LCM
An image generation model based on consistency distillation techniques. It achieves fast, high-quality image generation by substantially reducing the number of diffusion inference steps required.
Timeline (last 90 days)
Representative few-step distilled models include SDXL-Turbo, LCM (Latent Consistency Model), and Flash Diffusion.
LCM (Latent Consistency Model) can significantly reduce the number of diffusion steps.
Acceleration nodes in the ComfyUI ecosystem are typically implemented based on consistency distillation or flow matching acceleration techniques (such as LCM, TCD, or Hyper-series LoRA) to compress the number of inference steps.
Common inference acceleration techniques include consistency distillation, flow matching, and diffusion model step compression methods such as LCM and SDXL-Turbo.
Latent Consistency Models (LCM) apply consistency distillation to the latent space, enabling high-quality image generation in 1–4 steps, and have been widely adopted by the community.
The Turbo and LCM variants of Stable Diffusion compressed the number of sampling steps from dozens down to 1–4 steps through consistency distillation
All Facts (6)
Representative few-step distilled models include SDXL-Turbo, LCM (Latent Consistency Model), and Flash Diffusion.
50%UnverifiedLCM (Latent Consistency Model) can significantly reduce the number of diffusion steps.
50%UnverifiedAcceleration nodes in the ComfyUI ecosystem are typically implemented based on consistency distillation or flow matching acceleration techniques (such as LCM, TCD, or Hyper-series LoRA) to compress the number of inference steps.
50%UnverifiedCommon inference acceleration techniques include consistency distillation, flow matching, and diffusion model step compression methods such as LCM and SDXL-Turbo.
50%UnverifiedLatent Consistency Models (LCM) apply consistency distillation to the latent space, enabling high-quality image generation in 1–4 steps, and have been widely adopted by the community.
50%UnverifiedThe Turbo and LCM variants of Stable Diffusion compressed the number of sampling steps from dozens down to 1–4 steps through consistency distillation
50%