Text-to-Image
Diffusion Single File
English
anima
comfyui

Things I noticed during testing

#19
by DarionK - opened

I noticed Masterpiece and Great Quality tags drastically change the artist styles, if you avoid them the artist styles are more loyal. So something to keep in mind when prompting for this model, also some small number character data has been affected, dunno if repeating data from characters or artist below 50 entries and above 20 might be a good idea to improve the possibilities for the model to be able to reproduce them.
Something else, avoiding using _ might be useful to get artists or characters to appear better.

So, just some things I noticed and wanted to share, funny enough I have trouble recreating characters that supposedly have over 500 images, a example is Zoey from kpop demon hunter, for some reason I can't make it appear properly, which I find funny. Maybe because most of her art is in groups, but dunno tbh.

Hopefully this info is already known but wanted to share in case that wasn't the case.

If someone else wants to add to this thread, you are welcome to do so.

Sign up or log in to comment