Type | |
Stats | 144 658 3.1k |
Reviews | (34) |
Published | Jan 9, 2025 |
Base Model | |
Training | Steps: 2,132 Epochs: 27 |
Usage Tips | Clip Skip: 1 Strength: 0.9 |
Trigger Words | m0use_fstyle |
Hash | AutoV2 7A94B67F13 |
I STRONGLY ADVISE AGAINST USING THIS. IT DOES NOT WORK WELL AND IS NOT GOING TO BE UPDATED FOR THE FORSEEABLE FUTURE!
My first published attempt at a style LoRA for flux. Trained on a set of my recently posted images plus a few unposted ones to add a bit of diversity to the training set.
This is intended to add a cartoony style to images but given the dataset, will likely be somewhat bias toward making furries, especially mice and other rodents.
There is a trigger word ('m0use_fstyle') but having generated images both with and without it, I'm unsure if it does anything.
Known Issues:
Facial expressions are commonly a bit stiff. This doesn't affect all seeds but with how expensive flux is, I'd rather seed searching not be necessary.
Some prompts result in a noticeably different style with much brighter colours. Not sure exactly what triggers it but I'm quite certain that it is due to a particular image in the dataset which will be removed for the next version.
I've seen some talk of faces getting generally screwed up though I haven't encountered this myself so I can't really speak as to the cause.
Sometimes gives mouse features to other animals. This is also seed dependant. Seems we still need a greater variety of animals in the dataset.