What you may not make
A recorded owner decision, awaiting legal wording
Any content sexualising or exploiting children, in any form or style.
Intimate imagery of a real person without their consent, or sexual content depicting an identifiable real person without their permission.
Impersonating a real person — their face or their voice — so as to make a viewer believe they said or did something they did not.
Content intended to deceive in an election or a public process, or impersonating an official body.
Content inciting violence or hatred against people for who they are, or threatening or harassing a specific person.
Practical instructions for harming people — weapons, dangerous materials, or breaking into systems you do not own.
Fraud or forgery: fabricated documents, identities or evidence meant to pass as genuine.
Work protected by somebody else's rights, where you have no right to use it.
A base list, taken from what model providers require us to pass on to you. Matching it clause by clause against each provider's own policy is work waiting on counsel. What counts as a legitimate exception is the next clause, which the owner approved on 2026-09-15.
Journalism, research and satire
A recorded owner decision, awaiting legal wording
Sometimes the forbidden thing above is the subject of the work rather than its output: a report about a deception that cannot be understood without showing it, a study of how a model fails, a satire of a public figure that no viewer would take for a record. That work is allowed here, and this policy does not stand in its way.
The exception is narrow on purpose. Four things stay outside it whatever the purpose: sexual content involving children, intimate imagery of a real person made without their consent, practical instructions for harming people, and documents, identities or evidence made to pass as genuine. A study of a forgery does not need a forgery that works.
It does not cover work aimed at a private person. Satire addresses power; the same picture pointed at somebody who has none is harassment, and we read it as harassment. Nor does it cover work put in front of a viewer who cannot tell it is not real: satire that does not show itself to be satire is not satire.
Nothing here is claimed in advance. No sentence added to a description unlocks a refusal — a model provider answers the same way whatever reason is written for it, and we do not read descriptions looking for one. The exception is weighed afterwards, by a person and not a filter, on what has been put in front of them: the support desk cannot open your boards or read what you wrote, and sees your account, its accounting and your runs and nothing else. It is a defence, not a key.
The owner approved a narrow exception for journalism, research and satire on 2026-09-15 and left the wording to us; counsel has not reviewed it. Nothing in this clause changes what a model provider refuses on its own — see 'When a request is refused'.
When a request is refused
A model provider may refuse a request on content grounds. The refusal reaches you as a state on the node, and we do not retry automatically.
If the provider charges us nothing for a refused request, we charge you nothing. If it does charge us, that appears on your receipt exactly as it stands.
A provider refusing a request is not by itself an accusation that you broke this policy, and not everything their systems refuse is against it.
Where the service is offered
A recorded owner decision, awaiting legal wording
Generating is refused from some countries and regions, because the providers we depend on restrict where the end customer may be. A refusal on that ground accuses you of nothing, and wherever you pressed Generate — in the four workspaces or on the board — you are told that generating is not available from your current location.
The check sits on the request that is about to spend money with a provider, not on the door: signing in, reading your boards, exporting your work, deleting your account and stopping a run that is already going are none of them screened.
We infer your location from network data, and that can be wrong. Nothing on the screen carries a reference for you to quote, and the refusal is raised before any run or receipt exists, so nothing lands on your account to point at either. To tell us, write to info@upqar.ai and say when and from where you tried.
The refused-region list is recorded in the service's code, every entry records its source, and the owner confirmed it as it stands on 2026-09-15. It remains configurable, so changing it is a setting rather than a release. No country name appears in any message shown to a user.
What happens if you break this policy
A recorded owner decision, awaiting legal wording
A person decides, not a filter. Every measure below is taken by a named operator at the support desk, and the record keeps who took it, when and why — and, where the account already carried a ceiling of its own rather than the product's default, what that ceiling was before. There is no automatic suspension in this product, and a provider refusing a request is not one of these measures — see 'When a request is refused'.
The first time, unless it is one of the severe cases below, we write to you at the e-mail address UPQAR passed us when you signed in. The letter says which line of this policy is at issue and what has to stop. Nothing about the account changes. Not every account has such an address — UPQAR sends one or it does not, and there is nowhere here to give us one — and where there is none this step cannot happen at all.
