LoKr (works as a lora) trained on McBess style images. Somewhat creepy, early 1900s rubber hose animation. Trained in Onetrainer on Krea2 Base, dim 32, LoKr alpha 32, Prodigy with growth rate capped at 1.02 118 images mostly 1024^2 (aspects 3:2 wide to 1:2 tall) but with ~25% also at 768^2 and 512^2. 20 epochs, so 2360 steps. No trigger word, because Onetrainer doesn't use them, but every caption started with either "McBess style", "McBess complex style", or "McBess simple style" but I found that these don't do much/anything when using the lora. better to nudge the model a bit with things like "black and white illustration", "faceless", "no eyes", " vertical eyes" as you desire to get the effect you want. The lokr is quite flexible.

Files with SHA256: 6e45b8feef471974693eb42814aebb2810698f2ce94464d0353645d670a2c26e