Latent Consistency Models (LCMs) are a high-resolution image generation model that rapidly generates high-fidelity images through few-step inference. LCMs can be extracted from any pre-trained stable diffusion model, requiring only 32 A100 GPU hours of training to generate high-quality 768x768 resolution images. Additionally, LCMs introduce a new method called Latent Consistency Fine-tuning (LCF) which allows for fine-tuning on custom image datasets, enabling customized image generation.