms state-of-the-art methods in perceptual quality, distortion localization and human preference alignment, establishing a new paradigm for self-corrective and perceptually reliable T2I generation.