Ethical AI Video Generation

Almost every tool has an ethics page. Far fewer have a rule you could be removed for breaking. Here is the difference, and where we sit.

Read the Terms

Consent is the whole question

Nearly every real harm from AI video traces back to one thing: a face or a voice used without the permission of the person it belongs to. Everything else — watermarks, model licensing, training data — matters, but it is downstream of that.

So the useful test of a tool is not the tone of its ethics page. It is whether the thing you would be worried about is written down as forbidden, and whether anything happens when someone tries it.

What Percify prohibits

These are conditions of use, quoted from our terms rather than summarised:

  • Create avatars of individuals without their explicit consent
  • Create deepfakes or misleading content intended to deceive
  • Impersonate any person or entity

Enforcement is not only editorial. Generation requests pass through automated content moderation, and it does refuse work.

What we do not do

A page like this is worth less if it only lists strengths, so here are the limits.

  • We do not embed content credentials. Percify does not write C2PA provenance metadata into exports, so a file cannot be traced back to us by an automated checker. If your workflow requires provenance signing, it has to happen in your own pipeline.
  • We cannot verify consent. No tool can confirm that the person in an uploaded photo agreed to be there. The rule is real and enforceable after the fact; the judgement at upload time is yours.
  • Moderation is imperfect in both directions. Automated filters refuse legitimate portraits and cannot catch every misuse. Anyone claiming otherwise is describing a system that does not exist.

Five questions to ask any vendor

Does the tool require consent for a likeness, in the terms?
The operative question is not whether a vendor talks about ethics but whether creating an avatar of an identifiable person without their permission is actually prohibited. Percify prohibits it explicitly, alongside impersonation and deepfakes intended to deceive.
Is there enforcement, or only a policy page?
A rule with nothing behind it is a disclaimer. Ask what happens at generation time. Percify runs automated content moderation on generation requests, and it does reject work — including, sometimes, legitimate portraits.
Who owns the output, and does that change the ethics?
It does not. You own what you generate with Percify, but ownership of an output has no bearing on whether you had the right to the face, voice or footage that went into it. Those are separate questions and vendors sometimes blur them.
What happens to the photos and audio you upload?
Look for a stated purpose limitation. Percify takes a licence to your uploads solely to provide the service, and you retain ownership. The absence of any statement is the answer to watch for.
Is synthetic media disclosed to the people who see it?
This is the responsibility that sits with you rather than the tool. Most platforms now require you to label AI-generated content, and in several jurisdictions so does the law. A vendor cannot do this on your behalf at publish time.

Related: watermarks and commercial use, the AI video glossary and the privacy policy.