metadata
license: apache-2.0
pipeline_tag: image-to-image
VeraRetouch: A Lightweight Fully Differentiable Framework for Multi-Task Reasoning Photo Retouching
VeraRetouch is a lightweight and fully differentiable framework for multi-task reasoning photo retouching. It utilizes a 0.5B Vision-Language Model (VLM) to analyze image defects and formulate plans, which are then executed by a custom differentiable Retouch Renderer.
[Paper] [Project Page] [GitHub]
Overview
Existing photo retouching approaches often rely on non-differentiable external software, creating optimization barriers. VeraRetouch overcomes this with a fully differentiable Retouch Renderer, enabling direct end-to-end pixel-level training. It supports several modes:
- Auto Mode: Analyzes image defects and enhances them automatically.
- Style Mode: Retouches images based on user-provided text prompts.
- Param Mode: Executes retouching based on specific operator parameters.
Usage
Installation
# Clone the repository
git clone https://github.com/OpenVeraTeam/VeraRetouch.git
cd VeraRetouch
# Create and activate conda environment
conda create -n vera-retouch python=3.10
conda activate vera-retouch
pip install -r requirements.txt
Inference Examples
The model supports three inference modes. First, download the weights and place them in the ./checkpoints directory.
Auto Retouch Mode:
python inference.py --mode auto \
--model-path ./checkpoints/VeraRetouch \
--img_paths ./data_samples/input/sample_flower.jpg \
--save_dir ./data_samples/output/
Style Retouch Mode:
python inference.py --mode style \
--prompt "I want a dreamy bright pink style." \
--model-path ./checkpoints/VeraRetouch \
--img_paths ./data_samples/input/sample_flower.jpg \
--save_dir ./data_samples/output/
Param Retouch Mode:
python inference.py --mode style \
--instruction_path ./data_samples/param.json \
--model-path ./checkpoints/VeraRetouch \
--img_paths ./data_samples/input/sample_flower.jpg \
--save_dir ./data_samples/output/
Citation
@article{guo2026veraretouch,
title={VeraRetouch: A Lightweight Fully Differentiable Framework for Multi-Task Reasoning Photo Retouching},
author={Guo, Yihong and Lyu, Youwei and Tang, Jiajun and Zhou, Yizhuo and Wang, Hongliang and Chen, Jinwei and Zou, Changqing and Fan, Qingnan},
journal={arXiv preprint arXiv:2604.27375},
year={2026}
}