Runway AI Prompt Guide: How to Write Better Video Prompts
A practical guide to clearer text-to-video and image-to-video prompts using visual description, motion, camera direction and controlled iteration.
Good AI video prompting is less about writing the longest possible description and more about communicating a clear visual idea. Runway's current prompting guidance emphasizes direct language, useful motion descriptions and iterative refinement. The exact prompt you need also depends on whether you are generating a scene from text or animating an existing image.
This guide explains a practical workflow you can reuse when building Runway prompts. The examples are original PromptVanta examples designed to demonstrate the principles rather than guarantee an identical output from every generation.
1. Start With the Generation Mode
Before writing the prompt, identify what information the model already has. This changes what your text needs to explain.
Text to Video
With text-to-video, the prompt has to establish the visible scene as well as its movement. Describe the important subject, environment and visual direction, then explain the action and camera behavior that matter to the shot.
Image to Video
With image-to-video, the input image already supplies much of the subject appearance, composition, lighting and style. Your text can therefore concentrate primarily on what should move and how the shot should develop over time.
2. A Practical Runway Prompt Structure
There is no requirement to follow one rigid formula. A consistent structure is still useful because it makes prompts easier to write, diagnose and revise. For a text-to-video shot, try:
[Shot / camera] + [subject] + [action] + [environment] + [important supporting visual or motion details]
For example:
Medium tracking shot of a cyclist riding along an empty coastal road at sunrise. The camera moves smoothly beside the cyclist as ocean mist drifts across the road. Warm backlight, long natural shadows, cinematic realism.
For image-to-video, a motion-focused structure can be simpler:
The camera [camera movement] as the subject [subject action]. [Environmental motion or timing detail].
Example:
The camera slowly pushes toward the subject as she turns toward the window. Thin curtains move gently in the breeze while soft daylight shifts across the room.
3. Describe Visible Actions Instead of Abstract Intentions
Video models need to translate language into visible frames. Concrete physical actions therefore provide clearer direction than vague emotional or conceptual instructions.
Instead of relying only on a phrase such as “make the scene feel tense,” describe visible cues that create tension: the character slowly backs away, their hands tighten around an object, the camera moves closer, or a flickering light changes the environment.
This does not mean that mood words are useless. Terms such as cinematic, energetic or dreamlike can help establish direction, but they work best when the core action remains understandable.
4. Separate Subject Motion, Scene Motion and Camera Motion
Thinking about motion in three categories can make a prompt easier to control.
Subject motion
Describe what a person, animal or object physically does: walks forward, looks over a shoulder, raises a hand, falls, rotates or accelerates.
Scene motion
Describe changes in the environment: dust trails behind a vehicle, rain strikes a window, leaves move in the wind, smoke rises, or waves break against rocks.
Camera motion
Describe how the viewpoint changes. Useful directions include a locked camera, pan, tracking movement, dolly-style push-in, pull-back, handheld movement or an arc around the subject.
A runner moves through a narrow wet alley. The handheld camera tracks behind the runner while puddles splash under each step and loose paper moves in the wind. Natural nighttime street lighting.
Notice that the prompt gives each type of motion a clear role instead of stacking several unrelated movements into the same instruction.
5. Use Camera Language With a Purpose
Camera terminology can strongly change how a scene feels, but more camera terms do not automatically produce a better shot. Choose the movement that supports the action.
- Locked camera: useful when you want movement to come primarily from subjects or the environment.
- Tracking shot: follows a moving subject through the scene.
- Push-in: gradually brings attention closer to a subject or detail.
- Pull-back: reveals more of the environment or increases visual distance.
- Pan: rotates the view horizontally to follow or reveal something.
- Handheld: introduces natural camera movement that can create a documentary or immediate feeling.
If you need a mostly static composition, state that directly and make clear what is supposed to move within the frame. Generative video systems are built to create motion, so a scene with no described movement can still develop unexpected camera or environmental movement.
6. Keep a Single Shot Focused
A common prompting mistake is trying to compress an entire film sequence into one generation: several locations, multiple transformations, changing styles and numerous camera moves. Even when a model can follow complex instructions, a focused shot is easier to control and easier to revise.
If the concept contains multiple distinct story beats, split it into separate shots. Each prompt can then concentrate on one clear visual event.
Overloaded concept
A woman runs through a forest, enters a city, changes into a robot, flies into space and the camera spins around her before diving underwater.
