Type | |
Stats | 927 19,322 3.5k |
Reviews | (190) |
Published | Apr 21, 2024 |
Base Model | |
Training | Steps: 12,000 |
Usage Tips | Clip Skip: 2 Strength: 1 |
Trigger Words | shyguy shygal |
Hash | AutoV2 C4FBC59767 |
The shy ones from the Mushroom Kingdom!
This is a lora trained on 152 images taken from e621. It was trained for 12,000 steps at 1024x1024 resolution. Originally there was nearly a 1,000 images, so I had to manually sort through and delete the poor pic (I never realized there was so much art online of them...)
PonyDiffusion apparently knows a good amount of the "shyguy" tag, however it seems to not be able to properly generate the characters at all, so here I am to fix that! Technically there is no "shygal" tag, but I tagged all images in this dataset with both "shyguy" and "shygal" in case people don't use the right tag. So, try to use "shyguy" when prompting with this since that tag will worth with the model better!
NOTES:
The mask usually does the classic ":o" face, but with some clever prompting you can get emotions. Just make sure to prompt into "simple face" and maybe against "tongue" if you're having problems. See example pictures for examples.
There seems to be a slight bias towards red clothing and cyan/teal hair. You can easily prompt against this if you want more colorful generations! Using the dynamic prompts (wildcards) extension can make this fun for big batches.
Be careful with certain artist tags or when combining this with other artist loras, since they might mess with the eyes. Most artist tags seem to play nice, though.
As far as I could tell in testing, you cannot generate "normal" (canon, on-model) shyguys.
Credits:
minus8 - Made the original character
If I took that off, would you die?