BunnySlop AI Style.
big boobs 1 girls slop style MOSS/IL/Anima LoRA. from Slopia with love.|
MOSS v6.9/B : train with v6.9 dataset. feel still weak tho. maybe I will train again with more step
Krea2(test) : train with v6.9B dataset.
since I have skill issues with natural language prompt, that are best preview I can get, so go find someone else prompt to copy 😂
seem working better with non-Anime fine tune and too strong for Anime one, if you use with Anime fine tune try lower the weight down to 0.6-0.8 or lower.
Anima-2.9B ver. are converted for Anima-2.9B or something train/merge with it like Janima2.9b as name suggest
v6.9B : try to be more closer to older version
v6.9 : more vibrant color. thicker line
since most of my dataset are black haired girl, it may bias toward that.
This is my first attempt at create LoRa. this shit gonna be suck. you has been warned.
I create this intended for personal use because I'm tired of have to stack shit ton of style LoRas to get style I want and I'm suck at using artist tag. skill issues.
Preview image setting
Sampler: Eular a
Schedule type: Simple
Step: 30
CFG Scale: 4
Hires-fix upcaler: 4x_IllustrationJaNai_V1_DAT2_190k, x1.5, 20 step
gen with Forge neo. so no workflow for ya.
Description
hope this time isn't overtrain kekw
FAQ
Comments (4)
ump45！faker！
yeah, the real one must have the Berlin wall
Thanks! Your LoRa style looks amazing. But it feels a bit overfitted, the poses tend to be quite rigid and fixed. I tested both v404 and v3.14, and the issue persists unless I drop the weight below 0.2. I checked the LoRa's metadata and it showed around 3300 steps (tho that might be misread, since 72*20/1 should be around 1440). But if it really was trained for 3300 steps with only 72 images, that definitely feels too high. You might want to try bringing the total steps down to around 1100-1400 and test from there🤓☝️
thank you for advise.
I will try when I have time make next version