Use Romanized words as trigger phrases. You must add the trigger for it to look more accurate.
Trained on the Animagine3.1 model, it is not suitable for the pony model.
You can add the anime coloring tag, but it can sometimes appear blurry. High-resolution fix is needed.
If the clothes are sometimes not drawn accurately, it means your tags are not detailed enough.
Regarding color issues, since most of the dataset comes from an old anime, the colors can sometimes appear overly yellow. In this case, you need to reduce the strength or switch the sampler.