Content filter feedback #110
DDG Admins,
How about a feature to indicate the content filter's objection when it rejects a prompt? This would help immensely in debugging them, because sometimes the filter's choices are downright ludicrous.
Examples from personal experience: "treasure chest" in a pirate image failed repeatedly; "treasure box" was acceptable. C'mon folks... really? Similarly, "white décor" was refused, but the filter had no problem "alabaster décor".
More transparency on this point would be greatly appreciated. Thank you.
Resolved: A blocked prompt now says why, and a hiccup in our safety check no longer counts against you.
I agree.
I agree
I agree
It would be nice but not even ChatGPT knows precisely why some image prompts to it are refused - if you ask it, and it can guess, and that's about it. Most of the refusals come from the models, not from DDG.
Dilly dilly!
Done! When a prompt is blocked, the message now says why: the word that isn't allowed, a young person mentioned together with a body word (we name the body word), or that it reads as sexual content or violent death.
And when our safety check itself has a hiccup, you are asked to try again, and it no longer counts toward a temporary block. We also fixed "treasure chest" and "chest of drawers" being read as a body word. More in the release notes: https://deepdreamgenerator.com/updates