Few-Step Distillation for Text-to-Image Generation: A Practical Guide

TL;DR

This paper systematically studies and adapts diffusion distillation techniques for open-ended text-to-image generation, providing practical guidelines and open-source tools to enable fast, high-quality, resource-efficient T2I models.

Contribution

It is the first comprehensive analysis of diffusion distillation methods applied to open-ended T2I generation, with practical recommendations and pretrained models.

Findings

Identified key obstacles in applying diffusion distillation to T2I.

Provided practical guidelines for input scaling, architecture, and hyperparameters.

Established a foundation for deploying efficient T2I diffusion models.

Abstract

Diffusion distillation has dramatically accelerated class-conditional image synthesis, but its applicability to open-ended text-to-image (T2I) generation is still unclear. We present the first systematic study that adapts and compares state-of-the-art distillation techniques on a strong T2I teacher model, FLUX.1-lite. By casting existing methods into a unified framework, we identify the key obstacles that arise when moving from discrete class labels to free-form language prompts. Beyond a thorough methodological analysis, we offer practical guidelines on input scaling, network architecture, and hyperparameters, accompanied by an open-source implementation and pretrained student models. Our findings establish a solid foundation for deploying fast, high-fidelity, and resource-efficient diffusion generators in real-world T2I applications. Code is available on github.com/alibaba-damo-academy/T2I-Distill.