AI RESEARCH
Agentic Retoucher for Text-To-Image Generation
arXiv CS.AI
•
ArXi:2601.02046v3 Announce Type: replace-cross 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.