TOS interpretation. About real people.

I'm planning to publish embedding trained on photos of a real person. Due to lack of a close-up's in training data it's failed to reproduce any meaningful resemblance of that person. Basically I've got another face based on a "mask" learned by embedding.

So questions is:

1. Do I have to mention prototype name in embedding description?

2. Can I post NSFW images in preview? I want to demonstrate what kind of body features this embedding will produce.

If people can guess who it is, assuming they know the subject, then it must follow all rules for real people even if the quality is low or not plausibly photographic. If the resemblence was truly not captured then how could anyone even know what subject you used?

Hi!

Coolguy is correct - if we can see any resemblance between the name of the subject and a real-life person, then it would definitely need to conform to the nsfw rules. However, based on the way you're describing - if the face just looks like rando-face #1 then you'd be fine to post nsfw - we'd never know what training data you used.

Thanks!