Bubble ??????

V3: wove in some more detail and a bit more ?????? size, so small/flat ????? size tags no longer necessarily come out as chibi and... questionable.

V2: I was able to pull out more detail in my testing , particularly in hair, and with the VAE I use (color101) some more vibrant color. I still recommend using "??????????????" or "?????????????" in the positive prompt to avoid 'chibi' nudity.

My aim here was to land somewhere between cute and ????, chibi and ??????, in a modern/polished version of the ?????? style of the 90s used in series like Lunatic Night, Rei Rei, and F3, that were drawn in an older, more rounded, what I think of as a "bubbly" style.

I think it came out pretty well with a solid ?????? capability, though it occasionally does still lean farther toward chibi than I would prefer, that's what future versions are for.

Preview Images were made with some variation of these settings:
Sampler: I mostly use Euler, DPMs work fine too
Steps: 20-40
hires steps: ~20 (I use a mix of hires fix and Ultimate SD upscale)
denoise: .3-.5
Clip Skip: 2 (1 also works fine, but I prefer 2)

usual positive prompt: masterpiece, best quality, etc.
negative prompt: worst quality, normal quality, low quality

I would also recommend a ?????? tag, medium/large/huge/etc. as the style leans small anyway. chibi in the negative may help some with that as well.

By request, I've uploaded to tensor.art as well here for users who prefer their on-site generation.