@inproceedings{rassin-etal-2026-quantifying,
title = "Quantifying and Mitigating Default Behaviors in Text-to-Image Models",
author = "Rassin, Royi and
Slobodkin, Aviv and
Ravfogel, Shauli and
Elazar, Yanai and
Goldberg, Yoav",
editor = "Lee, Hung-Yi and
Dev, Sunipa and
Li, Jessy and
Elliott, Desmond",
booktitle = "Findings of the {A}ssociation for {C}omputational {L}inguistics: {EMNLP} 2026",
month = oct,
year = "2026",
address = "Budapest, Hungary",
publisher = "Association for Computational Linguistics",
url = "https://aclanthology.org/2026.findings-emnlp.1596/",
pages = "33994--34024",
ISBN = "979-8-89176-462-0",
abstract = "Text-to-image (T2I) models consistently collapse to stereotypical attribute values when prompts leave concepts underspecified, a phenomenon we term \textit{default behaviors}. We conduct a large-scale empirical study using , an LLM-driven measurement approach that quantifies diversity at the level of individual concept attributes. We find pervasive default behaviors across all evaluated T2I models, with stronger models tending to exhibit less diversity. We trace these defaults to reporting bias in training captions: when captions omit common attributes, the accompanying images are dominated by a narrow set of values, and model outputs closely mirror this skew. We confirm this link causally with controlled fine-tuning interventions. As a proof of concept, we show that making sampled attribute values explicit broadens outputs, with some loss of faithfulness to secondary prompt details. Together, our results recast low diversity not as an inherent limitation of T2I models, but as a measurable and steerable consequence of what their training captions leave unsaid. Code and data available at: https://royira.github.io/GRADE."
}<?xml version="1.0" encoding="UTF-8"?>
<modsCollection xmlns="http://www.loc.gov/mods/v3">
<mods ID="rassin-etal-2026-quantifying">
<titleInfo>
<title>Quantifying and Mitigating Default Behaviors in Text-to-Image Models</title>
</titleInfo>
<name type="personal">
<namePart type="given">Royi</namePart>
<namePart type="family">Rassin</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Aviv</namePart>
<namePart type="family">Slobodkin</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Shauli</namePart>
<namePart type="family">Ravfogel</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Yanai</namePart>
<namePart type="family">Elazar</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Yoav</namePart>
<namePart type="family">Goldberg</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>Findings of the Association for Computational Linguistics: EMNLP 2026</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">979-8-89176-462-0</identifier>
</relatedItem>
<abstract>Text-to-image (T2I) models consistently collapse to stereotypical attribute values when prompts leave concepts underspecified, a phenomenon we term default behaviors. We conduct a large-scale empirical study using , an LLM-driven measurement approach that quantifies diversity at the level of individual concept attributes. We find pervasive default behaviors across all evaluated T2I models, with stronger models tending to exhibit less diversity. We trace these defaults to reporting bias in training captions: when captions omit common attributes, the accompanying images are dominated by a narrow set of values, and model outputs closely mirror this skew. We confirm this link causally with controlled fine-tuning interventions. As a proof of concept, we show that making sampled attribute values explicit broadens outputs, with some loss of faithfulness to secondary prompt details. Together, our results recast low diversity not as an inherent limitation of T2I models, but as a measurable and steerable consequence of what their training captions leave unsaid. Code and data available at: https://royira.github.io/GRADE.</abstract>
<identifier type="citekey">rassin-etal-2026-quantifying</identifier>
<location>
<url>https://aclanthology.org/2026.findings-emnlp.1596/</url>
</location>
<part>
<date>2026-10</date>
<extent unit="page">
<start>33994</start>
<end>34024</end>
</extent>
</part>
</mods>
</modsCollection>
%0 Conference Proceedings
%T Quantifying and Mitigating Default Behaviors in Text-to-Image Models
%A Rassin, Royi
%A Slobodkin, Aviv
%A Ravfogel, Shauli
%A Elazar, Yanai
%A Goldberg, Yoav
%Y Lee, Hung-Yi
%Y Dev, Sunipa
%Y Li, Jessy
%Y Elliott, Desmond
%S Findings of the Association for Computational Linguistics: EMNLP 2026
%D 2026
%8 October
%I Association for Computational Linguistics
%C Budapest, Hungary
%@ 979-8-89176-462-0
%F rassin-etal-2026-quantifying
%X Text-to-image (T2I) models consistently collapse to stereotypical attribute values when prompts leave concepts underspecified, a phenomenon we term default behaviors. We conduct a large-scale empirical study using , an LLM-driven measurement approach that quantifies diversity at the level of individual concept attributes. We find pervasive default behaviors across all evaluated T2I models, with stronger models tending to exhibit less diversity. We trace these defaults to reporting bias in training captions: when captions omit common attributes, the accompanying images are dominated by a narrow set of values, and model outputs closely mirror this skew. We confirm this link causally with controlled fine-tuning interventions. As a proof of concept, we show that making sampled attribute values explicit broadens outputs, with some loss of faithfulness to secondary prompt details. Together, our results recast low diversity not as an inherent limitation of T2I models, but as a measurable and steerable consequence of what their training captions leave unsaid. Code and data available at: https://royira.github.io/GRADE.
%U https://aclanthology.org/2026.findings-emnlp.1596/
%P 33994-34024
Markdown (Informal)
[Quantifying and Mitigating Default Behaviors in Text-to-Image Models](https://aclanthology.org/2026.findings-emnlp.1596/) (Rassin et al., Findings 2026)
ACL