Overview
A LOCON for Stable Diffusion 1.5. This model was trained and tested on the Mistoon Anime model using Kohya SS.
Usage
Trigger keywords: ayumu uehara
Recommended keywords: pink hair, hair ornament, yellow eyes, single side bun
Recommended strength: 0.5 ~ 0.6
Recommended SD model: Mistoon Anime model
Description
FAQ
Comments (5)
That is a massive file for 1.5. Are LOCON usually this large, or still unusual?
I've been setting my LOCONs to high dimension. That seemed to give me better results back when I was running tests for deciding parameters, but might just be bias/fluke.
@CulturedDiffusion Is that how the kohya ss resizer works then, larger dimension = bigger file size and such? Cause the one time I resized a lora using kohya I kept the defaults (I think dimension or whatever was 4) and the size size went from over 800MB to under 20MB. There seemed to be no noticeable impact on the quality.
@Kaladae Yes, the resizing a LORA/LOCON generally means reducing the dimension. I haven't actually tried reducing the size of a model I've already trained to test the effect. Will look into that.
@CulturedDiffusion Hey thanks for teaching me something today. From my sample size of 1, it seems very effective. May be worth a look.