Using Stability AI’s Latest Text-To-Image Model
A short one this time!
Stability AI recently released a research preview of their latest model named Stable Cascade. The release scripts for training were a bit rough and very optimised, but we can’t expect much from what is essentially just a preview.
However, user 2kpr has created a tweaked fork of the SAI’s StableCascade repository that is more readily usable for training and inference. You can see some of the initial conversation here from the Stable Diffusion Training Discord Server.
The instructions in the readme are pretty self-explanatory, although for me personally, Under linux I had to additionally install a local version of the CUDA libraries to get it going:
conda install nvidia/label/cuda-12.1.0::cuda-toolkit
I will probably add more detail to this post at a later date that might be useful for those who are new to any of this.
Links
https://github.com/2kpr/StableCascade
https://discord.gg/emuZnSaEgf