Rethinking Global Text Conditioning in Diffusion Transformers

Fuente: arXiv
Saved in:
Bibliographic Details
Main Authors: Starodubcev, Nikita, Pakhomov, Daniil, Wu, Zongze, Drobyshevskiy, Ilya, Liu, Yuchen, Wang, Zhonghao, Zhou, Yuqian, Lin, Zhe, Baranchuk, Dmitry
Format: Preprint
Published: 2026
Subjects:
Online Access:
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866918330415185920
author Starodubcev, Nikita
Pakhomov, Daniil
Wu, Zongze
Drobyshevskiy, Ilya
Liu, Yuchen
Wang, Zhonghao
Zhou, Yuqian
Lin, Zhe
Baranchuk, Dmitry
author_facet Starodubcev, Nikita
Pakhomov, Daniil
Wu, Zongze
Drobyshevskiy, Ilya
Liu, Yuchen
Wang, Zhonghao
Zhou, Yuqian
Lin, Zhe
Baranchuk, Dmitry
contents Diffusion transformers typically incorporate textual information via attention layers and a modulation mechanism using a pooled text embedding. Nevertheless, recent approaches discard modulation-based text conditioning and rely exclusively on attention. In this paper, we address whether modulation-based text conditioning is necessary and whether it can provide any performance advantage. Our analysis shows that, in its conventional usage, the pooled embedding contributes little to overall performance, suggesting that attention alone is generally sufficient for faithfully propagating prompt information. However, we reveal that the pooled embedding can provide significant gains when used from a different perspective-serving as guidance and enabling controllable shifts toward more desirable properties. This approach is training-free, simple to implement, incurs negligible runtime overhead, and can be applied to various diffusion models, bringing improvements across diverse tasks, including text-to-image/video generation and image editing.
format Preprint
id arxiv_https___arxiv_org_abs_2602_09268
institution arXiv
publishDate 2026
record_format arxiv
spellingShingle Rethinking Global Text Conditioning in Diffusion Transformers
Starodubcev, Nikita
Pakhomov, Daniil
Wu, Zongze
Drobyshevskiy, Ilya
Liu, Yuchen
Wang, Zhonghao
Zhou, Yuqian
Lin, Zhe
Baranchuk, Dmitry
Computer Vision and Pattern Recognition
Diffusion transformers typically incorporate textual information via attention layers and a modulation mechanism using a pooled text embedding. Nevertheless, recent approaches discard modulation-based text conditioning and rely exclusively on attention. In this paper, we address whether modulation-based text conditioning is necessary and whether it can provide any performance advantage. Our analysis shows that, in its conventional usage, the pooled embedding contributes little to overall performance, suggesting that attention alone is generally sufficient for faithfully propagating prompt information. However, we reveal that the pooled embedding can provide significant gains when used from a different perspective-serving as guidance and enabling controllable shifts toward more desirable properties. This approach is training-free, simple to implement, incurs negligible runtime overhead, and can be applied to various diffusion models, bringing improvements across diverse tasks, including text-to-image/video generation and image editing.
title Rethinking Global Text Conditioning in Diffusion Transformers
topic Computer Vision and Pattern Recognition
url https://arxiv.org/abs/2602.09268