Runway Prompts: How to Write Them for Gen-4.5, With Examples

What Runway Gen-4.5 is and which version is current

Runway is an AI video platform, and Gen-4.5 is its own video-generation model, released on 1 December 2025 and available on paid plans. It has two modes: text-to-video, where you only describe the scene, and image-to-video, where you start from an image and the text says how it moves. For editing existing footage Runway has a separate model, Aleph 2.0. Checked in November 2026.

Runway's platform, and its API in particular, also gives access to models from other companies, such as Kling 4.0, Seedance, Veo, Wan 3.0 and MiniMax H3. This guide covers Gen-4.5 only. If you pick another model inside Runway, follow that model's own guide, because each one has its own syntax.

Clip length, available aspect ratios and whether Gen-4.5 generates sound aren't confirmed in the documentation we could review, and they depend on your plan anyway. Check what your account allows in the tool itself before you plan a video.

Rule number one: keep everything positive

Runway's prompting guide is explicit: negative phrasing isn't supported and may produce unpredictable or even opposite results. Typing 'no people' can fill the street with people. As far as we can tell there's no separate negative-prompt field either, so the fix is to describe what you do want to see.

• 'No people' → 'an empty street at dawn'.
• 'No text or signs' → 'clean, plain walls'.
• 'Don't move the camera' → 'a locked-off shot'.
• 'No shaky footage' → 'smooth, steady camera movement'.
• 'No bright colours' → 'a muted palette of greys and soft blues'.

Before you generate, scan the prompt for 'no', 'without', 'don't' and 'never'. Each one marks a sentence worth rewriting.

How to structure a Gen-4.5 prompt

Gen-4.5 handles plain film-set language well. A reliable order is shot and camera first, then subject and action, then setting, lighting and style, all in short sentences that describe what's on screen.

• Shot and camera: 'A locked-off medium shot…', 'The camera dollies in…'.
• Subject: who or what is in frame, with two or three concrete traits. 'A ceramicist with rolled-up sleeves'.
• Action: one action, in the present tense. 'She shapes a tall vase on a spinning wheel'.
• Setting: where it happens. 'A sunlit pottery studio with bare plaster walls'.
• Lighting and style: 'soft daylight from a side window, realistic cinematic look'.

A locked-off medium shot of a ceramicist with rolled-up sleeves and clay-streaked forearms shaping a tall vase on a spinning potter's wheel. Her hands rise slowly and the walls of the vase grow taller and thinner. The studio behind her is empty and quiet, with bare plaster walls. Soft daylight from a large side window, fine dust drifting in the light. Realistic cinematic look, shallow depth of field.

Camera and movement in Runway

Runway recognises the usual camera styles: locked, handheld, dolly, pan, subject tracking and focus shifts. For a completely still camera, the term its guide uses is 'locked-off'.

• Still: 'A locked-off wide shot of…'.
• Handheld: 'Handheld camera tracks the cyclist through a crowded market'.
• Dolly: 'The camera dollies in as the woman turns toward the window'.
• Pan: 'The camera pans slowly from the harbour to the lighthouse'.
• Focus shift: 'Focus shifts from the raindrops on the glass to the street behind'.

You can combine complex moves, but the bigger the move, the more the model has to invent. Help it by saying what comes into view in each phase: first this, then that, finally the other.

The camera starts on a close-up of a brass compass resting on a map, then pulls back and rises. First the wooden desk and an oil lamp come into view, then the round window of a ship's cabin, and finally the stormy grey sea outside. Warm lamplight inside, cold blue light from the window. Realistic cinematic look.

The camera tracks alongside a cyclist in a green rain jacket as she rides along a coastal road at golden hour. The sea glitters on her left and tall grass bends in the wind on her right. Warm low sunlight, long shadows across the asphalt, cinematic look.

Image-to-video: describe the motion, not the picture

When you start from an image, the subject's look, the colours and the set are already decided. Repeating them in the text doesn't help. What the model needs is what moves, how, in which direction and at what pace, and what the camera does.

The camera dollies back slowly from the lantern on the table. First the wooden walls of the cabin come into view, then the open front door, and finally the snowy pine forest outside. A gust of wind blows in through the doorway, and then the lantern flame bends and flickers. Snowflakes drift past the door frame.

Gen-4.5's known limitations, turned into prompt advice

Runway lists three weak spots for the model. Knowing them saves a lot of retries, because you can work around each one in how you write.

• Cause and effect: the effect sometimes shows up before the cause (the glass shatters before it falls). Put the cause in one sentence and the effect in a later one, joined with 'then' or 'after', and leave a beat between them.
• Object permanence: something that gets covered or leaves the frame may change or vanish. Describe it with the same traits every time you mention it and say where it stays ('the red umbrella stays leaning against the bench').
• Success bias: actions tend to succeed even when they should fail. If someone has to trip or miss a jump, describe the failure as a physical event that does happen ('his foot slips on the edge and he lands short') rather than 'he fails'.

A locked-off close-up of a wooden table in a sunny kitchen. A child's hand pushes a glass of orange juice toward the edge. The glass tips over the edge and falls, and then it lands on the tiled floor and the juice splashes outward in a bright arc. Natural morning light, realistic look.

Handheld camera follows a young skateboarder in a yellow hoodie as he rolls toward a low concrete ledge in an empty skate park. He pops the board, his back wheel catches the edge of the ledge and he stumbles forward, landing on his hands on the smooth concrete. The board rolls away on its own. Overcast soft light, documentary look.

Common mistakes

• Writing what you don't want ('no text', 'without people') instead of what you do want to see.
• Saying 'static' or 'the camera doesn't move' when the guide suggests 'locked-off'.
• Asking for a big camera move without describing what it reveals.
• In image-to-video, describing the image again instead of the motion.
• Packing several actions into one short clip, or leaning on empty adjectives like 'epic' or 'stunning'.
• Assuming a specific length, aspect ratio or audio track without checking your plan.
• Asking for the look or voice of real people, copyrighted characters or brands.

Using the PromptNow Runway prompt generator

PromptNow has a Runway Gen-4.5 prompt generator. You pick the mode (text or image to video), describe the scene, the type of action and the camera, and note anything you want to avoid. You get a prompt to paste into ChatGPT, Claude or Gemini, which returns the final Runway text written in positive terms only, with two variants, a breakdown of each part and a list of what to check in your plan.

If you haven't settled on a model yet, the general scene template and the image-to-video template work with any generator.