Update in AI Gen Speed with Latent Consistency Models (LCMs)

LCMs: The next generation of generative models Have arrived! 

Example of LCM gen on ComfyUI 

New, improved image-creating technology is here! And they’re changing the game. Imagine creating 10 images in one second using only a standard computer memory amount—that’s possible with a strong GPU like the 4090, thanks to something called Latent Consistency Models.

Right now we have Latent Diffusion models (LDMs) from the likes of Stable Diffusion (SDXL). Dalle-3 or Midjourney – They are great at making very clear, high-quality images, but they usually take a long time and a lot of computer power because they build these images step by step.

Think of the whole image-making process like a complex math problem inside a model (specifically, it’s called a PF-ODE). LCMs can find the answer super quickly without needing to work through the problem step by step. This means we can make high-quality images much faster! The AI world is moving so fast!

Read full paper here