[SD 2.1][TextualInversion]Smoose - a style embedding v2

0xSoul

Oose - Noun. Any soft downy substance, typically from soft fibers or hairs

smoose - the small oose embedding.

Smoose is a style embedding trained on my personal digital art as well as a set of gens I made when I first got SD running. It's mostly a 'mood', but it attempts to do the following:

• Roughly center the character, if there is one.

• Add a rimlight

• Add subsurface scattering

• Add more dynamic light

• soften background, or isolate against a plain background.

  

It pushes things toward full body characters, so sometimes it's best to turn down the weighting on it, or add "human" or "character" or something to the negative if you're trying to do a landscape.

This was the very first embedding I made, and I didn't know diddly about captioning, or dataset curation, so it's really RAW but I use it... a lot. It's become the salt on my table.

Version 2 was trained by a friend on sd 2.0 when it dropped. It isn't as strong in some ways, but is better behaved overall (doesn't push you around as much as v1).

It mixes VERY WELL with my CharTurner embedding.

While it will work in the base models, I use LOTS of models, and the examples are in many different models. I really do use this embedding on almost every gen.

The longer your prompt is, the more detail in it, the more Smoose will be a flavoring, and not a overwhelming force.