AI Safety Policy

YourBrand turns text, photographs and footage into video. Most of what people make with it is ordinary creative work, and this policy is not about that.

It exists because a tool that can put a face into a scene can also be used to hurt somebody, and the small number of ways that happens are serious enough to be worth naming, building against, and writing down.

You are responsible for what you type, what you upload, and what you do with the result. Using this service, you are confirming that your work:

• Complies with the law where you are and where your audience is.
• Respects other people's rights — their copyright, their face, their voice, their name.
• Is not built to deceive somebody into believing a real person said or did something they did not.

Where you upload a photograph or a clip of a person, you are confirming that it is you, or that you have their permission.

The full list, with the reasoning for each, is the Acceptable Use Policy. In summary, these categories are prohibited outright:

• Children's safety
• Sexual content
• Violence and gore
• Hate and harassment
• Self-harm
• Other people's rights
• Deception and fraud
• Illegal goods and activity

Two of those carry criminal liability that no platform protection covers. Child sexual abuse material is reported to NCMEC as a legal duty rather than a policy choice, and non-consensual intimate imagery is criminal in the US under the TAKE IT DOWN Act and in the UK under the Online Safety Act.

This is the part of an AI video tool that goes wrong most often, so it gets its own section rather than a line in a list.

You can put a real person into a generated scene here — a character built from your own photographs, a face composited into footage on the Edit page, a voice reading a script. That is a legitimate and common thing to want. It is also how a deepfake is made, and the difference is entirely consent and context, neither of which we can see from the request.

So the rules are the ones we can actually hold you to:

• Use a real person's face or voice only with their permission, or if it is you.
• Never present synthetic footage of a real person as a genuine recording of something that happened.
• Never use somebody's likeness to sell or endorse a product without their agreement — an advert is where a likeness claim carries real damages.
• Public figures are not an exception. Satire and commentary are legitimate; fabricated statements presented as real are not.

Where a prompt or an upload looks like it touches somebody else’s rights, the app stops and asks you to confirm you have them before it will render. That confirmation is recorded. It covers someone else's character, a real person, you uploaded a photo of someone.

• Provider classifiers are switched on. Every model that exposes a content filter is sent it explicitly on every request, rather than relying on whatever the endpoint happens to default to.
• Prompts are screened before they are sent, for the prohibited categories above, and a match blocks the render rather than warning about it.
• The rights gate interrupts a generation that names a franchise or a real person, or that uploads a photograph of someone, and asks for confirmation before spending anything.
• Every report has a route to a human, and the categories with a legal clock on them are sorted to the front of that queue.
• Removal is reversible. Content taken down is hidden everywhere rather than deleted, because counter-notices exist and destroying the evidence of a dispute is the worst possible answer to one.
• Repeat infringers lose their accounts.

Said plainly, because a safety page that only lists strengths is not a safety page:

• Nothing here is reviewed by a person before it renders. Review happens after a report.
• Prompt screening is defeated by describing a thing instead of naming it. It is a speed bump on the casual case, not a wall.
• We cannot tell from a request whether you have permission from the person in a photograph you uploaded. We ask, we record the answer, and we act on it if it turns out to be untrue.
• We do not watermark output, so a clip made here does not carry a marker identifying it as synthetic.

What we do scales with what was done:

• Content removed, and hidden from every public surface.
• Repeated or deliberate breaches: the account is suspended.
• The prohibited categories: the account is closed permanently, on the first instance.
• Child sexual abuse material: reported to NCMEC, and every associated account closed. Deleting it without reporting would itself be an offence.

Every public piece of work has a report button, and it reaches a real queue. You do not need an account to have something looked at — if you cannot reach the button, email safety@yourdomain.com.

You can report it as:

• Copyright — This uses my copyrighted work.
• My face or voice — This uses my likeness without permission.
• Intimate imagery of me — Sexual content of me that I did not consent to. Actioned within 48 hours.
• Content involving a minor — Sexual content involving a child. Actioned immediately.
• Harassment — This targets or threatens someone.
• Impersonation or fraud — This pretends to be someone, or is a scam.

Copyright notices have their own process and their own legal requirements — see Copyright and DMCA.

The models behind this product change every few months and the law around synthetic media is changing faster than that. This policy is revised when either does, and the categories above are rendered from the same definitions the product enforces, so the published rules and the running code cannot quietly drift apart.

If you think something here is wrong, or that we have missed a category that matters, tell us at safety@yourdomain.com.