Every moderation job evaluates content against four checks (one per moderation category plus the prompt prohibition check). You can disable any check that doesn’t fit your use case — disabled checks are skipped entirely, saving cost and latency.
By default, all four checks are enabled.
Open the Moderation page in the customer dashboard. Each check appears as a tile with a toggle. Changes are saved immediately and apply to all subsequent jobs.
Available checks
When underage is enabled, jobs are rejected if the check fires. Account termination is not
automatic — it is a manual admin action triggered separately. Disabling underage removes that
rejection signal, so make sure you have an alternative CSAM-detection control in place before
turning it off.
How disabled checks behave
- Text disabled — keyword filtering and the LLM prohibition check are skipped for prompts, so a prompt is never rejected on text grounds.
- A single VLM category disabled (e.g.
nsfw) — the per-category VLM prompt is skipped at enqueue time and the category does not appear in the result. For video, the corresponding per-frame VLM prompt is skipped too.
- At least one built-in VLM category must remain enabled. The API rejects an update to
/me/moderation-config that would disable nsfw, ai_adult_general, and underage all at once with a 400 invalid_request.
API access
You can also read and update your toggles from the API.
Read current settings
The enabled_categories object contains only the checks you’ve explicitly disabled — anything missing is enabled. The catalogue provides human-readable labels and descriptions for each check.
Update settings
The body must be a JSON object. Allowed keys are the four check names listed above; values must be booleans. Sending an empty object {} resets every check back to its default (enabled). 