Runway Prompt Generator
Build a Runway Gen-3/Gen-4 video prompt for text-to-video or image-to-video mode — with the right camera vocabulary for each.
Runway's video models respond well to prompts structured around a clear starting frame plus a described action or transformation, rather than an abstract concept alone. This tool builds a Runway prompt with an explicit starting image description and the motion or change that should happen from there.
How to use it
Be as specific as a photo description — this anchors what the video begins as.
What changes from that starting point — camera motion, subject action, or a transformation effect.
Structured as a starting-frame description plus the motion instruction.
Tips for better results
- Keep the transformation simple and singular. One clear change per generation performs more reliably than several stacked transformations.
- Describe lighting and mood in the starting frame, not the action line. This keeps the two parts of the prompt doing distinct jobs instead of repeating each other.
- Reference specific camera terms Runway recognizes, like dolly or pan, over vague description. Specific camera-movement vocabulary tends to produce more predictable motion than phrases like "moving around."
Example output
“Starting frame: a lit candle on a windowsill at night, rain on the glass behind it. Action: slow dolly in toward the flame as the rain intensifies.”
TL;DR
Runway’s Gen-3 and Gen-4 video models work in two distinct modes that need genuinely different prompts: text-to-video, where you describe an entire scene from nothing, and image-to-video, where you upload a starting frame and only need to describe what changes.
Why image-to-video prompts should describe motion, not appearance
When Runway already has your starting image, re-describing the subject’s static appearance in the prompt wastes the model’s attention on something it can already see, and occasionally introduces small conflicts if your text description doesn’t perfectly match the image. The more effective prompt in this mode focuses entirely on what should happen next — the motion, the camera move, the mood shift — and leaves the visual identity of the subject to the image itself. This generator’s image-to-video mode is built around that principle specifically, dropping the scene field and keeping only motion, camera, and style.
What each field contributes
- Mode changes the entire structure of the output, not just which fields show.
- Scene (text-to-video only) describes the full shot from scratch.
- Motion is the one field both modes share — what changes over the course of the clip.
- Camera movement uses vocabulary Runway’s models specifically recognize — orbit, push in, pull out, crane — rather than vaguer phrasing.
- Visual style sets an overall aesthetic, most useful in text-to-video mode where there’s no reference image to anchor the look.
Text-to-video — Scene: “a red sports car drifting around a mountain curve,” Motion: “dust kicks up behind the tires as it turns,” Camera: orbit right, Style: cinematic. Produces: “a red sports car drifting around a mountain curve, dust kicks up behind the tires as it turns. Camera: orbit right. cinematic style.”
Image-to-video — same motion field, no scene. Produces: “Starting from the uploaded image: dust kicks up behind the tires as it turns. Describe only the motion and change — the image already defines the subject’s static appearance. Camera: orbit right. cinematic style.”
Who this is for
Anyone using Runway’s Gen-3 or Gen-4 models for either text-to-video generation or animating an existing image, who wants a prompt structured correctly for whichever mode they’re actually using rather than one generic template for both.
FAQ
Does this work for both Runway Gen-3 and Gen-4?
Yes, the underlying prompt principles (motion-focused image-to-video, full-scene text-to-video, recognized camera vocabulary) apply across both model generations.
Why does the Scene field disappear when I switch to image-to-video mode?
Because in that mode the starting image already defines the subject and setting — re-describing it in text is unnecessary and can occasionally conflict with what the image shows.
Can I combine a style choice with an uploaded image that already has its own style?
You can, but be aware the model will try to reconcile both — if your image already has a strong look, an added style instruction works best when it complements rather than contradicts it.
Is this useful for Runway’s other tools, like Act-One?
This generator is built specifically around Gen-3/Gen-4 video generation prompting; other Runway tools with different inputs (like performance-capture features) aren’t its focus.
Does this tool store my prompt or image anywhere?
No. It only generates text in your browser — it doesn’t handle or store image uploads at all.
