A port of my 1.5 embed and Lora, using the same dataset after upscaling it.

PDXL Version: Important note

Mistakes were made in tagging, so to have her look proper, you must run it at higher strengths (1-1.3) and have this in the prompt:

sh0ggoth, purple skin

For SDXL version, you can skip purple skin most of the time unless it gets washed out by the rest of the prompt.

Use of Adetailer is also highly recommended.

The dataset also had no backgrounds, so getting one sometimes does require forcing it with enough words; see the showcase images for examples.