How Runway reads a prompt
Runway's video models, from Gen-3 Alpha onward, turn text or images into short cinematic clips. They respond best to prompts that read like a director's note: what the camera does, what we see and how things move. Runway's own guidance suggests a structure that leads with the camera movement, then the scene, then extra details. Keep each prompt focused on one shot rather than a sequence of events.
When you start from an image, the picture already defines the look, so your prompt should focus on motion: how the subject moves, what the camera does and how the environment reacts. Avoid negative phrasing such as "no blur", which can produce the opposite effect. Duration, aspect ratio, seed and reference options are set in the interface and vary by model generation, so check Runway's help center for current options.
The prompt formula
[camera movement]: [scene and subject]. [subject motion]. [lighting and atmosphere]. [style]
Example
Slow dolly in: a lone astronaut stands on a red desert dune at dusk. Her cape ripples in the wind as she turns to look at two moons rising. Warm low sunlight and drifting dust. Cinematic, anamorphic film look.
Runway parameters and controls
| Control | Syntax | What it does |
|---|---|---|
| Camera movement words | Plain language at the start of the prompt | Describe the move directly: dolly in, truck left, crane up, handheld, orbit, static.e.g. Low angle tracking shot: ... |
| Image to video | Upload a first frame | Uses your image as the starting frame. Prompt the motion rather than re-describing the image.e.g. The woman slowly turns her head and smiles as wind moves her hair. |
| Duration | Duration selector | Sets clip length. Short durations such as 5 or 10 seconds are typical at the time of writing.e.g. 10 seconds for a slow reveal |
| Keyframes | First / last frame inputs (model-dependent) | On models that support it, set start and end images so Runway animates between them.e.g. Day exterior as first frame, night exterior as last |
| Aspect ratio | Ratio selector | Chooses landscape, portrait or square output depending on the model.e.g. 9:16 for vertical social video |
| Seed | Seed setting | Fixes randomness so you can compare prompt changes more fairly.e.g. Fixed seed while testing camera wording |
Do this
- Lead with one clear camera move; combining several moves in a short clip often looks chaotic.
- Describe motion with strong verbs: drifts, sprints, unfurls, ripples, tilts.
- For image-to-video, prompt only the motion and camera, not the details already in the image.
- Keep each prompt to a single shot and location.
- Phrase everything positively; describe the calm sky rather than writing "no clouds".
- Add atmosphere words (dust, mist, rain, light flicker) to make scenes feel alive.
Avoid this
- Writing a whole storyline with multiple scene changes in one prompt.
- Using negative instructions like "no camera shake", which may add shake.
- Asking for fast, complex action and many characters at once, which increases warping.
- Re-describing an uploaded image in detail instead of describing how it should move.
Runway prompt examples
Cinematic reveal
Crane up: a misty pine forest at dawn slowly gives way to a vast mountain lake. Sunlight breaks over the ridge and sparkles on the water. Cool blue shadows turning to warm gold. Cinematic, slow and majestic.
Product spin
Static camera: a matte black wireless earbud case rotates slowly on a glossy turntable. Soft studio light glides across its curved surface. Clean gray background, premium commercial look.
Character moment
Handheld close-up: a young violinist on a rainy street corner closes her eyes and begins to play. Raindrops bounce off her umbrella, passersby blur in the background. Moody streetlight, intimate documentary feel.
Image-to-video motion
The camera slowly pushes in as the old lighthouse beam begins to sweep across the waves. Clouds drift quickly overhead and sea spray rises against the rocks.
Runway prompt FAQ
How should I structure a Runway prompt?
A reliable structure is camera movement, then scene, then details: for example, "Slow tracking shot: a cyclist rides along a coastal road at sunset. Warm light, long shadows." Keep it to one shot and describe motion with clear verbs. Runway publishes prompting guides for each model generation, so check them when a new model is released.
Should I write a prompt when using image-to-video?
Yes, but keep it about motion. The image already defines subjects, colors and composition, so describe what moves and how the camera behaves. Repeating visual details can make the model fight the image. A short prompt like "the camera slowly orbits as leaves fall around her" usually works better than a long description.
Does Runway support negative prompts?
Runway does not rely on a negative prompt field, and its guidance warns that negative phrasing can backfire. Writing "no blur" may introduce blur because the model latches onto the word. Instead, describe the result you want: "sharp, steady shot" or "clear blue sky". If problems persist, simplify the motion or try a different seed.
How long can Runway videos be?
Individual generations are short, typically around 5 to 10 seconds depending on the model and plan at the time of writing. Longer sequences are built by extending clips or editing several generations together. Plan shots as separate prompts, keep lighting and style wording consistent, and check Runway's current documentation for limits.