SAGA: Learning Signal-Aligned Distributions for Improved Text-to-Image Generation

Fuente: arXiv
Salvato in:
Dettagli Bibliografici
Autori principali: Grimal, Paul, Soumm, Michaël, Borgne, Hervé Le, Ferret, Olivier, Sugimoto, Akihiro
Natura: Preprint
Pubblicazione: 2025
Soggetti:
Accesso online:
Tags: Aggiungi Tag
Nessun Tag, puoi essere il primo ad aggiungerne!!
_version_ 1866917206997073920
author Grimal, Paul
Soumm, Michaël
Borgne, Hervé Le
Ferret, Olivier
Sugimoto, Akihiro
author_facet Grimal, Paul
Soumm, Michaël
Borgne, Hervé Le
Ferret, Olivier
Sugimoto, Akihiro
contents State-of-the-art text-to-image models produce visually impressive results but often struggle with precise alignment to text prompts, leading to missing critical elements or unintended blending of distinct concepts. We propose a novel approach that learns a high-success-rate distribution conditioned on a target prompt, ensuring that generated images faithfully reflect the corresponding prompts. Our method explicitly models the signal component during the denoising process, offering fine-grained control that mitigates over-optimization and out-of-distribution artifacts. Moreover, our framework is training-free and seamlessly integrates with both existing diffusion and flow matching architectures. It also supports additional conditioning modalities -- such as bounding boxes -- for enhanced spatial alignment. Extensive experiments demonstrate that our approach outperforms current state-of-the-art methods. The code is available at https://github.com/grimalPaul/gsn-factory.
format Preprint
id arxiv_https___arxiv_org_abs_2508_13866
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle SAGA: Learning Signal-Aligned Distributions for Improved Text-to-Image Generation
Grimal, Paul
Soumm, Michaël
Borgne, Hervé Le
Ferret, Olivier
Sugimoto, Akihiro
Computer Vision and Pattern Recognition
State-of-the-art text-to-image models produce visually impressive results but often struggle with precise alignment to text prompts, leading to missing critical elements or unintended blending of distinct concepts. We propose a novel approach that learns a high-success-rate distribution conditioned on a target prompt, ensuring that generated images faithfully reflect the corresponding prompts. Our method explicitly models the signal component during the denoising process, offering fine-grained control that mitigates over-optimization and out-of-distribution artifacts. Moreover, our framework is training-free and seamlessly integrates with both existing diffusion and flow matching architectures. It also supports additional conditioning modalities -- such as bounding boxes -- for enhanced spatial alignment. Extensive experiments demonstrate that our approach outperforms current state-of-the-art methods. The code is available at https://github.com/grimalPaul/gsn-factory.
title SAGA: Learning Signal-Aligned Distributions for Improved Text-to-Image Generation
topic Computer Vision and Pattern Recognition
url https://arxiv.org/abs/2508.13866