Why prompts from this tool look different
Most AI video prompts fail for the same reason student films do: nobody decided what the camera is doing. A model given "a woman walks in an alley" invents the shot size, the movement, the light and the mood — and you re-roll five times hoping for luck. Working directors don't describe scenes that way, and neither should you. This generator forces the four decisions every real shot starts with — framing, movement, light, style — and phrases them in the film vocabulary the models were trained on.
How each model wants its prompt
| Model | Sweet spot | Special powers |
|---|---|---|
| Veo 3 | Long cinematic paragraphs (60–120 words) | Native audio: dialogue in quotes, sound effects, ambient noise |
| Sora | Rich visual description | Strong scene physics and multi-subject staging |
| Kling | Concise prompts, strong motion verbs | Separate negative-prompt field; good motion control |
| Runway Gen-4 | Short, positive phrasing, camera move first | Excellent image-to-video; keywords over prose |
| Seedance | Action first, then camera, then atmosphere/sound | @image references for character consistency; native Chinese & English; strong multi-shot narrative |
Want the full theory — shot grammar, movement vocabulary, per-model quirks and worked examples? Read our companion guide: How to Write AI Video Prompts Like a Filmmaker.
Prompt like you'd brief a DP
The fastest mental model: you're not "writing a prompt", you're briefing a director of photography who has never seen your storyboard. Name the subject and what they do, where they are, how big they are in frame, how the camera moves, what the light source is, and what the finished frame should feel like. Everything this tool asks maps to that briefing — which is also why its output transfers between models so well: good shot description is model-agnostic; only the formatting changes.
FAQ
Do the same prompts work for Veo, Sora, Kling and Runway?
The ingredients are the same — subject, action, setting, shot, camera move, light, style — but each model prefers a different recipe. Veo 3 rewards long cinematic descriptions and understands audio cues; Runway’s official guidance asks for short, positively-phrased prompts with the camera move up front; Kling responds best to concise prompts with a strong motion description; Sora likes rich visual detail. That’s exactly why this generator outputs a separately-formatted version per model.
Should I write AI video prompts in English?
Yes, for best results. All major video models are trained predominantly on English captions, and film vocabulary ("dolly-in", "golden hour", "anamorphic") carries precise meaning the models have seen millions of times. Write in English even if your team works in another language — this tool does the phrasing for you.
How long should a video prompt be?
Veo 3 and Sora handle 60–120 word cinematic paragraphs well. Kling and Runway generally do better under ~50 words — over-stuffed prompts cause them to ignore instructions or produce muddled motion. One idea per shot: if you need two camera moves, generate two shots.
Why does my AI video ignore my camera movement?
Usually because the prompt buries the movement mid-sentence, or asks for several movements at once. Put the camera instruction early (this generator does it for Runway automatically), use standard film terms rather than vague ones ("slow push-in" beats "camera gets closer"), and limit each generation to a single move.
What is a negative prompt and do I need one?
A negative prompt lists what you don’t want (extra fingers, text artifacts, warped faces). Kling and some platforms expose a separate negative field — the generator’s Kling output includes a suggested one. Veo and Sora work with positive phrasing only, so describe what you want instead of what you don’t.