Hey there, it's HornyBoobstard and Si, petite dragon girl from Arknights: Endfield. This is first published AI-Toolkit Lora and result I think is rather bad at the moment. More thoughts about it in the end.
Si is one of the new playable characters in Arknights: Endfield. Somehow she is thirteen sister of Sui (I have heard theory that she is artificial body combined with energy of thee Sui sisters). This time I think that is all the information because I still haven't started new story mission. Last story missions are really boring in their "we have three really important items (that would be forgotten after this mission) and you need to collect/activate all of them". I hope when I finally play it, everything would not be like this again.
Recommended strength is 0.7 or higher. Local tests shows some really blurry results even after additional training. Less trainings makes images even more blurry while more training makes details better but still keeps blur in some extent. If I'll figure how to work with very detailed characters I will probably retrain it in the future.
For body use:
Si_Endfield, green hair, blue eyes, long hair, double bun, black streaks, black hairband, black ribbons, hair ornament, pointy ears, tag earrings, yellow earrings, dragon girl, black and yellow horns, eastern dragon horns, multicolor tail, eastern dragon tail, black tail, tail ornament, coin-shaped ornament, glowing holed coin, arms camouflage, For her dress use:
si_dress, white dress, blue stripes, green trim, cleavage cutout, microdress, black high collar, sleeveless, black armbands, detached sleeves, long sleeves, yellow rope, bead bracelet, yellow bracelet, white thigh strap, white socks, black shoes, And that's all for now. I think I kind of figured out how AI-Toolkit works and already trained few Lora on it (because Kohya literally locked from usage). New Lora shows that it might work rather good with characters that are rather simple and has low amount of additional details. When it comes to more detailed character (like this one) it works rather awfully. Still I was asked to make this one so I guess I have to post at least some results even if they are not amazing. Why not collab? I am already using collab for Anima, if I move Pony to collab I would sacrifice model that seems to be basically upgrade and more popular. I will try to spend some time on AI-Toolkit but I have doubts that I would get better results. Maybe this method is just worse then Kohya or maybe CivitAI just failed to implement it correctly or maybe there is some rules that I don't know because failed to find. In any case more Anima soon and maybe Pony. You may support me with crypto. Or at least react, download, generate and buzz. See you next time!

