This lora I did, a long time ago, but abandoned, now there is no time so its time has come. I have managed to do quite extensive testing, but I don't have time to fully describe the results.
It can be used both as a style and as a concept. The outline of the style is there, but I would not say that it is full-fledged
Versions
NAIv1.1 is quite a bit different, maybe it's the best version of all, but since I don't have time to fully test it, I'll leave it on the right. Yes this is a re-training at the moment of the new nai 1.1 version, will it give a positive result? I don't know, I haven't had time to test it, but it will surely increase compatibility in the future.
You could try chibi for it maybe it will have an interesting effect.
pony
It's better than what I scrapped sometime before, the style is at least noticeable. The version is optimized, but I don't have time to describe it. It probably comes first, but that doesn't mean it's better than NAI, it's just pony-based and yes pony is more comfortable with anthropomorphic animals, inside the dataset they are.
Description
And so, this is a general post for all my loras, it's universal because I really have very little time. If you see this, it is likely that the NAI version has had little testing and was actually just ported so that you could apply it to this model. NTR MIX v777 for lor has just been released. So, some of my thoughts here are speculative
for styles
ntrmix v4 doesn't really require quality tags if it's a lore style, the style will be most original without them, see my generation there most likely correct.
The cfg scale with styles is about 3-5 (4) for v4, maybe for the new version you can specify more or the quality tags have less influence on the style.
I probably posted a few epochs due to lack of testing time; they should be optimal or at least work somehow, take your pick, but most likely the leftmost version is better. The styles themselves are trained with good methodology and there should be no problems, in extreme case increase or decrease the lore weight most often optimally between 0.8 and 1.2