A test lora following Knxo's guide (https://civitai.com/articles/138/making-a-lora-is-like-baking-a-cake) using an entirely synthetic dataset.

Trigger: saimaki (it's supposed to be s4imaki but works better without the '4' for some reason)

Extra tags: green_micro_bikini_green_pelvic_curtain_highleg , white_virgin_killer_sweater_white_thighhighs

All images were generated with the following:

Size: 1024x1024 , 896x1152, 1152x896 (+Hi-Rez x2)

Sampler: Euler A Karras 20 steps + HiDiffusion

CFG: 6

Model: AutismMix_pony

VAE: none

Lora has a strong tendency to make arched doors, windows, gradient green background, sun, earrings (all of which can be put in negative prompt if needed)