v8.0 Release!
Improved Realism
Better human characters
Better in-between transformation aspects
More varied backgrounds
No VAE Required
DPM++ 3M SDE Exponential Recommended
50 Steps
CFG 2
Moved to NoobAI in fresh new merge!
v7.0 Release!
Improved Realism
Better human characters
Better in-between transformation aspects
More varied backgrounds
Moved to Clip Skip 1
Same generation parameters as previous versions
No VAE required
DPM++ 3M SDE Recommended!
Model Specifications
Sampling Method: DPM++ 2 M SDE or DPM++ 3M SDE
Schedule Type: Exponential or Karras
Steps: 30-75
Resolution: 1024x1024 or 1024x1152
CFG Scale: 3-6 (4 is a sweet spot)
Clip Skip: 2
General Information
This model is intended to produce photorealistic images of furry characters. It is very good at creating NSFW images. It can be used to produce non-furry characters, but that is the main purpose of this model. I heavily encourage experimentation with this as I have been testing this merge for about a week, and I've had pretty good results. Please see example images for prompt templates. Some example images have had minor inpainting done, and all have been upscaled using x4 NMKD Superscale.
Extra Information
So, it has been a while since I've been actively posting on CAI, and I somewhat fell off the face of the earth. My absence for this long was caused by some mental health concerns that led to me taking down my previous models. These issues are being taken care of, and this upload should stay up. Some of the old Scrimblo Sauce models remain archived on CAI, and they should stay here. I will hopefully be able to upload some of the missing original models that I still have copies of in due time. For now, enjoy this new SDXL version of Scrimblo Sauce.
Online Generation
v2.0 PixAI: https://pixai.art/model/1807403823970331577
v7.0 Tensor: https://tensor.art/models/881161578831938708
Description
Improved Realism
Better human characters
Better in-between transformation aspects
Fresh merge using NAI models
CFG Scale Sweet Spot reduced to 2-5. (2 is the real sweet spot)
DPM++ 3M SDE Exponential
50 Steps
FAQ
Comments (7)
I got only noise with nobAI 8.0 model with any settings
Steps: 50, Sampler: DPM++ 3M SDE, Schedule type: Exponential, CFG scale: 2, Seed: 2997208860, Size: 960x1280, Model hash: 08d97d3169, Model: scrimbloSauceXL_v80NAI, FP8 weight: Enable, Cache FP16 weight for LoRA: True, Version: v1.10.1
What was your prompt?
@ScrimbloHere extremely realistic, hyperrealism, natural lighting, portrait, (anthro goat female, detailed white fur, fangs:1.3), biting lip, long white hair, chubby, fat, purple robe, large breasts,
I just copy one of your picture settings and prompt
@loki9, ensure your UI supports v-prediction, I assume the model uses that as you can't experiment with training loras until you enable the v-prediction flag.
esse DPM++ 3M SDE procurei mais que tudo porem nao achei
Really love v8.0. It's become my go to realistic Illu/Noob model over WAI Realism. Thanks!
Very cool model! Needs a bit of prompt manipulation to get lighting right (it tends towards high contrast/shadow night scenes, I hope this is improved later) but it is probably the best photorealistic furry SDXL model I've used so far.
It also has some very spicy and specific anatomical knowledge, which is a welcome change from 90% of the other realistic furry models that tend to just add big circumcised human wangs to every male character. Must be an interesting dataset lol.
Great work OP.