/vt/ - Virtual Youtubers

>>65843163
>>65843741
As for running at resolutions that it hasn't been trained on, that is the purpose of using hires fix (img2img on a lower resolution image) to avoid multiplication of images. if you try to gen straight up to a high resolution you will either get chimaeras, people standing on each other's shoulders or boxed generations like picrel.
If it's close, it will work fine, like 768x512 but 1024x512 will very often result in multiple images/subject/etc.
Landscape vs vertical will show that landscape creates prettier backgrounds but worse humans at least in anime models.