Write prompts that don't get flagged
Most flagged requests in Flick are ordinary film scenes that a model's filter misread: a fight, a period execution, a bath, a superhero homage. The filter does not know your story. It reacts to certain words, to what a reference image looks like, and to the finished result. Write around those triggers and the same scene usually runs.
A flagged generation costs nothing. The credits held for it are returned automatically.
How the filters work
- Each model company runs its own filter. Flick sends your request to the model you picked, and that company's filter decides. This is why one model can refuse a request another accepts.
- Filters check three things. The words in the prompt, every reference image, and the result after it is made. A prompt can pass and the finished video can still be refused.
- Filters read words, not intent. "Execution", "blood" or "undressing" reads the same in a historical drama as anywhere else. A respectful project gets no exception.
What gets flagged most
1. Famous characters, franchises and brands
Names like Superman, Spider-Man, Darth Vader or a Pokémon are refused by several image and video models, and so are stories about two franchise characters meeting.
Removing the name is not enough. A prompt that says "a hero in a red and blue suit with web patterns and a mask" still describes Spider-Man, and is refused the same way.
The same copyright check reads brands and real media. A real logo, a brand name on a product, or a still from an existing film used as a reference can get a video refused, sometimes only after it is made.
Instead: make the character your own. Give them a new name, their own costume, colours and silhouette, and build a Character Reference for them. An original hero also stays consistent across shots, because the model is not pulled toward the famous version. For products, describe an unbranded one ("a red soda can"), and use references you made rather than frames from someone else's film.
2. Real people
Prompts that name a celebrity, politician, athlete or other public figure are often refused, and so is "looks like" a named actor.
Some video models also check reference images for real people. A photo of a real person is refused, and occasionally a very realistic AI character is misread as one. The message says the image "may contain a real-person likeness".
Instead:
- Describe the look rather than naming someone: "a man in his 50s with a grey beard, deep-set eyes and a weathered face".
- Use a character generated in Flick as the reference, not a photo of a real person.
- If a generated character is misread as a real person, use a slightly more stylized version of its reference, or try another model.
- Edits that put one real person's face onto another are refused.
- Never use someone's face or voice without their consent.
3. Violence, weapons and injuries
Graphic violence is the most common trigger. Wounds, blood, executions, a weapon aimed at someone, and words like "kill" or "kick" in a prompt can all trip a filter, even in a non-graphic scene.
Instead: write what the camera sees, not the harm.
- Show the moment before or after the blow, or cut to a reaction.
- Keep injuries non-graphic: "a bruise on his cheek, a torn and muddy sleeve" rather than "a swollen, bleeding eye".
- Hold a weapon as a prop ("she grips the sword at her side") rather than aimed at a person.
- Do not stack triggers. An injury plus a Horror style plus a dark prompt is far more likely to be refused than any one of them. Make the clean version first, then add detail with an edit.
4. Nudity and suggestive content
Bathing, undressing, "sexy", body-focused descriptions, and flirtatious dialogue are often refused, even when nothing explicit is shown. So is naming a character's profession when it implies sex work.
Instead:
- State the clothing: "wearing a white robe", "wrapped in a towel".
- Frame from the shoulders up, or let steam, water or a doorway cover the body.
- Describe the mood rather than the body: "a quiet, intimate moment" instead of how someone looks.
- Describe what a character does, not a label: "two elegantly dressed women wave to the crowd".
5. Children
Scenes that put a child in danger, injury, bathing or romance are often refused. Edits to a photo with a child in it are often refused too, even removing the child from a family photo.
Age is also read from wording. "A girl in her late teens" next to a description of her looks can be refused where "a woman in her twenties" is not.
Instead: give adults an explicit adult age, keep child characters out of dangerous or intimate moments, and do not upload photos of real children for editing.
6. Charged words in dialogue and emotion
Some models check the spoken lines and emotional wording too. A harmless argument, an insult, a threat said as a joke, or a pile of distress words ("smudged, wet cheeks, panic, terrified") can be refused.
Instead: keep what the line means and change the words. "I'll kill you" can become "You'll regret this". One clear emotion ("she is close to tears") works better than a list.
7. Sound and voices
Video models that make sound with the picture check the generated sound as well. If the message says the generated audio was flagged, turn off Generate Audio in the model's settings and add the sound separately, or reword the dialogue.
Voice and dialogue generation checks the spoken text, so the rewording in the section above applies there too. Clone only a voice you have permission to use: a celebrity's voice is treated like their face.
What no rewrite fixes
Some requests are refused because they are against Flick's Terms of Use, and there is no wording that makes them acceptable: harming or sexualizing minors, harassing or threatening real people, hate against a group, using someone's likeness or voice without consent, and fake real-world events meant to mislead. Trying to get around a content filter is against the Terms too.
Quick rewrites
| Often refused | Try |
|---|---|
| Superman flies over the city | A hero in a dark green suit with a silver crest flies over the city |
| A man who looks like Keanu Reeves | A lean man in his 40s with shoulder-length dark hair and a short beard |
| He stabs the guard and blood sprays | He lunges at the guard; cut to the guard falling out of frame |
| The prisoner is executed on the guillotine | The crowd falls silent as the blade is raised; the camera stays on their faces |
| A naked woman in the bath | A woman in a steaming bath, seen from the shoulders up |
| A teenage girl in a bikini | A woman in her twenties in a swimsuit, walking along the beach |
| Her face is swollen and bloody after the crash | After the crash: a bruise on her cheek, dust in her hair, a torn sleeve |
| Blood pools on the floor | A dark stain/water spreads across the floorboards |
When a generation is flagged
- Change one thing, the smallest one. Find the word or image most likely to have tripped the filter, using the list above, and change only that. Keep the rest of the prompt so the scene still looks like your scene.
- Then try the references. If the words look fine, run once without the reference image most likely to be read as a real person or a famous character.
- Watch how long it took. A refusal within seconds came from the prompt or a reference, and the same request will be refused again. A refusal after the video was made came from a check on the result, and one more try sometimes passes.
- Ask the chat agent to help. Tell it which shot was flagged and ask it to reword only what is likely to be refused, keeping the scene the same, and to tell you what it changed.
- Try another model only for a misread. When the request is within the rules and one model keeps misreading it, another model may accept it. Content against the rules is refused everywhere, and misspelling or disguising words to get past a filter is against the rules.
Some long video jobs only report a refusal after 15 to 30 minutes. Leave the job running; it will show the result. See Where is my generation.
Next
- Why a generation was flagged for what a refusal means and how to report a mistake.
- Image Generation Prompts Guide for writing prompts that describe what is seen.
- Character Reference for building an original character that stays consistent.