This is my first LoRa-model.
v1.0
I think that it is kinda overtrained and not perfect, but still allows you to perform a sufficient number of manipulations.
Tips: 0.6 - 0.8 scale will be good
training model (also used to generation): abyssorangemix2_Hard
Keywords: open gag mouth
v2.0
Complete model overhaul, from learning model to learning level. Gronon helped me with changing numbers and stuff for this version of the model.
Tips: 1 scale must be good, for better results u can use 768x768 resolution
Training model: NAI
Images generated on model: abyssorangemix2_Hard
Keywords: circle gag mouth
If u know how to improve model, let me know
Description
Arrived
FAQ
Comments (4)
man. I've been waiting for someone to make this for a while.
I'm not too happy with the results though, fine for a first LoRA.
Don't train LoRAs on a merge, finetuning a model that already had it's weights fucked with makes it really hard to use on anything else
@gronon Hi, ty for your feedback!
Can you explain to me in more detail and clearly what the merge is about? If I understand what you are talking about, it may have a positive effect on my future results.
@DWeoss you wrote on the description that you trained the LoRA on AOM2, people usually just used the leaked NAI model for training because all of the mixes have a little bit of it mixed in them, so most LoRAs trained on that model work pretty well no matter what checkpoint you use.
Beyond that, there are also many great concept LoRAs trained on it and you can just copy the settings used on them with the "Additional Networks" extension to the sd webui. Usually those settings work fine if you train on NAI.
@gronon can we contact in private chat, like discord for better respondings? (only text chat, not voice). If yes, pls contact with me (Weoss#4120), i have questions, that, i think, u can explain for me, thanks :)