SD3.5-Flash: Distribution-Guided Distillation of Generative Flows

Fuente: arXiv
Saved in:
Bibliographic Details
Main Authors: Bandyopadhyay, Hmrishav, Entezari, Rahim, Scott, Jim, Adithyan, Reshinth, Song, Yi-Zhe, Jampani, Varun
Format: Preprint
Published: 2025
Subjects:
Online Access:
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866915513950535680
author Bandyopadhyay, Hmrishav
Entezari, Rahim
Scott, Jim
Adithyan, Reshinth
Song, Yi-Zhe
Jampani, Varun
author_facet Bandyopadhyay, Hmrishav
Entezari, Rahim
Scott, Jim
Adithyan, Reshinth
Song, Yi-Zhe
Jampani, Varun
contents We present SD3.5-Flash, an efficient few-step distillation framework that brings high-quality image generation to accessible consumer devices. Our approach distills computationally prohibitive rectified flow models through a reformulated distribution matching objective tailored specifically for few-step generation. We introduce two key innovations: "timestep sharing" to reduce gradient noise and "split-timestep fine-tuning" to improve prompt alignment. Combined with comprehensive pipeline optimizations like text encoder restructuring and specialized quantization, our system enables both rapid generation and memory-efficient deployment across different hardware configurations. This democratizes access across the full spectrum of devices, from mobile phones to desktop computers. Through extensive evaluation including large-scale user studies, we demonstrate that SD3.5-Flash consistently outperforms existing few-step methods, making advanced generative AI truly accessible for practical deployment.
format Preprint
id arxiv_https___arxiv_org_abs_2509_21318
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle SD3.5-Flash: Distribution-Guided Distillation of Generative Flows
Bandyopadhyay, Hmrishav
Entezari, Rahim
Scott, Jim
Adithyan, Reshinth
Song, Yi-Zhe
Jampani, Varun
Computer Vision and Pattern Recognition
Artificial Intelligence
We present SD3.5-Flash, an efficient few-step distillation framework that brings high-quality image generation to accessible consumer devices. Our approach distills computationally prohibitive rectified flow models through a reformulated distribution matching objective tailored specifically for few-step generation. We introduce two key innovations: "timestep sharing" to reduce gradient noise and "split-timestep fine-tuning" to improve prompt alignment. Combined with comprehensive pipeline optimizations like text encoder restructuring and specialized quantization, our system enables both rapid generation and memory-efficient deployment across different hardware configurations. This democratizes access across the full spectrum of devices, from mobile phones to desktop computers. Through extensive evaluation including large-scale user studies, we demonstrate that SD3.5-Flash consistently outperforms existing few-step methods, making advanced generative AI truly accessible for practical deployment.
title SD3.5-Flash: Distribution-Guided Distillation of Generative Flows
topic Computer Vision and Pattern Recognition
Artificial Intelligence
url https://arxiv.org/abs/2509.21318