From the source
LoRA trains in minutes, produces 5MB models, and allows combining multiple concepts.
It is available now on Replicate.
Introducing LoRA: A faster way to fine-tune Stable Diffusion
Replicate introduces LoRA, a faster method to fine-tune Stable Diffusion.
Capability change
From the source
A few short months later, Simo Ryu has created a new image generation model that applies a technique called LoRA to Stable Diffusion. Similar to DreamBooth, LoRA lets you train Stable Diffusion using just a few images, and it generates new output images with those objects or styles. Unlike DreamBooth, LoRA is fast: While DreamBooth takes around twenty minutes to run and produces models that are several gigabytes, LoRA trains in as little as eight minutes and produces models that are around 5MB.
replicate.com
