5/12/26
- Version 3.1 Enhanced Faces
Change log:
- Trained LoRA on Miiisnai’s faces making faces more accurate
- Changes to LoRA to fix Images having back facing viewer
- Re-added supporting trigger words for body type
- This Version has essentially upgraded how face generation works making it more true to Miiisnai’s style, Haruka V2, Hoshino V2 and Nagi are still the best models to use for this LoRA, The next version of this LoRA would aim to be the last update so I can get rid of the (WiP)/ fully release it.
To do List:
- Finalize face consistency with sharp faces
- Release “ideal” prompt for body to work well with Miiisnai’s style (body hair included)
5/11/26
- Release of Version 3
- This version has made some significant improvements on using other models, specifically Haruka V2 and Hoshino V2 are currently on par with its shading and line art but still needs improvements on its face. That being said Haruka and Hoshino now work extremely well in imitating the artstyle the best without solely relying on the Nagi model but Hoshino still needs refinements. Additionally I’ve removed the supporting body tags for now to test the waters (they will be re-added soon)
To do list:
- Improve Face Consistency to imitate Miiisnai more accurately
- Fix back focused default generation
- Adjust consistency for Hoshino model
5/10/26
This is a LoRA meant to immitate the nsfw artist Miiisnai’s art style. I heavily recommend you use the Haruka v2 or Nagi model when using this LoRA because I found it yielded the best results to immitate their art style but the crystalize model works too as viable alternative. Ideally the Lora should be in the strength range of 0.7 and up but since this is a WIP toy around and let me know what works.
