𝐈𝐌𝐏𝐎𝐑𝐓𝐀𝐍𝐓 𝐓𝐎 𝐑𝐄𝐀𝐃:
The original LoRA was trained on TensorArt, but now I'm uploading it here so I hope it works well.
I made a V2 that is way better than the V1. The eyes are detailed, the style is defined without copying the hair and the clothes of the original images... I think it's almost perfect.
If you want to test this style, ignore the V1 and go directly to the V2. Works for males and females, but Danbooru is 98% females so the models will always struggle to generate males. Unfair, but with good prompting it can be solved.
DO NOT COPY THE CHARACTERS IN THE SHOWCASE. They are my Ocs. :P
𝐔𝐏𝐃𝐀𝐓𝐄 𝟐:
Recommended weight: 0.6. Also, try to avoid generic prompts like "ultra detailed" or "high quality".
The E10 version is better for clothing and full body. The original (E14) is better for eyes and faces. 🩷
E10 is the same LoRA, just a "flexible" version based on the Epoch 10. I've noticed this one does better with clothes and hairstyles so... Yeah. It works better for fashion/half-full body scenarios while the E14 is better for face close ups and eye focus.
I would appreciate any advice, feedback, and for you to share your work. Thank you very much.
