Disclaimer
This model applies Terminal SNR to its training noise schedule, as described here.
This will not work out-of-the-box (as of June 2023) with Automatic1111's Stable Diffusion WebUI.
It works great using a Diffusers pipeline, such as InvokeAI, or a simple Diffusers python script using DDPMScheduler. One such script is available, here.
I hope that Automatic1111's support for terminal SNR pipelines is improved someday, because SDXL will require it.
Description
This version of Pseudo's Journey was trained on a smaller subset of the synthetic data in efforts to reign in the text encoder during training.
Small subset means about 3300 images.
This was trained without any prior preservation on an A100 80G to 4,000 steps, at a cost of $5.
Used no special tricks. The text encoder was completely trained, no frozen layers.