v1.1 is trained on Citron Anime Treasure model. Small fixes to dataset.
v1.0 is trained on NTR MIX model.
The model is made from a mix of different artist's styles merged into one. All of the training images are from AI generated images. Finely curated and altered to produce clean outputs.
It's been specifically trained on NTR MIX (Illustrious base) because I feel it's given me the most stable clean outputs out of all the available so far.
It does not use trigger prompt. The intensity is linked to the Lora's weight.
Recommend using DPM++ 2M or Euler A with SGM Uniform.
- 28 sampling steps
- w/e width and height you want
- CFG 5 - 5.7
>Hires. Fix
- Upscaler: None
- Steps: 12
- Denoise: 0.45 - 0.49
- Upscale by 1.5
- Hires CFG: 5.7 (same as base CFG)
Description
This version was made on the Citron Anime Treasure model. I decided to try a new model since the NTR mix model tends to make multiple views/angles. :(
There's been slight fixes in the dataset as it tended to default to a few things, like issues when you add "nose".
Training was slightly more intense so set at 0.5 to get the same-y results as previous version. You can higher to get closer to the dataset's style.