Content rules
What this API won't make, and what happens when a request asks for it.
What we don't make
- Sexual content. Nudity, sexual acts, fetish content, or removing someone's clothes.
- Anything sexualising a minor. A person who looks under 18 in any sexual, romantic or revealing context. Never, in any form.
- Real people without their consent. Use photos of yourself, or of people who agreed to it.
- Public figures made to say or do something. Named celebrities and politicians.
- Gore and graphic injury.
- Hatred against a protected group.
- Fake IDs, documents or money.
Ordinary portraits, fashion, jokes and flirting without nudity are fine. A photo of a person is fine by itself.
What happens
Every request is checked three times:
- At the door. A short list of words that are never legitimate here answers at once with
input_refused. You're charged nothing. - By our check. A model looks at the prompt and the photos before any work starts. A refusal ends the generation with
content_refused, and the credits come back. - By the model. Models have rules of their own, and some refuse more than we do, including photos they take for real people. That also ends in
content_refusedorinput_refused, with the credits back.
Refusals count. Twenty refused requests within 24 hours pause the account, and every call then answers account_suspended.
If you're an agent
Tell the person what was refused. Don't reword a request to get past the rules: a refused idea stays refused, and each try counts toward the pause.
What you're responsible for
You need the right to use every file you send. When your own customers use your product, pass a pseudonym in end_user (never an email), so that one misbehaving customer can be told apart from the rest.