A request was rejected by a filter
Why a harmless description sometimes gets rejected and how to rephrase without risking your account.
Every model has its own safety filter, and their strictness differs. Sometimes a filter triggers on a perfectly harmless description — for example, because of a single word or an ordinary portrait.
What happens
- credits for the rejected generation go back to your balance;
- if a model’s filter rejected only the text of the prompt, no violation is counted automatically;
- if our filter rejected the request, or a model’s filter rejected it because of an attached file, it counts as a violation and the service shows how many attempts are left.
How to rephrase
- remove words that could be read ambiguously (weapons, blood, undressed and so on) if they aren’t essential to the scene;
- describe the same idea more neutrally;
- try another model — filters differ.
Don’t try dozens of wordings: many rejections within an hour pause generations automatically. Details in Violations and blocking.