Heads-up to genners, this is an amazing merge made by a literal who:
https://civitai.com/models/835655?modelVersionId=1023901It's primarily trained on Danbooru so strict tagging is necessary, but with a little patience in reading through the Danbooru tags wiki, you can bet extremely specific scenarios.
I never use LORAs with this, just throw an artist tag (with at least a hundred images on Danbooru) and the name of a girl with their usual descriptions, and you're off to the races. It's incredible, and the backgrounds tend to be way better than base NoobAI.
I tend to use:
25 steps
4.5 CFG scale
Euler A sampler, Simple scheduler.
It's fine with any image ratio I've thrown at it with SwarmUI.
I tend to gen a bunch, then take the favorites and re-run the same seed through the refiner (base model, 0.2 refinder control percentage).
