Delta-SVD: Efficient Compression for Personalized Text-to-Image Models

Fuente: arXiv
Gespeichert in:
Bibliographische Detailangaben
Hauptverfasser: Zhang, Tangyuan, Chen, Shangyu, Chen, Qixiang, Cai, Jianfei
Format: Preprint
Veröffentlicht: 2025
Schlagworte:
Online-Zugang:
Tags: Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
_version_ 1866912551249379328
author Zhang, Tangyuan
Chen, Shangyu
Chen, Qixiang
Cai, Jianfei
author_facet Zhang, Tangyuan
Chen, Shangyu
Chen, Qixiang
Cai, Jianfei
contents Personalized text-to-image models such as DreamBooth require fine-tuning large-scale diffusion backbones, resulting in significant storage overhead when maintaining many subject-specific models. We present Delta-SVD, a post-hoc, training-free compression method that targets the parameter weights update induced by DreamBooth fine-tuning. Our key observation is that these delta weights exhibit strong low-rank structure due to the sparse and localized nature of personalization. Delta-SVD first applies Singular Value Decomposition (SVD) to factorize the weight deltas, followed by an energy-based rank truncation strategy to balance compression efficiency and reconstruction fidelity. The resulting compressed models are fully plug-and-play and can be re-constructed on-the-fly during inference. Notably, the proposed approach is simple, efficient, and preserves the original model architecture. Experiments on a multiple subject dataset demonstrate that Delta-SVD achieves substantial compression with negligible loss in generation quality measured by CLIP score, SSIM and FID. Our method enables scalable and efficient deployment of personalized diffusion models, making it a practical solution for real-world applications that require storing and deploying large-scale subject customizations.
format Preprint
id arxiv_https___arxiv_org_abs_2508_16863
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle Delta-SVD: Efficient Compression for Personalized Text-to-Image Models
Zhang, Tangyuan
Chen, Shangyu
Chen, Qixiang
Cai, Jianfei
Computer Vision and Pattern Recognition
Personalized text-to-image models such as DreamBooth require fine-tuning large-scale diffusion backbones, resulting in significant storage overhead when maintaining many subject-specific models. We present Delta-SVD, a post-hoc, training-free compression method that targets the parameter weights update induced by DreamBooth fine-tuning. Our key observation is that these delta weights exhibit strong low-rank structure due to the sparse and localized nature of personalization. Delta-SVD first applies Singular Value Decomposition (SVD) to factorize the weight deltas, followed by an energy-based rank truncation strategy to balance compression efficiency and reconstruction fidelity. The resulting compressed models are fully plug-and-play and can be re-constructed on-the-fly during inference. Notably, the proposed approach is simple, efficient, and preserves the original model architecture. Experiments on a multiple subject dataset demonstrate that Delta-SVD achieves substantial compression with negligible loss in generation quality measured by CLIP score, SSIM and FID. Our method enables scalable and efficient deployment of personalized diffusion models, making it a practical solution for real-world applications that require storing and deploying large-scale subject customizations.
title Delta-SVD: Efficient Compression for Personalized Text-to-Image Models
topic Computer Vision and Pattern Recognition
url https://arxiv.org/abs/2508.16863