@inproceedings{fernandes-etal-2026-inference,
title = "Inference-Time Scaling in Diffusion Models: A Beam Search Approach to Visual Sequences Generation",
author = "Fernandes, Guilherme and
Ramos, Vasco and
Cohen, Regev and
Szpektor, Idan and
Magalhaes, Joao",
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.2107/",
pages = "44886--44909",
ISBN = "979-8-89176-462-0",
abstract = "While diffusion models produce high-quality images, they often lack consistency in sequences, especially for non-linear visual sequences. To solve this, we introduce an inference-time scaling beam search strategy for latent space exploration, enabling the conditional generation of coherent image sequences through beam search diffusion. In contrast to earlier methods that rely on fixed latent priors, our method dynamically samples past latents to search for an optimal sequence of latent representations, ensuring coherent visual transitions. As the latent denoising space is explored, the beam search graph is pruned with a cross-attention mechanism that efficiently scores search paths, prioritizing alignment with both textual prompts and visual context. Human and automatic evaluations confirm that BeamDiffusion outperforms other baseline methods, producing full sequences with superior coherence, visual continuity, and textual alignment."
}<?xml version="1.0" encoding="UTF-8"?>
<modsCollection xmlns="http://www.loc.gov/mods/v3">
<mods ID="fernandes-etal-2026-inference">
<titleInfo>
<title>Inference-Time Scaling in Diffusion Models: A Beam Search Approach to Visual Sequences Generation</title>
</titleInfo>
<name type="personal">
<namePart type="given">Guilherme</namePart>
<namePart type="family">Fernandes</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Vasco</namePart>
<namePart type="family">Ramos</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Regev</namePart>
<namePart type="family">Cohen</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Idan</namePart>
<namePart type="family">Szpektor</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Joao</namePart>
<namePart type="family">Magalhaes</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>While diffusion models produce high-quality images, they often lack consistency in sequences, especially for non-linear visual sequences. To solve this, we introduce an inference-time scaling beam search strategy for latent space exploration, enabling the conditional generation of coherent image sequences through beam search diffusion. In contrast to earlier methods that rely on fixed latent priors, our method dynamically samples past latents to search for an optimal sequence of latent representations, ensuring coherent visual transitions. As the latent denoising space is explored, the beam search graph is pruned with a cross-attention mechanism that efficiently scores search paths, prioritizing alignment with both textual prompts and visual context. Human and automatic evaluations confirm that BeamDiffusion outperforms other baseline methods, producing full sequences with superior coherence, visual continuity, and textual alignment.</abstract>
<identifier type="citekey">fernandes-etal-2026-inference</identifier>
<location>
<url>https://aclanthology.org/2026.findings-emnlp.2107/</url>
</location>
<part>
<date>2026-10</date>
<extent unit="page">
<start>44886</start>
<end>44909</end>
</extent>
</part>
</mods>
</modsCollection>
%0 Conference Proceedings
%T Inference-Time Scaling in Diffusion Models: A Beam Search Approach to Visual Sequences Generation
%A Fernandes, Guilherme
%A Ramos, Vasco
%A Cohen, Regev
%A Szpektor, Idan
%A Magalhaes, Joao
%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 fernandes-etal-2026-inference
%X While diffusion models produce high-quality images, they often lack consistency in sequences, especially for non-linear visual sequences. To solve this, we introduce an inference-time scaling beam search strategy for latent space exploration, enabling the conditional generation of coherent image sequences through beam search diffusion. In contrast to earlier methods that rely on fixed latent priors, our method dynamically samples past latents to search for an optimal sequence of latent representations, ensuring coherent visual transitions. As the latent denoising space is explored, the beam search graph is pruned with a cross-attention mechanism that efficiently scores search paths, prioritizing alignment with both textual prompts and visual context. Human and automatic evaluations confirm that BeamDiffusion outperforms other baseline methods, producing full sequences with superior coherence, visual continuity, and textual alignment.
%U https://aclanthology.org/2026.findings-emnlp.2107/
%P 44886-44909
Markdown (Informal)
[Inference-Time Scaling in Diffusion Models: A Beam Search Approach to Visual Sequences Generation](https://aclanthology.org/2026.findings-emnlp.2107/) (Fernandes et al., Findings 2026)
ACL