Acceptable Use & Content Policy

Last updated 18 August 2026.

This page is not a list of aspirations. It is the written form of what our code actually refuses, so that the refusal you meet in the product and the rule you read here are the same rule. Where the two ever disagree, that is a bug in this page and we want to hear about it.

coMeMenity is a meme service. Memes are rude, political, unflattering and frequently in poor taste, and almost none of that is a problem here. What follows is the short list of things that are.

We check both sides, and so do the model providers

There are four checks between your words and a meme, not one:

1. Our own check on your input, before anything is sent anywhere and before you are charged a generation. This is where the four refusals below happen.
2. The model provider's own filter. Every provider we route to runs its own, and they do not all draw the line in the same place.
3. Our own check on the output, after the picture comes back and before it is delivered to you. A picture our own check holds back is stored undeliverable and never served by any route.
4. A human, when somebody reports something. See the Copyright and Repeat Infringer Policy.

Checks 1 and 3 are ours. We run them because the provider filters are not enough — one of the three image providers we route to declined nothing at all across 44 live test requests, including every photorealistic request naming a living public figure. On that route our own check is not a second opinion. It is the only one.

The four refusals you can meet

There are exactly four. If your request is refused by us before dispatch, it is one of these, and the product names which one.

Three notes on that table:

• The minor class is the one where we deliberately offer no alternative and no rewording. Every other refusal points at something you can do instead; that one does not, on purpose.
• We refuse hate for attacks on people and groups. We do not refuse violence, self-harm or illicit subject matter at our own layer — a meme about violence is most of political satire and most of internet humour, and refusing it would refuse the product. Those are left to the provider's filter, which may well refuse them.
• A refusal by our own check costs you nothing. It is deliberately placed before the point where your allowance is consumed. A refusal by the provider may cost you one generation, because the provider charged us for the attempt.

Real people

This is the part of the policy that gets the most careful treatment, because it is where the serious harms are and because what we originally believed about it turned out to be wrong.

If your situation names nobody, none of this applies. Describing a situation rather than a person is the ordinary case, the commonest case, and the thing this product is for.

When a situation does name a specific real person, the rule is:

• A private individual: never. Not photorealistically, not as a cartoon, not "just as a joke". A drawing of a named private person is still a depiction of that person. Consent would be the only thing that could change this, and this release has no mechanism for obtaining or verifying consent — so "never without consent" resolves, here and now, to never.
• A public figure, photorealistically: never. A convincing fabricated photograph of a real person is the deceptive case, and it is refused regardless of how funny the situation is. It is also contractually prohibited by one of our providers independently of any filter.
• A public figure, drawn — or where you have not said how: we stop and explain before spending anything. You are shown what the policy is and what continuing does. That is a real stop, not a cosmetic interstitial.
• Sexualised or deceptive depictions of any named person: never, in any register, whether the person is famous or not.

What "continue" actually does, and what it does not promise

If you continue past that explanation, we change which model your request goes to — to the one whose observed behaviour actually permits a drawn send-up of a public figure.

We do not change your words. There is no code path in this product that takes a prompt, rewrites it to be more acceptable to a filter, and sends the rewritten version. Rewording a request to get past a safety filter is filter evasion, and we treat it as out of bounds for us as much as for you.

And continuing does not promise you a picture. Two things you should expect:

• The model may still decline. We measured this rather than assuming it, and here is the uncomfortable finding: which real person you name matters more than how you ask for them to be drawn. One provider refused one named public figure every single time both photorealistically and as a clearly drawn newspaper cartoon, while refusing three other named public figures never, in either register. So we will not tell you that going non-photoreal is a route that works. For some people it is not a route at all, on some models.
• If the model declines, it may still have charged us, and in that case it costs you a generation. Our own refusals are free. A provider's are not always.

Do not try to get round any of this

Attempting to evade a safety check is itself a violation of this policy, whether or not it succeeds. That includes rephrasing around a refusal, splitting a refused request across two fields, describing a person without naming them in order to get a likeness anyway, and using a different style chip in the hope of finding a more permissive model.

We hold ourselves to the same rule, which is why we say it so directly. When a provider refuses one of our requests, the only thing we do is tell you and ask you to describe something else. We do not retry it, we do not soften it, and we do not send it to a different provider. Routing a blocked prompt to a more permissive vendor is filter evasion with extra steps: it voids what protection there is, it creates real legal exposure for exactly the harms this page is about, and it is the pattern that gets API access revoked. Our build fails if that code is written.

What we do not allow, in one list

Beyond the four refusals above, do not use this service to:

• create or distribute sexual imagery of anyone, or any depiction of a minor in a sexual context;
• create non-consensual intimate imagery of any real person, including by "undressing" or sexualising a real person's likeness — see the removal path;
• impersonate a real person or falsely suggest that a real person said, did or endorsed something;
• harass, threaten, or incite violence against a person or a group;
• infringe somebody's copyright or trade mark — see the Copyright and Repeat Infringer Policy;
• attempt to extract, reverse-engineer or override our system instructions through the text you submit;
• automate access in order to exhaust the allowance, the spending ceiling, or a provider's rate limit.

What happens if you do

• A refused request is refused, recorded, and that is usually the whole story. One refusal is ordinarily a misunderstanding, and we treat it as one.
• Repeated refusals are recorded against the requester, including the text of the refused request. That record is an abuse record and it is described in the Privacy Policy.
• A strike is recorded against an account for a substantiated violation, and strikes accumulate. The copyright strike ladder is set out in the Copyright and Repeat Infringer Policy; the same counter carries strikes for the violations on this page.
• Enough strikes and the account is terminated, and its work goes with it.

Reporting something

Anybody can report any meme, with no account, from the page the meme is on. If what you are reporting is an intimate image of yourself or of a real person who did not consent, use the non-consensual intimate imagery path — it has a published deadline and a named person behind it.

You can always simply write to abuse@comemenity.com.