Your AI Video Prompt Was Blocked. Here's Why — and the Fix.
Content refusals are usually a vocabulary problem, not a story problem. Four things trip video-model filters: named real people, trademarked properties, age ambiguity, and violence described as injury rather than as action. Rewrite the shot in physical description instead of names, state that adult characters are adults, and stage violence as movement and reaction — most refused shots render on the next pass with the scene intact.
Why the error message doesn’t help
Provider refusals are intentionally uninformative. A specific message — "blocked because of the word X" — would let anyone map the filter in an afternoon, so you get "content policy violation" and nothing else. That leaves you guessing, and guessing is expensive: people re-render the same shot repeatedly with cosmetic changes, hitting the same trigger every time.
The useful move is to stop treating it as a mystery and check the four usual suspects in order.
1. Named real people
Any real person’s name is likely to refuse, and the ones that don’t refuse are worse — a rendered likeness of a real actor is a publicity-rights problem regardless of which tool made it.
Rewrite as physical description: not the actor, but "a lean man in his fifties, silver stubble, deep-set eyes, permanent half-smile." You get a castable original who can then be locked with reference images and reused for the whole production — which is more useful than a lookalike anyway.
2. Trademarked properties — including ones you didn’t name
The obvious version is naming a franchise. The version that catches people out is implicit: a caped figure on a gothic rooftop at night produces a recognisable superhero without a single trademarked word in the prompt, because the model has learned that combination of elements as one thing.
This is why you have to inspect the output, not only the input. If a render comes back looking like intellectual property, change the distinctive silhouette — the cape becomes a long coat, the emblem goes, the palette shifts — rather than editing the words that didn’t contain the problem in the first place.
Worth knowing. A prompt containing no IP terms can still generate an unmistakable copyrighted character. Reviewing renders for accidental likeness is part of the job, not a sign something went wrong.
3. Age ambiguity
Anything that could involve a minor is filtered hard and with a wide margin, which means adult scenes get caught by omission. If a prompt says "a student," "a young couple," or just doesn’t say, the filter has to assume the worst case.
State ages explicitly for adult characters. "Two adults in their late twenties" is four extra words that convert a large share of unexplained refusals into successful renders. Storing age on the character record so it’s applied to every shot automatically is better than remembering to type it.
4. Violence framed as injury
Drama needs conflict, and conflict gets refused when it’s written like a medical report. The filters respond much more to wound detail, blood, and weapon-to-body specifics than to the presence of a fight.
Write it as an editor would cut it: the shove, the stagger, the hand catching the doorframe, the impact off-screen, the cut to a witness’s face. That renders reliably, and it’s also stronger filmmaking — implication has always beaten depiction in this genre.
When rewriting isn’t the answer: route it elsewhere
Thresholds differ meaningfully between providers and even between models from the same provider. A legitimate dramatic beat refused by one model routinely renders on another with an unchanged prompt.
That makes model routing a real capability rather than a convenience. If your tool is locked to one provider, its policy is your policy; if a refused shot can be re-submitted to a different model automatically, a whole category of dead end stops being one.
How ShowMaker Studio handles it
Refusals come back as plain language telling you which of these four categories was probably hit and what to change, instead of a raw provider error code. Character ages and descriptions live on the cast record and are applied to every shot, so age ambiguity doesn’t creep in per prompt. Shots can be re-routed across providers when one model declines. And prompts are compiled per model rather than sent raw, so the same directorial intent is expressed in whatever vocabulary the target model accepts — see the storyboard guide.
Related questions
Why was my AI video prompt blocked with no explanation?
Providers deliberately keep refusal messages vague so the filter can't be reverse-engineered. In practice the trigger is usually one of four things: a named real person, a trademarked property or brand, wording that could imply a minor in an unsafe context, or violence described in injury terms rather than action terms. Almost all of these are vocabulary problems you can rewrite around without changing the story.
Can I use a real actor or celebrity in AI video?
No. Naming a real person will be refused by most providers, and where it isn't refused it exposes you to publicity-rights and likeness claims. Describe the physical type you actually want — age, build, colouring, bearing — and cast a consistent original character from that description instead.
Why does my prompt get blocked when it contains no banned words?
Because the filters read intent and combinations, not a word list, and because reference images are moderated too. A prompt with no flagged terms can still generate a recognisable copyrighted character if the model associates your described elements with one — a caped figure in a dark city produces a specific superhero whether or not you named him. Check what came back, not just what you sent.
How do I write violence for a drama without being refused?
Describe action and reaction rather than injury and gore. A shove, a fall, a flinch, an off-screen impact, and a cut to a reaction shot carry a fight perfectly and read as filmed drama. Detailed wounds, blood volume, and weapon-to-body specifics are what trips filters — and in edited drama they're rarely the shot you needed anyway.
Do different providers block different things?
Yes, significantly. Thresholds vary by provider and by model, so a shot refused by one may render on another with the same prompt. A pipeline that can route a refused shot to a different model gets you the shot without a rewrite, which is why single-provider tools feel more restrictive than they technically are.
Should I write character ages into every prompt?
For adult characters, yes — it's cheap insurance. Ambiguity about age is one of the most aggressively filtered signals, and a scene involving young characters or any physical contact will refuse far less often when every character is explicitly described as an adult. Storing age on the character record rather than retyping it per shot makes this automatic.
Every fix on this page is a feature, not a workaround.
ShowMaker Studio is built around these problems — reference-locked casting, location look signatures, coverage-based scenes, delivery-timed captions, and per-shot model routing. Three free films a month, no card.