Content Moderation
How Nunchux checks requests, and how an approved account turns the check off
What Nunchux checks
Nunchux checks the prompt of every request to a Nunchux Optimized model. It sends the prompt to OpenAI’s moderation API while the model runs. If the check flags the prompt, Nunchux withholds the result and returns a 422content_policy_violation. You are not charged for a withheld result.
Moderation is on by default for every account and every request.
Turning moderation off
An account that Nunchux has approved can turn the check off for one request. Sendcontent_moderation: false in the request body:
content_policy_violation. This check reads the prompt text, not uploaded images.
Some models come from third-party providers, which apply their own moderation that Nunchux does not control.
Requests without the field, or with content_moderation: true, are moderated as usual. An approved account can keep moderation on for some requests and turn it off for others.
Who can turn it off
Nunchux approves an account to turn moderation off. Approval covers the whole account. A personal account can apply as well as a team. An account that is not approved gets a 403content_moderation_not_approved when it sends content_moderation: false. Nunchux does not run the request:
true or false. Any other value, including a string or null, returns a 400 invalid_request.
Where the field works
Partner model routes, such as
/v1/google/, /v1/klingai/ and /v1/bytedance/, do not read the field. Nunchux runs no content filter on them. The partner applies its own moderation.
With the OpenAI SDK, send the field in extra_body, as with other Nunchux fields. See OpenAI compatibility.