Just a question about the text-to-image side of things.

80% of the time when I put in text prompting for a mature adult (SFW/NSFW) image it returns patently underage/child results.

Is there any language I can use to eliminate underage results?

Negative prompts seem to be included with positive, so if I put 'child, cub, underage, immature' in the negative prompt it seems to up the ratio to 100% underage content.

So, clearly, my prompt usage is incorrect or the AI engine is doing something opposed to the given prompts - and it did the same with the 'older' engine (whatever that was)

you are viewing a single comment's thread
view the rest of the comments
[–] 2 points 1 year ago

Sadly, this was one of the "downgrades" of the new model.

The old model was pretty darned accurate when simply using numeric ages (at least for humans/goblins), the new one is horrible.

My previous age prompt: [if (mlf.checked && cgr.checked) {"{40-58} year old"} else if (mlf.checked) {"{59-69} year old"} else if (cgr.checked) {"{40-58} year old"} else "{20-35} year old"]

Now, it's a total crapshoot, and it definitely defaults to younger. Adding "mature" and/or "older" seems to make little to no difference.

  • source