Training Tagging Specifications
We all know we can use Danbooru and Natural Language in Anima, but is there a specific format the Anima team used on their database? I think knowing this will help anyone who is trying to finetune the model or train Loras, since proper Database tagging could help to not disturb the original model that much if so desired.
Has that info been released or are examples available?
Either way, I just wanted to ask out of curiosity.
Based on the Lora model I’ve trained (using only Danbooru tags, my English isn’t very good), you simply need to write the caption in the official tag order ([quality/meta/year/safety tags] [1girl/1boy/1other etc] [character] [series] [artist] [general tags]); there are no other issues.
When writing captions, I only add [1girl/1boy/1other etc], [character] [series] [artist]; as for [quality/meta/year/safety tags], I only add ‘highres’, and the order of the general tags doesn’t matter at all.