I couldn't find any succubus LoRA meant for photorealism, the closest being Rogue Girl SDXL. So I generated images using that and a combination of checkpoints and LoRA I made to make several images to train this LoRA on.
I'm happy with the results on generating different skin colors consistently. Previously when I was trying to generate images of red skin I mostly ended up with sunburns. It does a decent job with separating out different horn colors, but this mostly went untagged. The results can likely be improved with the use of prompt cutoff.
There did appear to be some bias in the initial training images where any sort of 'colored skin' prompts tended to gravitate towards a base subject of dark-colored skin that would recolor to red,green, or purple while the base subject of light-colored skin would recolor to pink, blue, or yellow. I couldn't shake this from my prompting results and had trouble rotating colors in photoshop to get a decent result for training in the end, so I was mostly unsuccessful in overwriting the seemingly already trained-in bias.
... also there's some bias that the "squatting" and "squatting cowgirl" position defaulted to sexual activity because of the base model I was using. Adding 'vaginal penetration' and 'dildo riding' to the negative prompt may help if generating these poses while not wanting that.
blue,green,hot pink,pink,yellow,light,dark,purple are the keywords for skin, hair, and horn color.
demon wings or bat wings along with low wings are also usable.
This time I did try and include a better variety of small and medium-size breasts in the training sample to make this more widely usable. Every training image was nude as I couldn't get the prompts for the first generation to work well with clothing. Plus it's a succubus, so being nude goes with the territory a bit...
I'll use the same settings to train an SD1.5 version next. I've included the training set in case anybody wants to use it and update the tags or create a flux or pony version since I don't have the setup for those model types.
Description
Same training set with SD 1.5 settings. Appears to need stronger weighting on the skin color in order to get a consistent effect. It also seems like it isn't doing much to help with photorealism overall, just that it doesn't push things into anime territory as much as the others do.
I don't recommend it compared to the SDXL version
FAQ
Looks like we don't have an active mirror for this one right now.
This archive is kept alive by community contributions, so there are still some gaps, especially for newly removed content.
Some files do get recovered over time through contributions. If you're looking for this one, feel free to ask in Discord, or help preserve it if you have a copy.