stable-k-4d
jupyter/colab implementation of Stable Diffusion and Katherine Crowson's k-diffusion wrapper in order to use the k_lms sampler.
by karan4d aka DamascusGit. The notebook utilizes CPU draw manual seeding with k_lms in an attempt to best replicate the results from Simulacra/Dreambot using the stable-diffusion weights.
Using The Notebook
Provided you have the weights, store them in
/stable-diffusion/models/ldm/stable-diffusion-v1/
Change your desired model's name to model.ckpt.
Run the cells in order except for "DEBUG SHIT IGNORE THIS".
Comments
implemented with the help of Katherine Crowson, palp, wbrown, and litexev.
This would not exist without the good folks over at EleutherAI + StabilityAI.
Inspired by the very special work of the SimulacraBot Team.