Runway Gen-4 Prompting Guide: How to Get the Best Results

Runway Gen-4 is a fast, flexible video generation model built for creators who need output that holds up next to live-action footage, animation, and VFX work. From a single image and a text prompt, Gen-4 produces 5 or 10-second video clips that are visually coherent, physically grounded, and ready to drop into a real production pipeline.

But like any AI model, the quality of what you get out is directly tied to what you put in. This guide breaks down exactly how to write prompts that get the most out of Gen-4 — from foundational principles to specific motion techniques and ready-to-use prompt patterns.

Start Simple, Then Build

The most common mistake new users make is loading up their first prompt with every possible detail. Gen-4 actually performs better when you start minimal and iterate toward complexity.

Begin with the single most important motion in your scene. Get that working cleanly. Then layer in additional elements one at a time:

1. Subject motion — what the character or object does
2. Camera motion — how the camera moves or holds
3. Scene motion — how the environment responds or behaves
4. Style descriptors — the visual or cinematic tone

This incremental approach gives you visibility into what's driving your results — and makes it much easier to troubleshoot when something doesn't land the way you expected.

The Four Prompt Elements

Subject Motion

Subject motion is the core action — what a character or object is doing physically. This includes movement, gestures, facial expressions, and reactions.

When referring to characters, use general terms like "the subject" or a simple pronoun rather than restating the full description from your image. The model already has that information from the image itself. Repeating it in the prompt tends to confuse the model or reduce motion quality.

The subject slowly turns to face the camera, squinting against the wind.

For scenes with multiple subjects, use clear spatial or descriptive identifiers to assign actions to specific characters:

The figure on the left crouches behind the wall. The figure on the right continues walking forward, unaware.

Scene Motion

Scene motion describes how the environment itself moves or reacts — whether in response to subject actions or independently.

There are two ways to trigger scene motion:

Implied motion — letting the environment react naturally through description:

The subject sprints across the sand dunes.

(Implied: dust, disturbance, wind-blown sand)

Explicit motion — directly stating what the environment should do:

The subject sprints across the sand dunes. A cloud of fine dust rises and trails behind them.

Implied motion tends to produce more organic-looking results. If it's not coming through clearly, try stating it directly or reinforcing it with additional descriptive language.

Camera Motion

Camera motion defines how the camera moves through the scene. Gen-4 responds to standard cinematography terminology, so you can use familiar filmic language directly in your prompts.

Some effective terms to work with:

Example:

A handheld camera tracks the subject as they move through the crowded market. The motion is loose but intentional, weaving between stalls.

Style Descriptors

Style descriptors set the broader visual or motion language of your clip. They're useful for communicating things like motion speed, rendering aesthetic, or cinematic genre — elements that influence the overall feel rather than specific actions.

Slow motion. Cinematic live-action. Warm golden hour light.

Stop-motion animation aesthetic. Slightly jerky but deliberate movement.

Style descriptors can be woven into the body of your prompt or appended at the end as a finishing layer.

Best Practices

Use positive phrasing only

Gen-4 is designed to understand what should happen — not what shouldn't. Negative or prohibitive language ("don't move", "no shaking") tends to produce unpredictable results or even the opposite of what you want.

❌ Avoid:

No camera movement. The camera doesn't move. NO MOVEMENT.

✅ Instead:

Locked camera. The frame holds completely still throughout.

Be direct and physical

Abstract or poetic descriptions force the model to interpret your intent, which often leads to random or surprising motion. Always translate emotional or conceptual ideas into specific physical actions.

❌ Avoid:

The subject radiates an aura of warmth and welcomes the visitor with open-hearted presence.

✅ Instead:

The subject smiles broadly, opens their arms wide, and pulls the visitor into a hug.

Focus on motion, not the image

Your input image already tells the model what the scene looks like. The text prompt should focus on what changes — how things move, what happens next, how the camera behaves.

Restating visual details that are already visible in the image (clothing, hair color, setting) tends to reduce motion output or cause the model to over-focus on appearance rather than action.

