Runway: your first 5-second clip (2026)
Runway does not fix a weak frame — it animates it. This flow takes 15 minutes and produces a 5 s clip you can show in a pitch, not a campaign deliverable. You start from a still you already like — Midjourney or a real photo — and describe motion only.
The problem it solves
People open text-to-video, stack the still’s adjectives into the prompt and burn credits on 10 s Gen-4.5 attempts. They get weird hands, impossible physics and a «wow» clip that is not deliverable. The first clip comes from image-to-video: the frame is already in the JPG.
What you will learn
- Start from a still that already works, not text-to-video
- Upload the frame in Generate Video and pick Gen-4 Turbo for tests
- Write a motion prompt: one action + one camera, 5 s
- Generate 2–3 takes and keep the gesture that reads
- Export the keeper without cutting a 60 s film
Before you start
- An account at runwayml.com with credits (the Free plan is enough for 2–3 Turbo takes)
- A still that already works: a Midjourney export or your own photo with a clear frame
- 15 minutes. If you have no still yet, do the Midjourney tutorial first
Steps 1
Start from a still that already works
Open the JPG you already like: a product in three-quarter view, a space with decided light, a figure from behind. If it comes from Midjourney, use the grid keeper, not the «almost». Do not open Runway for text-to-video yet: without an input frame, the model invents composition and burns credits on shots you did not ask for.
What you should see: An image file on disk with clear frame, light and subject. Generate Video tab not open yet.
Steps 2
Open Runway → Generate Video → Gen-4 Turbo
Go to runwayml.com and sign in. From the homepage, choose Generate Video (or Video in the menu). In the model selector, open the Runway group and pick Gen-4 Turbo: it costs less per second (~5 credits/s) than Gen-4 (~12/s) or Gen-4.5. Upload your still as the first frame — drag the JPG or use Upload. Gen-4 requires an input image; Turbo is the right place for your 2–3 test takes.
What you should see: The Generate Video panel with your still as the initial frame, Gen-4 Turbo selected and an empty prompt field.
Steps 3
Write motion, do not re-describe the still
The prompt asks for movement only. One subject action and one camera move. Shape: «The camera [move] as the subject [one action].» Example: «The camera slowly pushes in as steam rises from the coffee tin.» Do not repeat light, style or composition — they already live in the image. Set duration to 5 seconds, not 10 on the first try.
What you should see: A 1–2 sentence motion prompt, no still adjectives. Duration at 5 s visible in the controls.
Steps 4
Generate 2–3 takes and pick the gesture
Hit Generate. Wait in the queue. Repeat with the same action or a tiny tweak (slower, slightly shorter push). Do not change model or duration between test takes. Play all three. Keep the take where the gesture reads — rising steam, a turn you understand — not the shiniest or noisiest one.
What you should see: Two or three ~5 s clips in history. One take mentally marked as keeper for gesture legibility.
Steps 5
If physics or hands fail, change the action
Hands, on-screen type and fine physics still fail in 2026. If steam goes through the tin or fingers melt, do not stack adjectives: change the action («gentle steam wisps» instead of «thick steam cloud») or remove hands from the still. At most, one more take on Gen-4 or Gen-4.5 only if the gesture already worked in Turbo and you want more sharpness.
What you should see: Either a revised prompt with a simpler action, or a fourth take on a higher model with the same motion that already read in Turbo.
Steps 6
Export the keeper and stop there
Download the clip you picked (Export or Download depending on plan). Name it by shot and action (`tin-steam-push5s.mp4`). Do not cut a 60 s film in Runway — this is a prototype. If you need on-screen copy or a headline, that is ChatGPT + an editor (CapCut, Premiere, whatever you use), not this step.
What you should see: An ~5 s MP4 in your downloads folder, ready to drop into a deck or a stories mock.
Real use cases
Pitch prototype
Product still from Midjourney. In Runway: slow push-in + steam. 5 s in the deck. The client argues gesture, not 4K resolution.
Product turn for an internal ad
Real photo of the SKU on a table. Prompt: «The camera arcs 15 degrees as light catches the edge.» Three takes in Turbo, one in Gen-4 if the arc already reads.
Midjourney still to 5 s motion
The keeper from the Midjourney tutorial. One action (leaves moving, a door opening). Export to the moodboard with JPG and MP4 together.
Common mistakes
Starting with text-to-video
Without a still, the model invents the frame and burns credits. Image-to-video first.
Re-describing the still in the prompt
Light, style and composition are already in the JPG. The prompt is motion: action + camera.
10 s and Gen-4.5 on attempt 1
More duration and a pricey model do not fix a bad gesture. 5 s in Turbo to learn.
Treating the clip as a campaign film
Runway prototypes motion. The final deliverable goes through edit, copy and color elsewhere.
Close
Your first useful Runway clip comes from a directed still and a short motion prompt. Gen-4 Turbo at 5 s teaches which gestures read before you spend on expensive models. If the base frame is weak, go back to Midjourney — Runway will not save it.
Next steps
- If the still does not exist yet, follow the Midjourney tutorial and come back with a keeper.
- If you need the ad headline, open ChatGPT with the still as visual context.
- To compare when Runway vs Midjourney, see the catalog comparison.
Free guide
Want the 15-tools guide?
We send it free. We will also ping you when the next guide in this path goes out.
Unsubscribe whenever you want. We do not sell the list.