Runway image to video camera motion is a separate decision from what the subject does. For a Gen-4.5 shot, choose one main action, describe the intended camera behavior and check whether the starting image supports both. This guide explains how to diagnose a mismatch, budget a few comparisons and decide whether the final frame is suitable for continuation.
This guide covers Runway’s Gen-4.5 workflow, from planning a shot to troubleshooting the result.
Set up one shot before opening the generator
Imagine a photograph of a ceramic cup on a table. The intended video has steam rising while the camera stays still. That is a complete shot brief. Asking simultaneously for a room reveal, a rotating cup, changing daylight and a moving camera creates several independent ways for the result to miss the brief.
Write three short acceptance criteria: the handle stays on the same side, steam is the only subject movement, and the table edge does not shift. These criteria are our suggested review method, not capabilities guaranteed by Runway. They make a failed generation actionable: you can identify which instruction needs work instead of deciding that the whole clip merely looks wrong.
| Planning choice | Write down | Why it matters |
|---|---|---|
| Starting composition | The cup occupies the left third; space remains for a caption | A later crop should not remove the subject or caption area |
| Main motion | Steam rises slowly | A single visible change is easy to evaluate |
| Camera | Fixed position | Subject movement and camera movement remain separate decisions |
| End condition | The cup is still identifiable and fully visible | A beautiful opening does not excuse an unusable ending |
| Delivery | Vertical clip with text added afterward | The output needs to fit the intended edit |
The Gen-4.5 settings that affect your budget
As of October 1, 2026, Runway’s Gen-4.5 creation documentation lists web access on Standard and higher plans, 720p output, two-to-ten-second duration, and 24 or 25 fps. Image-to-video includes portrait, landscape and square framing. The base rate in Credits Mode is 12 credits per generated second.
At that rate, five seconds × 12 credits per second = 60 credits. Three completed five-second alternatives total 180 credits, even if you keep only one. Budget these as generated seconds, not accepted footage. This calculation excludes format surcharges, upscaling and other operations; inspect the displayed charge before submitting.
That arithmetic does not apply to every mode. Runway’s legacy Unlimited documentation says eligible subscribers can use Explore Mode without consuming credits, with variable waiting times and concurrency limits. Check your own subscription and active mode; access to Gen-4.5 alone does not establish Explore entitlement.
Browser and API budgets also differ. The developer pricing page separately lists Gen-4.5 at 12 credits per second. Runway’s API FAQ treats web and API purchases as separate. Its Enterprise documentation describes transfers for enabled, linked organizations, so check that arrangement with your administrator rather than assuming a shared balance. The examples here are browser credit calculations, not dollar quotes.
A useful budget is an experiment allowance rather than a promise of finished seconds. Allocate a small number of attempts to the motion, then reserve another attempt for a specific correction. If all attempts fail because the source image is unsuitable, improve that source instead of consuming the remaining allowance on synonyms.
Prepare the image for the motion you want
Runway’s image-to-video prompting guide treats the uploaded image as the starting frame and the prompt as direction for motion. It also warns that source artifacts can become more apparent during generation. That distinction makes image preparation part of the video task.
Look especially at boundaries that movement will expose. A cup partly hidden behind a plant asks the model to infer its concealed shape if the camera travels sideways. A hand already blurred in the photograph is a weak reference for a precise gesture. Do not rely on a generic request for higher quality to restore those missing details.
For the cup example, choose a source where the handle, rim and base are visible. Leave enough room above it for steam. Use an uncropped source if important parts are missing. If the source implies fast movement, choose another image for a still shot instead of writing several increasingly forceful requests for stillness.
This is a risk-reduction approach, not a guarantee of physical accuracy. The purpose is to remove avoidable ambiguity before paying to discover it in a video.
Generate a baseline, then change one variable
Follow the browser controls in the creation guide:
- Search for Gen-4.5 in Apps View, or select Video in Tool Mode and choose Gen-4.5.
- Drag your image into the prompt window and add the movement instruction.
- Check the aspect ratio and crop, choose duration, and set FPS under Advanced.
- Confirm the cost and mode, then click Generate. Download the completed result using the controls below it.
An original baseline instruction for our example is: “Thin steam rises from the cup. The camera remains fixed. The cup stays on the table throughout the shot.”
Review the first result against the three acceptance criteria. If the camera travels, adjust only the camera instruction or source composition. If the steam is too vigorous, simplify the action wording. Keep the image, duration and other settings stable while evaluating that change. Changing everything at once makes it harder to learn from the result.
| What you see | Next experiment | Avoid doing immediately |
|---|---|---|
| Desired movement, unsuitable framing | Revisit crop and source margins | Rewriting the entire action |
| Attractive scene, wrong movement | Reduce the prompt to the essential action | Adding more decorative adjectives |
| Shape changes near an obscured edge | Use a clearer source image | Assuming resolution alone fixes geometry |
| Action starts but does not finish | Simplify the action or allow more time | Adding another action to the sequence |
The table is an editorial diagnostic sequence. It does not establish that any one adjustment will fix a particular generation.
Continue a clip without hiding the join
Runway documents a specific continuation path: move the playback scrubber to the end, select Use, then Use current frame. That frame becomes the next generation’s image input. Join the completed clips in an editor and remove the shared frame. Inspect the join; this process does not guarantee seamless continuity.
Before extending, ask whether the last frame deserves to become the next source. If the cup has already changed shape, another generation starts from that changed object. Instead, find an earlier usable frame or restart the shot. Preserve the original source and each accepted clip so that an unsuccessful extension does not replace the best version.
For the cup shot, compare the object’s position and shape on either side of the cut. If the join remains distracting, consider a deliberate cut to a different composition.
Finish with an acceptance check, not another prompt
Watch the downloaded clip at normal speed and pause near its beginning, middle and end. Check object boundaries, unwanted changes in the background and whether the final framing leaves space for the intended text. Then place it in the actual delivery format. A clip can look acceptable in a generator preview and still crop poorly in the final layout.
Add precise typography in the editor rather than treating generated lettering as the only copy of a product name. Retain the prompt, model label, duration and credit charge with the chosen file. That small record makes a later revision easier to reproduce and explains where the production budget went.
If your project instead starts with an original AI companion, Cherrypop’s character creator and video workspace are separate workflows worth examining. Cherrypop is free to start with limits; relevant features can require Premium or Cherries. Check the operation cost before generating. These are character-based workflows, with no claim of equivalent Runway controls or output quality.
Runway image to video camera motion: isolate the problem
| Observation | Next controlled comparison |
|---|---|
| The cup is still but the camera drifts | Keep the source and duration fixed while simplifying the camera direction. Compare table edges at the start and end. |
| Steam and camera both move too much | Choose which motion is essential; remove the other request for the next attempt. |
| The shot cuts before the action completes | Reduce the number of requested actions, or compare a longer duration within your credit limit. |
| The final frame has changed the cup | Do not use it as the continuation source; choose a usable frame or revisit the shot. |
Record the specific change and whether it helped.
Practical questions
Should I describe everything already visible in the picture?
Usually the useful addition is what should happen next. Keep details that resolve an ambiguity, but avoid a long inventory that competes with the movement instruction. A prompt for a turning object needs a clear turn; it does not need a paragraph rediscovering the background.
Does a longer clip necessarily solve a failed action?
No. Additional time can help an action fit, but it does not repair a contradictory instruction or an unclear source. Decide whether the failure is timing, movement, framing or source geometry before increasing the duration and its associated credit cost.