Apparently it's incredible difficult for the existing models to generate images with this fantastic, nerdy and practical piece of garment, so I decided to have a go at training a Lora. I seem to get the best results around 0.8-0.9. Updated 26/7-23.

Anorak

Fewer and better files, so now you don't need to stick your hands in the pockets, just because you are wearing a anorak. Seem to give best results around 0.8-0.8.

Anorak_Lora.safetensors

v1.1

v1.0

First (ish) version.. surprisingly good, but apparently when you wear an anorak, you also have your hands in the pckets.

Files with SHA256: b40ec04e95e5970f1b93cd6cad43b6189184f0dfbc77e6180d83d7b30106a2c7