This repo swaps out the original vae in stable diffusion v1.5 for the new MSE vae found at https://huggingface.co/stabilityai/sd-vae-ft-mse-original.

The results seem very minimal, perhaps i'm improperly swapping the decoder weights...

This is mostly for those curious, so break stuff, try it out, see what works, etc...

NOTE: usage of this model implies accpetance of stable diffusion's CreativeML Open RAIL-M license