The lines not worth crossing: likeness, consent and synthetic media

Operators in this niche tend to treat all risk as the same kind of risk: something that might get an account banned. That framing is fine for most decisions and catastrophically wrong for a few.

Policy risk

A platform may restrict or remove your account. Painful, survivable, and largely a business-continuity problem. You mitigate it with diversification, documentation and compliance.

Legal exposure

Consequences that follow you personally regardless of what any platform does. Not mitigated by having a backup account, by operating through a company, or by the platform never noticing. There is no continuity plan for this category.

Almost everything in this business sits in the first category. The rest of this page is about the small set that does not.

This is the central one, and it covers more ground than people assume. Face swaps onto a real person. Training an adapter on photographs of a real person. Generating a synthetic character deliberately built to resemble a specific individual. So-called nudify or undress tooling applied to a real photograph.

Platform position is unambiguous: this category is described as zero tolerance with immediate permanent bans across the platforms in this space. Legislators in multiple jurisdictions have been moving the same way, treating distribution of non-consensual intimate synthetic imagery as a criminal matter and not a terms dispute. We have not summarised specific statutes here, for the reason given in the provenance note at the foot of the page; get advice for the jurisdiction you actually operate in.

The practical rule is simple and has no edge cases worth exploring: generate synthetic identities, train only on your own synthetic output, and never reference a real individual anywhere in the pipeline. Our consistency methods guide covers how to build an identity from scratch, which is the only defensible route.

This is the category with the least tolerance for error anywhere in the stack: platforms, payment processors, hosting providers and law enforcement all treat it as absolute. Policy language across the major platforms covers content whether real, fictional, simulated, cartoon, AI-generated or edited.

The operational implication is that this cannot be handled by intent. A generation pipeline does not know what you meant. It has to be handled by a review gate that rejects anything ambiguous instead of assessing whether it is probably fine.

Weaker than the first two in legal terms but it belongs on the list, because it is the one operators walk into gradually rather than deliberately.

It typically starts with a slightly vague bio, continues through chat scripts that imply a physical person, and ends somewhere that looks like inducing payments through a false representation. The platform framing is misleading fans. The consumer-protection framing, in some jurisdictions, is less forgiving than that.

The mitigation is proper disclosure, and specifically extending it to everyone with account access, not treating it as a profile-settings task.

For balance, because treating everything as existential is its own failure mode and it leads people to ignore the real lines.

The moment a chatter, manager, editor or agency touches your accounts, your compliance position becomes whatever the least-briefed person believes. Written rules are the only version of this that survives contact with a real team.