Guide
Why a prompt gets refused
Three different things stop a generation and they arrive looking identical. One will never move, one may well render on a different route, and one is a wording problem you can fix in a minute — so the first useful step is telling them apart.
The first kind
A platform rule, which does not move
Some things are refused everywhere on this platform regardless of model, route, wording or intent. Anything depicting a minor is the clearest case, and it is absolute: no artistic framing, no stated age beside a contradicting description, and no claim that a depicted person is fictional changes it.
There is no technique for this one and it is worth saying plainly rather than leaving people to experiment. A refusal of this kind is the system working, and repeated attempts around it are recorded.
The second kind
The model's own policy, which is narrower than ours
Every model this platform routes to belongs to someone else, and each has its own content rules on top of these. Those rules differ from model to model and are frequently stricter than the platform’s — which produces the confusing case where something permitted here is still declined, and the same request succeeds later on a different route.
This is the refusal most worth recognising, because it is the one where trying again is sensible rather than stubborn. Rewording often clears it: the filter is reading the phrasing, not your intention, and a description that leads with the setting and the people rather than with the act tends to pass where the reverse does not.
Nothing is charged for it. A refusal produces no output, so the credits return automatically at the moment it fails, and a refused attempt does not consume a free trial attempt either.
The third kind
An ambiguity you wrote without meaning to
This is the largest group and the most frustrating, because the request was entirely legitimate and the refusal still looks like an accusation.
Screening has to judge a description on what it contains. When a description leaves something unstated, the check has to decide without it, and on anything touching age it is built to err toward refusing. A scene can be completely lawful and still be stopped for saying nothing either way.
Two habits remove most of these. State the age or age range of everyone in the scene, and do it in every shot of a film rather than only the first, because each shot is checked on its own and inherits nothing from the one before it. Name relationships between adults explicitly rather than leaving them implied — family and in-law wording has historically been misread as a signal about age when the ages were not stated.
Words that describe a child stop a generation whatever else is written, including when an adult age is given beside them. That is not a filter to argue with by rephrasing around it; it is the first kind of refusal, and it is deliberate.
Output screening
Sometimes the prompt passes and the result does not
A generation can be accepted, run, and then be withheld. That is not inconsistency — the prompt and the output are screened separately, because what a model produces is not always what was asked for.
Video is screened by sampling frames, which is why a clip is occasionally stopped over something that appears briefly and is not what the scene is about at all. When that happens the credits are returned, and on a multi-shot film it costs that shot rather than the film.
If a refusal looks plainly wrong, it is worth reporting rather than working around — the screening is tuned from cases like it, and the compliance page sets out what is permitted and how to reach someone.
Asked often
Short answers
- Why was my AI prompt rejected?
- There are three separate reasons and they are worth telling apart. A platform safety rule, which is absolute and will not move. The model provider's own content policy, which is narrower than the platform's and varies between models, so a different route may well render the same thing. Or an ambiguity in the wording that made the request read as something it is not - and that one is fixed by writing more precisely.
- Do I get charged for a refused generation?
- No. A refusal produces nothing, so the credits are returned automatically at the moment it fails rather than on request, and a refused attempt does not count against a free trial allowance either.
- My prompt was clearly fine. Why was it still stopped?
- Most often because an age or a relationship was left to be inferred. Screening has to judge from what is written, and when a description gives it nothing it errs toward refusing. Stating the ages of everyone in the scene, in every shot rather than only the first, removes the ambiguity that causes most of these.
Keep reading
The rest of the series
- Does the language of your prompt matter?The script matters more than the language, and the gap is larger than anyone tells you.
- Why AI changes the face every timeThe mechanism behind the problem: no memory, and a prompt that names a category rather than a person.
- How to get the same character in every imageFive approaches, what each costs, and the cases where training a LoRA is still the right answer.
Where this comes from
Unlimited Studio is built around this one problem
Everything above is what the product does: a character is locked from eight reference angles, each generation conditions on whichever angle its prompt implies, a score checks the result against the reference, and a film is shot one frame-chained shot at a time. What it will and will not generate is set out in the content policy.