Latent Consistency Models: Synthesizing High-Resolution Images with Few-Step Inference

TL;DR

Latent Consistency Models enable rapid, high-quality image synthesis from pre-trained diffusion models with minimal inference steps, significantly reducing computational costs while maintaining state-of-the-art performance.

Contribution

We introduce Latent Consistency Models (LCMs) that allow fast, few-step inference on pre-trained latent diffusion models, and propose Latent Consistency Fine-tuning for customized datasets.

Findings

LCMs achieve high-fidelity image generation with only 2-4 inference steps.

LCMs trained on LAION-5B-Aesthetics outperform existing methods in few-step generation.

Training LCMs requires only 32 A100 GPU hours for high-resolution images.

Abstract

Latent Diffusion models (LDMs) have achieved remarkable results in synthesizing high-resolution images. However, the iterative sampling process is computationally intensive and leads to slow generation. Inspired by Consistency Models (song et al.), we propose Latent Consistency Models (LCMs), enabling swift inference with minimal steps on any pre-trained LDMs, including Stable Diffusion (rombach et al). Viewing the guided reverse diffusion process as solving an augmented probability flow ODE (PF-ODE), LCMs are designed to directly predict the solution of such ODE in latent space, mitigating the need for numerous iterations and allowing rapid, high-fidelity sampling. Efficiently distilled from pre-trained classifier-free guided diffusion models, a high-quality 768 x 768 2~4-step LCM takes only 32 A100 GPU hours for training. Furthermore, we introduce Latent Consistency Fine-tuning (LCF), a novel method that is tailored for fine-tuning LCMs on customized image datasets. Evaluation on the LAION-5B-Aesthetics dataset demonstrates that LCMs achieve state-of-the-art text-to-image generation performance with few-step inference. Project Page: https://latent-consistency-models.github.io/