I censored a nsfw model ( https://civarchive.com/models/13565/criarcys-fantasy-to-experience ) by making everyone wear heavy coats.
This technique does not require a dataset.
Trained in 500 steps (about 1 hr on 2 a6000s) using only text with:
https://github.com/ntc-ai/conceptmod
The training prompt is:
"badword=fully clothed wearing a parka" for multiple badwords.
None of the example prompts mention coats.
Description
FAQ
Comments (9)
rude
Hey I'm just demoing my conceptmod repo. I can remove your content as base.
@ntc they were just doing model merges as well, so i don't get what the deal is. Looks like someone is coveting what they did recently. You even cited them in your text as the originator of the model. Just gotta laugh sometimes I guess. GJ getting what you wanted from the model. Maybe we could get a lora rip though. That would be fun. Got have them undress from something.
I have another one trained towards vibrant colors I was going to release based on your model.
Let me know if you disapprove and I will withhold the model and retrain with a different base.
@ntc nah im joking lol, the picture just caught my eye from familiarity
@artificial_genius There are options to train only the self attention layers, which can be extracted to a lora. I don't know how to do it yet but it's a good idea.
@Criarcy ;) New model incoming then.
I actually like the idea of safe-itizing models.