"The gateway to your tooni-muses"
Introducing: TooniVerse, the first model mix that came out of my hands.
This model aims to blend realism with a cartoon-appearance. It should be capable of bringing your craziest prompts to life!
During my testing, my favourite settings to use while generating were:
Generation Size: 512x768 or 512x512
Samples: DPM++ 2M Karras or Restart
Alternatively, DPM++ 3M SDE Exponential also gave good results during testing, Euler A had a tendency to start drawing additional subjects. ymmv.
Steps: 30
Clipskip 1 or 2 (1 tends to be more detailed, but could give your subjects an older appearance, 2 might sacrifice details bit smoothes out the subject)
CFG: 7 (6 - 8 should work fine, lower or higher is untested by me)
VAE: Either VAE-FT-MSE-840000-EMA-PRUNED or kl-f8-anime2
Hires. Fix: On
Upscaler: 4x_NMKD-Superscale-SP-178000_G
Hires steps: 20
Denoise: 0.4
Upscale by 2
Adetailer: On
1st model: face_yolov8n_v2
(Optional): mediapipe_face_mesh_eyes_only (I get picky about eyes for some reason)
FaceRestore seems optional, the model outputs great faces with and without.
I'd like to thank a few people that have tested MANY pre-releases of this mix (In no particular order):
Another special thanks goes to the people in the UnstableDiffusion discord who reacted to the images that I shared during testing- you gave me a sense of awe during this process which fuelled my desire to continue working until the merge was ready.