@inproceedings{liu-etal-2026-adaretouch,
title = "{A}da{R}etouch: An Interleaved Vision-Language Agent for Adaptive Multi-Turn Image Retouching",
author = "Liu, Ziang and
Zhang, Yongheng and
Wen, Qi and
Zhao, Kai and
Zhang, Jianwei and
Dong, Yuezhou and
Hao, Yiren and
Li, Bo and
Dai, Peishan and
Zhang, Hao",
editor = "Lee, Hung-Yi and
Dev, Sunipa and
Li, Jessy and
Elliott, Desmond",
booktitle = "Proceedings of the 2026 Conference on Empirical Methods in Natural Language Processing",
month = oct,
year = "2026",
address = "Budapest, Hungary",
publisher = "Association for Computational Linguistics",
url = "https://aclanthology.org/2026.emnlp-main.1565/",
pages = "35673--35692",
ISBN = "979-8-89176-459-0",
abstract = "Image retouching requires sophisticated aesthetic understanding and precise parameter control, yet existing agents typically rely on fixed, open-loop execution or software-specific supervision. We introduce AdaRetouch, an interleaved vision-language agent for adaptive, multi-turn image retouching. Given an instruction and an image, AdaRetouch dynamically diagnoses the visual state, invokes non-destructive tools, and iteratively observes the results until the intent is satisfied. To train this closed-loop behavior, we propose a software-agnostic pipeline that constructs trajectories featuring curriculum-driven tool sequences and step-wise reasoning. AdaRetouch is initialized via supervised fine-tuning and further optimized using process-level reinforcement learning for quality and efficiency. Experiments on AdaRetouch-Bench show that AdaRetouch outperforms general MLLMs, existing agents, and diffusion editors in intent alignment, perceptual quality, and calibrated tool planning."
}<?xml version="1.0" encoding="UTF-8"?>
<modsCollection xmlns="http://www.loc.gov/mods/v3">
<mods ID="liu-etal-2026-adaretouch">
<titleInfo>
<title>AdaRetouch: An Interleaved Vision-Language Agent for Adaptive Multi-Turn Image Retouching</title>
</titleInfo>
<name type="personal">
<namePart type="given">Ziang</namePart>
<namePart type="family">Liu</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Yongheng</namePart>
<namePart type="family">Zhang</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Qi</namePart>
<namePart type="family">Wen</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Kai</namePart>
<namePart type="family">Zhao</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Jianwei</namePart>
<namePart type="family">Zhang</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Yuezhou</namePart>
<namePart type="family">Dong</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Yiren</namePart>
<namePart type="family">Hao</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Bo</namePart>
<namePart type="family">Li</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Peishan</namePart>
<namePart type="family">Dai</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Hao</namePart>
<namePart type="family">Zhang</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<originInfo>
<dateIssued>2026-10</dateIssued>
</originInfo>
<typeOfResource>text</typeOfResource>
<relatedItem type="host">
<titleInfo>
<title>Proceedings of the 2026 Conference on Empirical Methods in Natural Language Processing</title>
</titleInfo>
<name type="personal">
<namePart type="given">Hung-Yi</namePart>
<namePart type="family">Lee</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Sunipa</namePart>
<namePart type="family">Dev</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Jessy</namePart>
<namePart type="family">Li</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Desmond</namePart>
<namePart type="family">Elliott</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<originInfo>
<publisher>Association for Computational Linguistics</publisher>
<place>
<placeTerm type="text">Budapest, Hungary</placeTerm>
</place>
</originInfo>
<genre authority="marcgt">conference publication</genre>
<identifier type="isbn">979-8-89176-459-0</identifier>
</relatedItem>
<abstract>Image retouching requires sophisticated aesthetic understanding and precise parameter control, yet existing agents typically rely on fixed, open-loop execution or software-specific supervision. We introduce AdaRetouch, an interleaved vision-language agent for adaptive, multi-turn image retouching. Given an instruction and an image, AdaRetouch dynamically diagnoses the visual state, invokes non-destructive tools, and iteratively observes the results until the intent is satisfied. To train this closed-loop behavior, we propose a software-agnostic pipeline that constructs trajectories featuring curriculum-driven tool sequences and step-wise reasoning. AdaRetouch is initialized via supervised fine-tuning and further optimized using process-level reinforcement learning for quality and efficiency. Experiments on AdaRetouch-Bench show that AdaRetouch outperforms general MLLMs, existing agents, and diffusion editors in intent alignment, perceptual quality, and calibrated tool planning.</abstract>
<identifier type="citekey">liu-etal-2026-adaretouch</identifier>
<location>
<url>https://aclanthology.org/2026.emnlp-main.1565/</url>
</location>
<part>
<date>2026-10</date>
<extent unit="page">
<start>35673</start>
<end>35692</end>
</extent>
</part>
</mods>
</modsCollection>
%0 Conference Proceedings
%T AdaRetouch: An Interleaved Vision-Language Agent for Adaptive Multi-Turn Image Retouching
%A Liu, Ziang
%A Zhang, Yongheng
%A Wen, Qi
%A Zhao, Kai
%A Zhang, Jianwei
%A Dong, Yuezhou
%A Hao, Yiren
%A Li, Bo
%A Dai, Peishan
%A Zhang, Hao
%Y Lee, Hung-Yi
%Y Dev, Sunipa
%Y Li, Jessy
%Y Elliott, Desmond
%S Proceedings of the 2026 Conference on Empirical Methods in Natural Language Processing
%D 2026
%8 October
%I Association for Computational Linguistics
%C Budapest, Hungary
%@ 979-8-89176-459-0
%F liu-etal-2026-adaretouch
%X Image retouching requires sophisticated aesthetic understanding and precise parameter control, yet existing agents typically rely on fixed, open-loop execution or software-specific supervision. We introduce AdaRetouch, an interleaved vision-language agent for adaptive, multi-turn image retouching. Given an instruction and an image, AdaRetouch dynamically diagnoses the visual state, invokes non-destructive tools, and iteratively observes the results until the intent is satisfied. To train this closed-loop behavior, we propose a software-agnostic pipeline that constructs trajectories featuring curriculum-driven tool sequences and step-wise reasoning. AdaRetouch is initialized via supervised fine-tuning and further optimized using process-level reinforcement learning for quality and efficiency. Experiments on AdaRetouch-Bench show that AdaRetouch outperforms general MLLMs, existing agents, and diffusion editors in intent alignment, perceptual quality, and calibrated tool planning.
%U https://aclanthology.org/2026.emnlp-main.1565/
%P 35673-35692
Markdown (Informal)
[AdaRetouch: An Interleaved Vision-Language Agent for Adaptive Multi-Turn Image Retouching](https://aclanthology.org/2026.emnlp-main.1565/) (Liu et al., EMNLP 2026)
ACL
- Ziang Liu, Yongheng Zhang, Qi Wen, Kai Zhao, Jianwei Zhang, Yuezhou Dong, Yiren Hao, Bo Li, Peishan Dai, and Hao Zhang. 2026. AdaRetouch: An Interleaved Vision-Language Agent for Adaptive Multi-Turn Image Retouching. In Proceedings of the 2026 Conference on Empirical Methods in Natural Language Processing, pages 35673–35692, Budapest, Hungary. Association for Computational Linguistics.