Hello there. Summer is coming and soon we'll need something refreshing. So I present to you this model. It is a purely 2d, very retro-inspired tool that goes well with pastel colors, provides sharp and crisp images without upscaling, and it is very compatible with various LoRAs, LoCons, and TIs. The colors are not too vivid, adding to retro feeling. It can be pushed even further by using ClearVAE, but I personally prefer blessed2vae. It is attached for your convenience, but not baked in so you may choose freely; don't forget to rename it after downloading. Or don't use any VAEs whatsoever, it'll give a very old VHS tape look, if that's your thing. Blessed2 gives the effect of a Blueray remaster of an old anime title. The exact goal I was aiming for.

I was testing this model on everything I could think of. Some pics in the gallery require loras/TIs to be reproduced. Numeratic's loras I renamed to keep them all in one place, so if you need them, adjust the prompt accordingly. Eyes are mostly good (apply detailed if that's not the case) and eyebrows mostly follow the hair color without prompting anything. Landscapes can be beautiful and detailed. This model can be used for NSFW pictures as well, but it's not its main focus (nor should it draw them on its own, at least the Cream version).

Main drawbacks of this models are:

hands: they are always a pain but this model is completely rotten at them, so I really advise you hiding hands behind back, in pockets or out of frame;
zoom layer/multiple views: the main blight of this model apart from hands. It really likes doing these, so I'd advise you to add them to negative prompt as well;
backgrounds: they can go missing despite being described, this can actually be used to produce cutouts easier;
bar censorship: very prominent for any hardcore pics, add these to negative if you don't enjoy them;
privates: they can be discolored or weird, that isn't much of a problem for softcore.

Some of these are alleviated in the later versions, so check those too.

I did a further revision of my prompts and came to a conclusion you really don't need much of negative TIs, at least for this model. Pick one and build around it. And no, badhand doesn't help at all here, that was basically the first thing I checked. So you don't need to be verbose in either part of your prompt. The only quality prompt I use is detailed. It is quite helpful as this model tends to produce plain images on its own (which can be both good and bad, and I'm glad to have that option).

Pruned files behave very differently when compared with full ones, up to a point of being perceived as an entirely different model. In some cases they even perform better (especially if we're talking about hands). So I decided to split them up into different subpages. Every gallery would be absolutely the same, drawn with the respective model, all other settings intact. All things considered, I do recommend trying pruned files first.

My recommended settings remain roughly the same:

Steps: 25-35;
Scale: 5-15;
Sampler: DPM++ 2M Karras;
Clip skip: 2 (or 1 if you need colored skin).

Credits

I used an extensive list of models in my merge, painstakingly selecting those that would work towards my goal. Influence of some of these might already be miniscule to a point of extinct but I still want to name them and give my thanks to their authors. So, in no particular order here they are: