Skyrim style, trained on mix of vanilla and modded images

Version 2 has a much larger dataset and some better captioning than version 1. For reference, version 1 was trained with around only 35 images. Version 2 was trained with almost 200 images.

With this comes better knowledge of Skyrim races. It should respond better to race specific names. Here are the example names you can use in prompts now:

Nord Woman\Nord Man

Redguard Woman\Redguard Man

Argonian Woman\Argonian Man

Khajiit woman\Khajiit man

Dunmer woman\Dunmer man

Bosmer Woman\Bosmer man

Altmer Woman\Altmer man

orsimer man\orsimer woman

You can use this in addition to the trigger word for specific races. May add more races later. Also since the dataset has many more elves than before, you may see humans with elf ears, putting "pointy ears" or "Elf ears" in the negative seems to help in that case.

A simple example prompt looks like this: klein-skyrim-style, Argonian man standing, in a forest

Trained with ai toolkit, default settings, 9000 steps.