LCM - Latent Consistency Models

Latent consistency models(LCM) are used for image generation, similar to stable diffusion but much faster, often used for “real time AI image generation”.

Good(ish) article giving a high-level overview of how LCMs work here. Complete with anime girl diagrams, in true stable diffusion vibes.

Essentially, my stupid takeaway:

Latent consistency models still use the same underlying latent space -> denoising process that Stable Diffusion pioneered. Usually these models use a distillation of Stable Diffusion, to gain from all of the knowledge and advancements gained in image generation by SD. The difference with LCMs: instead of doing multiple iterations to remove the noise iteratively from the noised image/starting point, LCM does this in one pass. It’s a one-shot denoising process.