Synthetic people, handled seriously.
This product generates human beings who do not exist and puts words in their mouths. That deserves a clear position rather than a paragraph buried in terms of service.
The cast is generated, not scraped.
Actors in the shared library are generated identities. They are not built from a specific real person, and they are not trained on a named individual's likeness. An actor built from a real person is an enterprise conversation precisely because consent and licensing have to be handled properly rather than assumed.
AI generated, and we say so.
TikTok and Meta both require AI content disclosure. Meeting that requirement needs more than good intentions: it needs to be knowable which model produced which part of a given asset. The platform records that per asset, so a disclosure is a lookup rather than a guess.
Layers, including the ones that are inconvenient.
Provider filters
Every generation passes through the safety classifiers of the underlying providers. Rejections surface with the real reason rather than a generic failure.
Wardrobe checks
Reference images are reviewed for visible clothing before they reach a video model, because an under dressed reference gets rejected and because it is the right default.
Separated buckets
Public showcase assets live in a physically separate bucket from customer generations. There is no path from private account content into the public one.
What we are still working on
Transactional email is not yet provisioned, which means account notifications do not send today. Automated content review catches the categories described above and is not a substitute for your own judgement about whether a specific ad is appropriate for a specific audience. If you find output that should not have been produced, tell us and we will look at it.
Questions about any of this?
Ask directly. A real answer is more useful than a policy page.