You need three seconds of a fictional workshop at dawn to open an explainer. You do not need a thirty-second sequence with a moving camera, five characters, and a complete story.
Runway added WAN 3.0 to its API on August 26. Its changelog describes reference-driven video, first and last keyframes, and native audio, with output up to thirty seconds. Those controls create options, but an available duration is not a target for every shot. Runway API changelog
For creators making illustrative footage, the useful unit is a shot with a defined role in the edit. The process below is a planning approach, not a report of measured model performance.
Describe the sentence the footage supports
Begin with the narration. Suppose the line is: “The idea started as a rough sketch on a workbench.”
The accompanying image should help the viewer understand that sentence. A hand setting down a sketch may be enough. A sweeping tour of a futuristic factory introduces information the story does not require.
Write four things before generating: what the viewer should notice, the action, the approximate usable length, and how the shot enters and exits the edit. This makes a result easier to accept or reject.
Keep documentary evidence distinct from illustration. If the video describes a real maker's workshop, use authentic footage for claims about that place. A generated scene can support a clearly illustrative passage; it should not masquerade as a record of something that happened.
Build references for continuity
For a fictional sequence, define the recurring objects before requesting motion. Choose the sketch, the workbench surface, the lighting direction, and any visible clothing.
Use references you have permission to use. Keep a record of where each came from and why it belongs in the project.
Runway's general reference guidance recommends clear subjects, good source quality, and coherent style references. Those are useful preparation principles, though supported inputs and limits still depend on the specific model and endpoint. Runway reference-media guidelines
Create a small continuity sheet with the reference images and a short description. When reviewing two shots, compare them with that sheet rather than relying on memory.
Test one motion before adding complexity
Start with the simplest action that communicates the idea. In the workbench example, the camera can remain still while the sketch enters the frame.
Review the movement itself. Does the hand keep its shape? Does the paper bend plausibly? Does the object remain the same when it passes behind another object? Does the ending give you a clean place to cut?
If the result fails, change one part of the request. Simplify the hand interaction, reduce the camera movement, or choose a shot where the paper is already on the desk. An editorial adjustment may be faster than repeatedly asking for an increasingly complicated generation.
Keep rejected attempts in a review folder long enough to record why they failed. “Wrong object” and “unusable motion” point to different fixes.
Judge the clip inside the sequence
Place the candidate between its neighboring shots, with the actual narration and music. Check color, scale, direction of movement, and the viewer's attention at the cut.
A clip that is attractive alone can feel unrelated in context. If the surrounding footage is quiet and observational, a dramatic simulated crane move may break the visual language.
Listen to any generated audio separately. Decide whether it contributes to the scene, conflicts with the existing sound, or should be replaced. Do not assume the soundtrack is approved because the picture passed.
Record the usable seconds, the total attempts, and the time spent repairing the shot. That gives you a practical cost for this kind of B-roll.
For the next video, repeat the shot types that worked. Keep difficult interactions for scenes that truly need them. The goal is a dependable way to fill an illustrative gap while preserving the meaning and rhythm of your edit.