language: en
license: unknown
tags:
- stable-diffusion
- text-to-image
- not-for-all-audiences
.safetensor model for automatic1111 webui (and similar inference software).
Strange_Dedication_v5 is probably my final improvement of this model series. It kinda evolved into having it's own style, but as it is still based on sd1.5, I guess it will be overtaken soon. I trained some loras to complement what I was missing in v4 and merged the loras with the model to create this result. Due to the loras and it's heritage it's obviously better at drawing women. But it's also quite good at drawing landscapes. It will react heavily on "portrait", this is to make after detailer and face inpaint work better. Also, it's biased into drawing shiny skin and fox ears, sometimes even without asking for it (it's a feature).
I use it mainly with danbooru tags, but you can prompt using natural language as well. Some concepts may require usage of the specific tag tho. Of course it has the sd1.5 common issues with multiple limbs and strange hands, but failure rate is not too bad. I recommend the usage of hires fix, I used it for almost all example images. Also, even if I didn't use face inpaint for the examples, the model will inpaint faces very well.
Without further ado, here some examples (prompt in metadata, more examples in files):
No loras, embeds etc were used for the examples, also no inpainting.