This is my very first model training. I already effed up because I messed up a tag in the training data, lol. Gonna retrain soon. I think I will also play with the LoRA network settings to in order to shrink the file down a bit, as long as it doesn´t impact the results negatively.

Comments, critique and suggestions are welcomed and encouraged.

I trained this model on 66 3D renders created by myself.

SDXL
Trained at CLIP STOP:2

I use Foocus Moonride to generate images. Turns out a CLIP Skip of -4 works best there.

Trigger: ohwx3d

other words used:

greeble
iso
cubic
cubes
tech
soft
holo
neon
orb
fluid
matte
city
hex
metal
mograph
glossy
plastic
sss
dots

and, because of a typo: "i" and "so" (gonna fix soon)