HUGGY niji style/pose

Trigger Words: some loras include multiple trigger codes (aaXXXxx, aaXXXxx, etc), representing different styles / characters, use ONE trigger code each time for best result. Check my sample prompts for characters reference.

This is a lora complex which contains various unique styles and poses. Currently I have a list of ideas to add into this lora complex. It takes enormous time for me too prepare source images. Each Lora will need 2-6 iterations to fix glitches.

Today (24 July) I just realised two things: source images with less captions get better result; training as concept is better than training as style, which means I have to start all over again. I usually like to test my own lora with all available checkpoints online before publishing, but I cannot do any of these since my buzz is running oooout, donate buzz if you like what you see.

TESTING STAGE

Issues to be fixed in the next version:

1) the quantity of source images for current version is two times of previous one, faces are less blurry, but fingers are worse, the only solution is to manually fix the hands from source image by inpaint (huge task, not anytime soon)

2) character generated from the later version looks ugly than the first version, which makes me confused.

3) I was told it was ok not to unify the dimension of source images, which did save hips of time. But I do notice issues on extra limbs which didn't happened in the previous version.

4) I removed the source images which have finger pointing gesture in this version, but this gesture still randomly appears which is weird.

aaEURm1n1 (2.5d style)

aaFLAm1n1 (flat style)
aaPO8m1n1 (painting style)

aaSKTm1n1 (sketchy style)