Runway Gen-4 Image to Video: Settings and Prompts

Runway's Gen-4 turns a single still image into a 5 or 10 second clip, keeping the look of your frame while adding motion you describe in a text prompt. Image-to-video is the most controllable way to work in Runway because you decide the composition, lighting and subject before any motion is generated. This guide covers the model choice, the settings that matter and how to write a prompt that describes movement rather than restating the picture.

Which Runway model to pick

Runway currently offers Gen-3 Alpha, Gen-3 Alpha Turbo, Gen-4 and Gen-4 Turbo in its video tool. Gen-4 is the newer family and generally holds a start image together better, with more natural motion and fewer melting faces. Gen-4 Turbo is the faster variant, which makes it the right choice while you are still testing whether an image wants to move at all. Gen-3 Alpha is still worth knowing about because it accepted a last frame as well as a first frame, so you could define where a shot ends; Gen-4 works from a single start image.

The model selector sits near the prompt box in the current interface, though its exact position changes from time to time. Because Runway bills credits per second of video, a practical workflow is to draft in Gen-4 Turbo at 5 seconds, then re-run the version you like in Gen-4 at the final length. Check Runway's current pricing page for what each model costs per second.

Preparing the start image

The start image does most of the work, so it is worth spending more time on it than on the prompt. A few habits make a large difference:

• Match the aspect ratio first. Generate or crop the image to the ratio you will export in. If the image and the chosen ratio disagree, Runway has to crop or pad, and you lose control of the framing.
• Leave room for movement. A subject pressed against the frame edge has nowhere to go. Give a walking figure space in front of them and a panning shot something to reveal.
• Keep the lighting simple. One clear light direction animates more convincingly than a scene with mixed, competing sources.
• Avoid tiny text and fine patterns. Lettering, lace and chain-link fencing tend to shimmer once they move.
• Upload at a sensible size. A crisp image at roughly the output resolution is enough; an enormous file is downscaled anyway.

If your still came from another image model, you can keep the whole pipeline in one place with a consistent character. See Gen-4 References for consistent characters for how to carry a face from stills into video.

Settings at a glance

The fixed-seed option deserves a word. With the seed locked, two generations that differ only in the prompt will share their underlying randomness, so you can see what a wording change actually did. With it unlocked, you cannot tell whether a better result came from your edit or from luck.

Writing a motion prompt

The commonest mistake is to describe the picture again. Runway can already see the woman in the red coat on the pier; what it needs to know is what happens next. A reliable structure is three short clauses:

1. Subject motion. One verb for the main subject: she turns to look over her shoulder; the dog shakes water from its coat; steam rises from the cup.
2. Camera motion. Slow push in, handheld drift, static tripod, gentle pan left. Say "static camera" if you want none; otherwise the model may add drift.
3. Pace and atmosphere. Slow and calm, quick and jittery, wind moving the hair, light flickering.

Keep it to one or two sentences. Long prompts with several actions produce clips that try to do everything and finish nothing within 5 seconds. If you want a sequence, generate separate clips and cut them together. Negative wording ("no camera shake") is less reliable than positive wording ("locked-off camera"), so phrase the thing you want.

For wider context on when image-to-video in Runway is the right choice compared with other tools, see which platform for what.

Try it

She slowly turns her head towards the camera and a faint smile appears. Gentle wind lifts strands of hair. Slow push in, shallow depth of field, calm and quiet.

Swap the first sentence for whatever your subject should do, keep the camera clause, and change the final clause to set the mood. If the face drifts, remove the head turn and let only the wind and camera move; small motions are the ones Gen-4 does best.

Fixing common problems

• The subject morphs. Shorten the clip to 5 seconds, ask for smaller movement and keep the camera still. Large motion on a 10 second clip is where identity slips.
• Nothing moves. Add a specific verb and a camera move. "Cinematic, beautiful" tells the model nothing about motion.
• The camera does something you did not ask for. State the camera explicitly every time, even when you want it static.
• Background crowds behave oddly. Prefer start images with few background people, or ask for them to be out of focus.
• Results vary wildly between runs. Turn on the fixed seed and change one thing at a time.

Keep a note of the model, duration, seed and prompt for every clip you like. When a client asks for "the same but longer", you will be able to reproduce the setup instead of guessing.