aaPO8il, aaEURm1n1, aaFLAm1n1, aaPO8m1n1, aaSKTm1n1, score_9,score_8_up,score_7_up, (masterpiece, best quality:1.3), realistic
Trigger Words: some loras include multiple trigger codes (aaXXXxx, aaXXXxx, etc), representing different styles / characters.The one triggering word each time for best result;Use CFG between 4.5 (complex scene) to 7.5 (simple);Start training steps from a lower value like 19 with Euler a sampler, if you are satisfied with the result, use hi-res fix / face fix (and set Euler to 29) to increase over all quality;You do need to lower lora weight for higher epoch version.The latest lora contains over 10 male characters, there's not enough resources to train them as separate characters, but by putting them into a single lora, it becomes difficult to distinguish one from the others even with seperate character code, perhaps I should train it again as character option next time??? I'm confused about regardlessly how many characters I threw into the source file, the characters generated looks so consistent. Base on my testing results, trigger word "aaPO8il" will magically prettify characters. The 10 unique character trigger codes do not work, that's why I didn't provide them. I'v not given the females specific tag yet, since it did not work on males. Everything is a work in progress.This is a lora complex which contains 9 (100 is my goal) unique styles and poses. Currently I have a list of ideas to add into this lora complex. It takes enormous time for me to prepare source images. Each Lora will need 2-15 iterations to fix glitches.Today (24 July) I just realised two things: source images with less captions get better result; train as concept is better than train 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.Issues to be fixed in the next version:1) character generated from the later version looks ugly than the first version, which makes me confused.2) training base on Pony get versatile results than training base on SDXL, but when prompts contain 1 human and 1 animal, Pony has a great tendency of blending two species, put source_pony, source_furry in the negative prompts won't help. Does anyone knows how to solve this?3) add captions to distinguish different styles (sharpie illustration, manga illustration, simple anime, cartoon anime, minimalist illustration, realistic anime, realistic photograph)4) add captions "bara"