❌ Avoid:

The tall man with dark curly hair in a grey suit and brown shoes reaches forward to shake someone's hand.

✅ Instead:

The man extends his arm for a handshake, then gives a polite nod.

Skip the conversation, describe the scene

Gen-4 is a visual model, not a chat interface. Conversational framing ("Can you add...", "Please show...") and command-based language ("Add a dog to the scene") don't give the model enough visual information to work with.

If you want something to appear in the scene, describe how it enters:

❌ Avoid:

Please add a dog running through the park.

✅ Instead:

A golden retriever bursts into the frame from the left, chasing a rolling tennis ball across the grass.

Keep it to one scene

Gen-4 generates 5 to 10-second clips — which is essentially one short scene. Trying to pack multiple scene changes, style shifts, or unrelated actions into a single prompt pushes the model in too many directions at once and typically results in inconsistent output.

Focus each generation on a single moment with one clear primary action.

❌ Avoid:

A cat turns into a phoenix, flies through a jungle that changes from day to night, then transforms into a submarine in a futuristic underwater city.

✅ Instead:

A cat crouches in tall grass, then leaps upward — and as it rises, brilliant orange feathers begin spreading across its body.

Working with Image Prompts

When using an input image alongside your text prompt, think of them as a team: the image handles the visual setup, and the text handles the motion.

A few principles to keep in mind:

• Let the image do the visual work — your text should focus entirely on how things move or change
• Keep motion grounded — physically plausible motion tends to produce cleaner results than extreme or impossible movement
• Add camera movement to create dynamism even in scenes with minimal subject motion

Portrait / character animation:

The subject slowly turns their head toward the camera. A faint breeze disturbs their hair. Their expression shifts subtly — from neutral to a quiet, guarded look. Locked camera, medium close-up.

Environment animation:

The landscape comes alive — clouds drift from right to left, casting slow-moving shadows across the valley. A gust of wind ripples the surface of the tall grass. Locked wide shot.

Product or object:

The camera performs a smooth orbit around the object, moving from the front to a three-quarter rear view. Soft studio lighting remains consistent. The subject sits still while the camera is the only thing moving.

Quick Reference: Prompt Structure

Here's a simple formula you can use as a starting point for any Gen-4 prompt:

[Subject] [action/motion]. [Scene environment reacts or behaves]. [Camera move]. [Style].

Example:

The woman stands at the edge of the cliff and raises her arms. Her coat billows violently in the wind behind her. Slow dolly push-in from a wide to medium shot. Cinematic live-action. Overcast natural light.

Build from there by adding specificity to each layer as needed.

Pro Tips

1. Start with motion, not appearance. Gen-4 doesn't need you to describe how things look — it needs to know how they move.
2. Use locking language for still cameras. Words like "locked", "static", and "the camera remains still" are far more reliable than "no camera movement."
3. Imply environment reactions before describing them. Adjectives like "dusty", "misty", and "wind-swept" can trigger subtle scene motion naturally.
4. One generation, one scene. Treat each 5-10 second clip as a single shot. Plan multi-clip sequences as separate generations.
5. Iterate in layers. Change one element at a time so you can isolate what's working.
6. Match prompt length to clip complexity. A simple landscape animation might need three lines. A character scene with a specific action might need eight. Don't over-write or under-write.
7. Use "the subject" as your default character reference. It keeps the model focused on motion rather than re-interpreting appearance.

Conclusion

Gen-4 is a powerful tool — and like most powerful tools, it rewards clarity and intentionality. The prompts that work best aren't the longest or most detailed ones; they're the ones that communicate the right things in the right order: what moves, how it moves, how the camera behaves, and what the overall visual language should feel like.

Start simple. Iterate methodically. Focus your text on motion, not appearance. And give the image the visual credit it deserves — your prompt's job is to set things in motion, not to redescribe what's already there.

With that approach, Gen-4 can produce footage that genuinely holds up alongside professional live-action and VFX content.