NSFW Marking
under review
l
lotte_peisch764
Right now, there are many MANY models on the site that REQUIRE one to put some version of 'nsfw' in the negative prompt to force the model to create SFW material. Nevertheless, almost all of these models are marked as SFW.
Here's my proposal:
In order for a model to be marked SFW, it must create SFW material without requiring additional negative tags to force it to be SFW. If the default, no negatives, no loras, no TIs version of the model can not create SFW images, it should absolutely NOT be marked SFW in the system. This can be automated just by looking at prompts submitted with the model, and it would FORCE model makers to make dual purpose models default to SFW while requiring tags to create NSFW content.
Reason being...
If I pick up a model that's marked SFW and it immediately makes NSFW without having a bunch of prompts to pull it back to SFW, I'm being sold a bunch of lies. If I have to use tokens to force your model to bs SFW, it's not SFW.
Civitai
under review
Civitai
Good idea, but not sure how we could enforce this yet.
l
lotte_peisch764
Civitai: The first step, to me, would be just scanning the negative prompts of the images shown in the showcase. If certain words show up in the negative, the model automatically gets shifted to NSFW. It'd be a double whammy with the AI image classification filter. People will still try to get around it, but it's at least a first step.