Chibi style loRA that I will merge/add to my model in the future. For now I'm playing around with it, and sharing it here. The settings are still kinda experimental as I haven't had that much time to check it all out. It seems loRA strength 0.3 is ideal, but increasing it can enhance the 'chibi' effect to almost cartoony levels.

Re the trigger: There are two triggers, one is for the regular chibi style that I trained the model on and the trigger is: (chibi style) -- with brackets. If you want a semi chibi (more realistic looking) style, use 'chibistyle' --without the quotes.

Recommended settings:

Sampler: Euler, Euler A, DPM++2M

Scheduler: Karras

Steps: 15+

CFG: 3.0 - 8.0 (though havent tested this out properly yet)

NB: Strength: 0.3 (thats the sweet spot)

Trigger: (chibi style) - for normal/trained chibi look

chibistyle - for semi-realistic chibi look