This model is a fine-tune of Stable Cascade stage C on 5000 images. The training dataset focuses for realism and consists of a lot of NSFW pictures, so the resulting model is capable of generating some NSFW content. The uploaded checkpoint has been converted to work with the standard ComfyUI Stable Cascade workflow.
Description
Trained additional 20,000 steps.
FAQ
Comments (6)
Please message me on Discord. MACK#1451 mack_is_kool
I want to send you details on how everything is broken from your Comfy workflow and even others when trying to use your model. I am very upset by all of this right now.
The workflow in the images isn't mine, it's the official stable cascade workflow for ComfyUI.
@altron Then can you supply the one you used?
I've used the official workflow, it's the same one as the one here: https://comfyanonymous.github.io/ComfyUI_examples/stable_cascade/.
V1 is an improvement in terms of realistic images / portraits, over the vanilla version.
V2, with the extra training steps, gives a slight improvement over V1.
I finally got this working with diffusers. I had to do some scary stuff to make it work though, like writing a script to rename the weights to what diffusers expected (it was very picky with SC, unlike with SDXL).
However, I only dropped in the prior weights. Were the other weights (vae, text_encoder, etc) trained as well, or only the prior?