VeraRetouch: A Lightweight Fully Differentiable Framework for Multi-Task Reasoning Photo Retouching

2026-04-29

AI Summary

This work addresses the limitations of existing photo retouching methods, which rely on non-differentiable external tools, leading to optimization difficulties, parameter redundancy, and poor generalization. To overcome these issues, we propose a lightweight, end-to-end differentiable retouching framework that leverages a 0.5B-parameter vision-language model to interpret both image defects and semantic editing instructions. A fully differentiable Retouch Renderer enables pixel-level training, while decoupled control latent variables and inverse degradation-based data synthesis enhance model generalization. Our contributions include AetherRetouch-1M+, the first million-scale professional retouching dataset, the differentiable renderer itself, and DAPO-AE—a reinforcement learning–based post-training strategy. The proposed method achieves state-of-the-art performance across multiple benchmarks, with significantly reduced model size, enabling efficient multi-task inference and mobile deployment.

Abstract

Reasoning photo retouching has gained significant traction, requiring models to analyze image defects, give reasoning processes, and execute precise retouching enhancements. However, existing approaches often rely on non-differentiable external software, creating optimization barriers and suffering from high parameter redundancy and limited generalization. To address these challenges, we propose VeraRetouch, a lightweight and fully differentiable framework for multi-task photo retouching. We employ a 0.5B Vision-Language Model (VLM) as the central intelligence to formulate retouching plans based on instructions and scene semantics. Furthermore, we develop a fully differentiable Retouch Renderer that replaces external tools, enabling direct end-to-end pixel-level training through decoupled control latents for lighting, global color, and specific color adjustments. To overcome data scarcity, we introduce AetherRetouch-1M+, the first million-scale dataset for professional retouching, constructed via a new inverse degradation workflow. Furthermore, we propose DAPO-AE, a reinforcement learning post-training strategy that enhances autonomous aesthetic cognition. Extensive experiments demonstrate that VeraRetouch achieves state-of-the-art performance across multiple benchmarks while maintaining a significantly smaller footprint, enabling mobile deployment. Our code and models are publicly available at https://github.com/OpenVeraTeam/VeraRetouch.

Problem

• photo retouching
• differentiable framework
• multi-task reasoning
• optimization barriers
• generalization

Innovation

• fully differentiable rendering
• vision-language model
• multi-task photo retouching
• inverse degradation
• reinforcement learning post-training