You'll need a lora for this, porn content rejection is heavy. Or you'll need a abliterated model, not sure if vision also works.
You might want to add something like yolo finetune to detect scenes + face recognition too.
You'll need a lora for this, porn content rejection is heavy. Or you'll need a abliterated model, not sure if vision also works.
You might want to add something like yolo finetune to detect scenes + face recognition too.
For GP's purpose, can face recognition techniques be repurposed for, um, other body parts recognition? Sometimes the actresses are facing away from camera. There are exposed lips, if that helps.
Yes, for actresses _and_ actors I'm sure you'd get the same level of performance as you would for any facial recognition use case. You can't do facial recognition on someone's back, but I'm sure there are other techniques/models that can be applied, many people have unique marks/features etc.
Vision still works perfectly fine in abliterated models.
Just because they don't refuse it doesn't mean they are useful.
I found a few pornographic pictures on the web to hand to Abliterated Gemma4 12B(literally just to test this) and it needs pushing just to accept that people can be naked.
It didn't refuse but it also didn't provide useful descriptions such as "this is a pornographic picture of a woman".
> G4: There is a person lying down in a scientific context, if I had to guess they are a biologist in a classroom
> me: Is she wearing any clothes?
> G4: No.
Also, it is obsessed with penises —seeing them in compositions where there is only a female. I suppose it's been trained to ban dick pics or something.
Prompting may help some but 12B seems to be a bit worse than E4B with the vision/audio model at voice and text reading so maybe that one would do better.
Never tried any of this for porn, just speaking out how I would go about it tbh!