This is my Futanari LoRA, focused on larger and more pronounced ??????? imagery. It has a harder time doing flaccid and non-erect imagery.
As usual, the strength is good from 0.6 to 0.8, but I use 0.75.
Though this LoRA heavily assists with making futanari imagery, if your model can't handle ????? generation well, it can only help so much. Granted that most models can only do ????? generation coherently around 50%-70% of the time, expect better but similar results.
Beware that I accidentally left in "large ?????????" in the training dataset tags, and applying this to your prompt makes it likely that you'll generate a woman with only testicles. This can possibly happen without this tag, but I've only gotten one generation like so.
The keywords are 'zfuta' and 'futanari'. These two in conjunction can produce good results, and often you do not need many more tags in order to get a good result.
All pictures were generated with my custom blends, available on my profile. I used the Orangemix VAE with Hires Fix on. I trained this LoRA with NAI and used 300 images for the dataset.
I can't promise that I can release a better version, this took multiple training attempts before I was satisfied enough. (I used 2 batches, so 6000 training steps is 12,000)
