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

2025/12/15 by Pu, Yifan, Han, Yizeng, Tang, Zhiwei +4
#Computer Vision and Pattern Recognition (cs.CV) #FOS: Computer and information sciences

paper · doi:10.48550/arxiv.2512.13006

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.

Citations

• Large Scale Diffusion Distillation via Score-Regularized Continuous-Time Consistency
• Qwen-Image Technical Report
• FLUX.1 Kontext: Flow Matching for In-Context Image Generation and Editing in Latent Space
• Mean Flows for One-step Generative Modeling
• SANA-Sprint: One-Step Diffusion with Continuous-Time Consistency Distillation
• Inductive Moment Matching
• Fast High-Resolution Image Synthesis with Latent Adversarial Diffusion Distillation
• Latent Consistency Models: Synthesizing High-Resolution Images with Few-Step Inference
• Consistency Models
• Diffusion-GAN: Training GANs with Diffusion
• Elucidating the Design Space of Diffusion-Based Generative Models
• Photorealistic Text-to-Image Diffusion Models with Deep Language Understanding
• High-Resolution Image Synthesis with Latent Diffusion Models
• Knowledge Distillation in Iterative Generative Models for Improved Sampling Speed
• Denoising Diffusion Implicit Models
• Denoising Diffusion Probabilistic Models
• Deep Unsupervised Learning using Nonequilibrium Thermodynamics

Cited by

• TwinFlow: Realizing One-step Generation on Large Models with Self-adversarial Flows

Related

• Few-Step Diffusion via Score identity Distillation
• SwiftBrush: One-Step Text-to-Image Diffusion Model with Variational Score Distillation
• CoDi: Conditional Diffusion Distillation for Higher-Fidelity and Faster Image Generation
• TeEFusion: Blending Text Embeddings to Distill Classifier-Free Guidance
• DiffBlender: Composable and Versatile Multimodal Text-to-Image Diffusion Models