>>51611922IIRC the AI is trained on 512x512 images (or images downsized to 512x512) so that's the default and optimal output size
Bigger resolutions would probably take much more time to build the AI
You can also ask for extra pixels in width or height (works well for scenery, but there's an comes at a risk of distorted characters)
>>51611853>en Lillie vs jp Riribasically the same thing
I won't budge unless you continuously write リーリエ in english contexts