Just a lora trained on a bunch of random chibi stickers, emotes, and poses. 0.8 seems to be the sweetspot on most models. While it was trained on NAI it works on most models even realistic (but the output won't be realistic!)
Keywords: Chibi emote
As for training tags and use, general human emotions and poses, hair colours, 1boy / 1girl, should know features like glasses, and can give both full body chibi or more emote style. I have no idea if it can work with NSFW models (no NSFW in training data) or how it plays with character Lora, as I never tested that.