Focused shot
Wide tracking shot of a woman running through a misty forest at dawn. The camera moves parallel to her as branches and ferns pass through the foreground. Soft diffused light, natural running motion.
7. Text-to-Video Prompt Example
For text-to-video, the model does not begin with an input image establishing the frame, so the visual description matters alongside the motion.
Wide cinematic shot of a small fishing boat crossing dark blue water before sunrise. A fisherman stands near the bow while the boat rises gently over slow waves. The camera tracks parallel to the boat from a low angle. Cold pre-dawn light, distant fog, realistic water movement, restrained cinematic look.
This prompt identifies what is visible, where it is happening, how the subject and environment move, how the camera behaves and what broad visual treatment is intended.
8. Image-to-Video Prompt Example
If your starting image already shows a woman standing beside a window in a softly lit room, repeating every detail of her appearance may be unnecessary. The prompt can focus on temporal change:
The woman slowly turns her head toward the window and takes one small step forward. The curtain moves gently in the breeze. The camera performs a subtle slow push-in. Continuous natural motion.
This approach lets the image carry the visual information it already contains while the text concentrates on animation.
9. Use Positive, Direct Instructions
Runway's prompting guidance recommends describing what should happen rather than building the prompt around negative commands. For example, instead of repeatedly saying that the camera must not move, describe the intended state directly as a locked or motionless camera.
Direct wording also reduces ambiguity. “The man walks slowly toward the doorway” communicates a visible event more clearly than a command asking the model to “make him more dramatic.”
10. Iterate One Important Variable at a Time
Prompting is an iterative process. When the first result is close but not correct, rewriting every sentence at once makes it difficult to know which change improved or damaged the result.
Start with the essential action. Then refine one important dimension at a time: subject motion, camera behavior, scene motion, timing or visual style. This creates a more controlled workflow and helps you discover how a particular model responds to your language.
11. Common Runway Prompting Mistakes
- Too many events in one shot: separate distinct story beats when the prompt becomes difficult to follow.
- Conflicting camera directions: avoid asking the camera to perform incompatible movements simultaneously unless the sequence genuinely requires them.
- Repeating the input image unnecessarily: in image-to-video, prioritize desired motion when the visual information is already established.
- Vague actions: replace abstract instructions with physical, observable behavior.
- Changing everything after a failed generation: revise the element that actually needs correction.
- Ignoring the starting image: poses, blur and other visual cues in an image can imply motion and may influence the animation.
12. Continuity Across Multiple AI Video Shots
When creating a sequence, each generated clip should still be treated as a controlled shot. Keep the important identity anchors consistent across prompts: the same character description when needed, clothing, key props, location, time of day and overall visual direction.
At the same time, avoid copying a huge block of irrelevant text into every shot. Repeat the details that genuinely need to remain stable, then clearly state what changes in the current scene. This makes the prompt easier to understand and edit.
For a deeper workflow, read the Prompt Continuity Guide.
13. Quick Prompt Checklist
Before generating, ask:
- Is the main subject or visual focus clear?
- Is the main action physically understandable?
- Does the environment need any specific movement?
- Have I chosen one useful camera behavior?
- Am I describing what should happen rather than mainly what should not happen?
- Does this prompt represent one manageable shot?
- If I am using an input image, am I focusing the text on the motion that is not already communicated by the image?
Frequently Asked Questions
Should Runway prompts be long?
Not necessarily. Clear, direct prompts that focus on the most important visual and motion information are easier to understand and iterate. Add more detail when it serves a specific purpose.
What should an image-to-video prompt describe?
When the input image already establishes subject appearance, composition, lighting and style, focus primarily on the desired motion: subject action, environmental movement, camera behavior, direction and timing.
What should a text-to-video prompt include?
Describe both the visible scene and its motion. A practical prompt can include the subject, environment, framing or style plus subject action, scene movement and camera movement where relevant.
How do I improve a prompt that is not working?
Return to the simplest version of the desired shot and refine it gradually. Change one important variable at a time so you can identify which instruction is affecting the result.
Create a Structured Video Prompt
PromptVanta can turn a short concept into a structured scene prompt that you can review and edit before using it in your preferred AI video workflow.
Related PromptVanta Guides
How to Write AI Video Prompts Cinematic AI Video Prompts Prompt Continuity Guide Kling AI Prompt Guide All AI Prompt Guides