If it happens again after that letter, generating is suspended for 7 days: your ceiling is set to zero, so no new work starts. Everything already made stays where it is — your boards open, your files download, an export still builds, and the account remains yours to delete whenever you want. Nothing lifts the suspension when a month turns; at the end of the seventh day an operator sets a ceiling again by hand. The desk can set a figure but cannot hand an account back to the product's default, so from that day your account carries a number of its own.
A third time and the suspension becomes indefinite: the ceiling stays at zero until we decide otherwise, and that is as far as this product goes. The account itself is an UPQAR account and only UPQAR can close one; nothing here asks it to, and no channel for asking has been built. You can still sign in, read, download and export.
Four things skip the ladder altogether, and the first occurrence is answered with that indefinite suspension: sexual content involving children, intimate imagery of a real person made without their consent, a credible threat against a specific person, and a deception aimed at an election or an official process. In the first of those we also do what the law requires of us; what that is, and to whom, is text still waiting on counsel.
A measure taken by mistake is undone on the day the mistake is shown: the ceiling goes back, and what the mistake cost you is returned to your account as a correction marked 'Our fault', which appears on your own account page in your own language. The accounting bounds that and not us: a correction can only give back against what your account is still counted as having spent inside the window your account page states, so a charge old enough to have left that window cannot be given back this way.
The owner set this ladder on 2026-09-15 and left the lengths to us; counsel has not worded it. What the product cannot do is written into the clause above rather than softened here: it sends no mail of its own, so a letter is a person's act and depends on there being an address; no timer lifts a suspension, and the desk can set a ceiling but cannot return an account to the product's default; and no measure here is reported to UPQAR — the product does reach UPQAR, to sign you in, and it carries none of this there. A suspended account also sees the ordinary 'you have used all your credits' message in the app rather than one saying a measure is in force.
If you think we got it wrong
A recorded owner decision, awaiting legal wording
You have 30 days from the day a measure takes effect to appeal it. Thirty is our own promise rather than a setting: it was chosen to sit inside the windows in which what an appeal is about still exists — a board you deleted stays restorable for a window stated on the trash page itself, and an account you asked to close stays reclaimable for a window stated on your account page. An appeal made inside them is answered while those things are still there — its runs on the record, the work still in your account — rather than against two people's memory of it.
An appeal is made by replying to the letter that told you about the measure, quoting the reference on it, or, if no letter reached you, by writing to info@upqar.ai. There is no form and no required wording: one sentence is an appeal. If no letter reached you, the 30 days run from the day you found out about the measure.
The measure stays in force while an appeal is open: an appeal that lifted a suspension the moment it was sent would not be a way to be heard, it would be a way to keep generating. It does not lengthen the measure either — a 7-day suspension ends after 7 days whether or not the appeal has been answered. Nothing is deleted because of an appeal, and the ordinary windows do not pause for one: a board you put in the trash yourself still leaves it at the end of the window the trash page states, as usual.
We answer in writing, and the answer says which way it went and why. If it goes your way an operator sets the ceiling back the same day, and what the measure cost you comes back to your account as a correction — under the same accounting bound as any other correction, described under 'What happens if you break this policy'.
The owner delegated this window to us on 2026-09-15 and counsel has not worded it. Thirty days was chosen to sit alongside the retention windows the product already keeps; those windows are deployment settings, which is why the pages that depend on them state them and this page points at those pages instead of printing a figure it cannot track. Writing to info@upqar.ai became a route for an appeal on 2026-09-29, using the address the owner named on 2026-09-28.
Reports from other people
A recorded owner decision, awaiting legal wording
To report work made here that breaks this policy or infringes your rights, write to info@upqar.ai with a link to it and what is wrong with it. If the report is about your rights in a work, say which work and why those rights are yours.
A person reads every report and answers it. Where the work breaks this policy and we can tell which account made it, that account is dealt with as 'What happens if you break this policy' sets out; a file copied out of the service carries nothing that names the account that made it, so a report of one alone may not lead to an account. A report of intimate imagery of a real person made without their consent is read within 48 hours of reaching us.
We tell the person who reported what we did. We give neither side the other's identity unless the law requires us to.
The owner's decision of 2026-09-29, written in our words as the owner's position rather than counsel's wording; the owner and counsel will revise it. Nothing in the product receives a report, so a report is a letter to this address, read by a person.