My first SDXL LoHa! I’ve been messing around with SDXL for a while now, and this is the first model I feel is worth sharing. I am working on others and I will improve this one, but I’m curious to see what people are able to do. SDXL is very flexible but very temperamental. Check the description for best models.
Description
This makes a variety of imps. It supports blue, red, pink, green, red, white, midna, and “dark look” imps. It’ll do pointy ears, wide ears, and furry ears. It also has a bit of support for horns, wings, tails, and eye color. It’ll do sexual and non-sexual generation.
Usage
To use it, pick an imp color and ear style for your prompt, ex: green imp, wide ears. Add in “short stack” for a more imp-like look. See gallery for example prompts. If you’re having trouble getting the color of imp, add in skin color to the prompt. A weight of 1-1.3 for the model generally works well. For img2img/upscaling I used 1.5x img2img and 2x SD upscale.
Notes
Due to the variety of the training data you may end up with mix-n-match features if you’re not specific. Lastly, if you’re using midna imp, you may need to use helmet in the positive or negative prompt, depending on what you’re looking for.
Models
As of writing this, I was only able to get this working well with a few models, and NSFW is very iffy but will probably start to work once better models come out. Here’s the ones I found work decently:
Realistic – SpectrumBlendX
Semi-Realistic – Dynavision XL
Anime – Explicit Freedom NSFW
I’ve been having fun results with a 70-30 mix of spectrum and dyna, and also a 25-25-50 of dyna-spectrum-explicit. Let me know if you find any good models and/or any way to improve NSFW results 😜