@inproceedings{yang-etal-2026-genesis,
title = "Genesis: Visual Augmentation Primitives for Multimodal Chain-of-Thought Reasoning",
author = "Yang, Fan and
Chen, Lifeng and
Wang, Yipu and
Wang, Han and
Chen, Feilong and
Chen, Zhiyang and
Ma, Yidong and
Zhao, Chaoyang and
Jiang, Hanqi and
Chen, Yibo and
Tang, Xu and
Hu, Yao and
Tang, Ming and
Jin, Xiaojie and
Zheng, Xiaolong and
Wang, Jinqiao",
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.2347/",
pages = "53310--53328",
ISBN = "979-8-89176-459-0",
abstract = "Recent multimodal large language models achieve promising results on visual reasoning tasks, approaching human-level performance on existing STEM benchmarks. However, these capabilities degrade on complex real-world scenes. We identify two intertwined causes: chain-of-thought reasoning remains largely confined to the linguistic space, and training data rarely requires each step to be anchored in visual evidence. In this work, we introduce \textbf{Visual Augmentation Primitives}, which leverages high-quality instruction-guided generation models to synthesize diverse photorealistic training samples and construct coherent multimodal reasoning chains throughout the generation process, requiring models to unify visual perception with logical reasoning at every step. We further introduce multimodal chain-of-thought rubric-based reinforcement learning, which extends rubric-based reward design from the text-only domain to visual reasoning, scoring intermediate reasoning steps for logical validity, completeness, and consistency, and construct the UNREAL benchmark to systematically evaluate visual reasoning on hyperrealistic scenes across controllable difficulty levels. Experiments show that a 4B model built on this framework surpasses 8B baselines, with the approach scaling consistently to 32B across visual reasoning and VQA tasks. All models, data, and code will be publicly released."
}<?xml version="1.0" encoding="UTF-8"?>
<modsCollection xmlns="http://www.loc.gov/mods/v3">
<mods ID="yang-etal-2026-genesis">
<titleInfo>
<title>Genesis: Visual Augmentation Primitives for Multimodal Chain-of-Thought Reasoning</title>
</titleInfo>
<name type="personal">
<namePart type="given">Fan</namePart>
<namePart type="family">Yang</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Lifeng</namePart>
<namePart type="family">Chen</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Yipu</namePart>
<namePart type="family">Wang</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Han</namePart>
<namePart type="family">Wang</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Feilong</namePart>
<namePart type="family">Chen</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Zhiyang</namePart>
<namePart type="family">Chen</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Yidong</namePart>
<namePart type="family">Ma</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Chaoyang</namePart>
<namePart type="family">Zhao</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Hanqi</namePart>
<namePart type="family">Jiang</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Yibo</namePart>
<namePart type="family">Chen</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Xu</namePart>
<namePart type="family">Tang</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Yao</namePart>
<namePart type="family">Hu</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Ming</namePart>
<namePart type="family">Tang</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Xiaojie</namePart>
<namePart type="family">Jin</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Xiaolong</namePart>
<namePart type="family">Zheng</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Jinqiao</namePart>
<namePart type="family">Wang</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>Recent multimodal large language models achieve promising results on visual reasoning tasks, approaching human-level performance on existing STEM benchmarks. However, these capabilities degrade on complex real-world scenes. We identify two intertwined causes: chain-of-thought reasoning remains largely confined to the linguistic space, and training data rarely requires each step to be anchored in visual evidence. In this work, we introduce Visual Augmentation Primitives, which leverages high-quality instruction-guided generation models to synthesize diverse photorealistic training samples and construct coherent multimodal reasoning chains throughout the generation process, requiring models to unify visual perception with logical reasoning at every step. We further introduce multimodal chain-of-thought rubric-based reinforcement learning, which extends rubric-based reward design from the text-only domain to visual reasoning, scoring intermediate reasoning steps for logical validity, completeness, and consistency, and construct the UNREAL benchmark to systematically evaluate visual reasoning on hyperrealistic scenes across controllable difficulty levels. Experiments show that a 4B model built on this framework surpasses 8B baselines, with the approach scaling consistently to 32B across visual reasoning and VQA tasks. All models, data, and code will be publicly released.</abstract>
<identifier type="citekey">yang-etal-2026-genesis</identifier>
<location>
<url>https://aclanthology.org/2026.emnlp-main.2347/</url>
</location>
<part>
<date>2026-10</date>
<extent unit="page">
<start>53310</start>
<end>53328</end>
</extent>
</part>
</mods>
</modsCollection>
%0 Conference Proceedings
%T Genesis: Visual Augmentation Primitives for Multimodal Chain-of-Thought Reasoning
%A Yang, Fan
%A Chen, Lifeng
%A Wang, Yipu
%A Wang, Han
%A Chen, Feilong
%A Chen, Zhiyang
%A Ma, Yidong
%A Zhao, Chaoyang
%A Jiang, Hanqi
%A Chen, Yibo
%A Tang, Xu
%A Hu, Yao
%A Tang, Ming
%A Jin, Xiaojie
%A Zheng, Xiaolong
%A Wang, Jinqiao
%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 yang-etal-2026-genesis
%X Recent multimodal large language models achieve promising results on visual reasoning tasks, approaching human-level performance on existing STEM benchmarks. However, these capabilities degrade on complex real-world scenes. We identify two intertwined causes: chain-of-thought reasoning remains largely confined to the linguistic space, and training data rarely requires each step to be anchored in visual evidence. In this work, we introduce Visual Augmentation Primitives, which leverages high-quality instruction-guided generation models to synthesize diverse photorealistic training samples and construct coherent multimodal reasoning chains throughout the generation process, requiring models to unify visual perception with logical reasoning at every step. We further introduce multimodal chain-of-thought rubric-based reinforcement learning, which extends rubric-based reward design from the text-only domain to visual reasoning, scoring intermediate reasoning steps for logical validity, completeness, and consistency, and construct the UNREAL benchmark to systematically evaluate visual reasoning on hyperrealistic scenes across controllable difficulty levels. Experiments show that a 4B model built on this framework surpasses 8B baselines, with the approach scaling consistently to 32B across visual reasoning and VQA tasks. All models, data, and code will be publicly released.
%U https://aclanthology.org/2026.emnlp-main.2347/
%P 53310-53328
Markdown (Informal)
[Genesis: Visual Augmentation Primitives for Multimodal Chain-of-Thought Reasoning](https://aclanthology.org/2026.emnlp-main.2347/) (Yang et al., EMNLP 2026)
ACL
- Fan Yang, Lifeng Chen, Yipu Wang, Han Wang, Feilong Chen, Zhiyang Chen, Yidong Ma, Chaoyang Zhao, Hanqi Jiang, Yibo Chen, Xu Tang, Yao Hu, Ming Tang, Xiaojie Jin, Xiaolong Zheng, and Jinqiao Wang. 2026. Genesis: Visual Augmentation Primitives for Multimodal Chain-of-Thought Reasoning. In Proceedings of the 2026 Conference on Empirical Methods in Natural Language Processing, pages 53310–53328, Budapest, Hungary. Association for Computational Linguistics.