Version 2.0 Details
This version has been trained on all new images with more controllable exaggerated features. Ask for a big mouth, big eyes, round or long face/head, big ears, etc. For some reason modifying the ears frequently manifests animal ears - there were none in the training images so I'm not sure where those are coming from :P
A strength of around 0.8 will help tone down horrible teeth and hands.
All the sample images above were created with SDXL Base 1.0. If you find it works well with another model, please let me know in the discussion below. Have fun!
Version 1.0 Details:
This LoRA creates goofball portraits with exaggerated proportions. For lack of a better word, the trigger is "caricature" and there are three framings/compositions intentionally trained into it.
Try starting your prompt with of of the following (replace the italicized words with your own):
caricature closeup portrait of subject in wearing clothing style in location
caricature half-body portrait of subject in wearing clothing style in location
caricature full body portrait of subject in wearing clothing style in location
If you don't specify a location, the default background is a generic studio backdrop or empty room
The LoRA will try to fill the frame so if you want a tall person use a tall portrait dimension, wide landscape for a wide face, etc.
Add emotion keywords to change the default unhappy expression
A CFG of 3 or 3.5 will help avoid contrast issues. The training images were a little "hot"
Finally I would like to thank Wizard Whitebeard for sharing his (at the time private) Gildenface LoRA with me! It was the inspiration for this LoRA, and the source for my initial training images of horrible and marvelous distorted faces.