Prompt writing is not about filling a text box with adjectives. It is about making the shot legible: one subject, one dominant action, one camera intention, and a clear endpoint. The current MotionVeo video models accept short clips, so a prompt that describes a single beat is easier to evaluate than a miniature screenplay.
The seven-part formula
- Subject: name the main object, person, or creature and its important visual features.
- Action: use a concrete verb such as turns, lifts, crosses, unfolds, or reaches.
- Environment: place the subject in a physical setting with one or two useful details.
- Camera: state whether the camera is locked, tracking, panning, tilting, or pushing in.
- Timing: describe the start, middle, and final beat when sequence matters.
- Light and style: give direction, contrast, color, texture, or a broad visual treatment.
- Output constraint: state 16:9 or 9:16 and any framing boundary.
Before and after
Before: “A cool cinematic robot in a city.”
After: “A small maintenance robot rolls beneath a glass transit station before sunrise. It pauses beside a puddle, projects a soft amber map, then turns toward the arriving train. Locked camera with a slow push in, cool blue shadows and warm map light, robot centered and fully visible, landscape 16:9.”
Use verbs that can be seen
Words such as “epic,” “beautiful,” and “professional” are broad intentions. Pair them with an observable choice: “high-contrast side light,” “slow lateral track,” “fabric lifts in the wind,” or “the subject exits frame left.” If a sentence cannot be represented by a change in pixels over time, it may not help the generation.
One action beats many actions
Short clips become harder to control when the prompt asks for a subject to run, jump, transform, speak, fight, and change location in the same moment. Pick the one beat that matters. You can build a sequence from several clips later.
Camera language without confusion
Separate camera movement from subject movement. “The camera tracks right while the cyclist moves left” is clearer than “dynamic movement.” If the camera should remain stable, say “locked camera” or “static framing.” Mention the endpoint: “the subject exits frame right by the final second.”
A source image can establish the first frame, but it does not remove the need to describe motion. Start with the visual fact you want preserved, then add one dominant change.
Troubleshoot by changing one sentence
| Symptom | Likely adjustment |
|---|---|
| The subject stays static | Replace abstract style words with a concrete verb and a start-to-end action. |
| The camera drifts | State “locked camera” or name one deliberate camera move; remove competing movement words. |
| Identity changes | Use a cleaner source/reference image and repeat the stable visual features once. |
| The framing is wrong | Choose 16:9 or 9:16 explicitly and describe where the subject stays in frame. |
| The scene feels overloaded | Remove secondary actions, extra characters, and decorative adjectives from the first pass. |
Respectful, usable prompting
Describe fictional characters or people you have permission to depict. Avoid requests that impersonate a real person deceptively, imitate a living artist as if the work were theirs, or rely on copyrighted brand elements you do not have rights to use. A clear prompt should also avoid promising unsupported audio, dialogue, or exact likeness controls unless the selected model documents them.
Try the formula in MotionVeo
Start with the AI video generator, then compare the catalog pages for Veo 3.1 Fast, Seedance 2.0, and Wan 2.6. Keep your first test simple enough that you can explain why the next version changed.