{ Back to journal }

Tutorial

Writing an image-to-video shot brief: action, camera and constraints

A focused way to brief one image-to-video shot, with a worked example that separates subject action, camera movement and acceptance criteria.

Character wearing earbuds beneath warm outdoor string lights in DELA Media’s Yoki film.

An image-to-video brief works best when one shot has one assessable event. Separate what the subject does from how the camera moves, then define how you will judge the result. This gives you a direction you can test instead of a long list of wishes.

Define the event before writing the prompt

Describe what the viewer should see happen in a single sentence. Avoid beginning with a list of style adjectives. A shot in which a paper sleeve opens to reveal a bottle is easier to assess than a request for an impressive cinematic product experience.

State the purpose of the shot outside the prompt: reveal the dispenser's silhouette, establish a calm atmosphere or connect two scenes. The purpose helps you reject a beautiful result that fails to communicate the required event. Keep that production note separate from instructions intended for the model.

Assign jobs to the image and the words

The starting image establishes what is initially in the frame. Inspect its crop, object placement and any existing defects before asking for movement. If the image already has a bent label, movement is not a substitute for correcting the source.

Runway's Gen-4 guide recommends concentrating its video prompt on motion and using a clear input image. That is guidance for the named model, not a universal syntax for every generator. Check your chosen tool's documentation before applying positive or negative prompt conventions.

Use a short shot brief alongside the input: subject action, camera behaviour, background behaviour and intended end state. That separation makes it easier to tell which instruction caused a change.

Work through an illustrative shot

Suppose the input shows a closed presentation box on a table, with a product already visible through a cut-out. The intended action is a slow lid opening while the viewpoint stays fixed. An illustrative prompt could be:

Illustrative shot prompt — not a tested result

The lid opens slowly on its rear hinge. The camera remains locked. Soft light moves gently across the lid as it opens, revealing the product inside.

This is a writing example, not a tested result or a guarantee of hinge accuracy. If the opening mechanism has to demonstrate a real product, a controlled 3D animation may be more suitable. Do not use the generator's plausible motion as engineering evidence.

Notice that the example does not ask the camera to orbit, push in and change angle while the lid opens. Those requests may describe several creative ideas rather than one assessable shot.

Keep acceptance criteria out of the wish list

Write a review note listing the visible result you need. For the box, it might say: the hinge relationship remains plausible, the product stays in its assigned position, the opening ends at a useful angle, and the final frame leaves space for the next edit.

These are acceptance criteria, not magic phrases that make a model obey. They give the reviewer a common basis for deciding whether to use, revise or abandon a candidate.

When iterating, change one meaningful factor at a time. If you alter the input image, action and camera together, you lose a clear basis for understanding why the output changed.

Choose the next action from the failure

If the camera moves when it should not, simplify the camera description and inspect whether the input suggests an unstable viewpoint. If the intended action is unclear, reduce competing actions. If the object changes identity or mechanism, consider a different input or a more controlled production route.

Record the input, model/version and prompt with the selected candidate. Preserve the rejected result that best explains an important limitation, rather than claiming that the first attempt worked.

Before handing the shot to an editor, check the complete clip and a few frames around its intended cut points. A good opening frame and an attractive thumbnail do not establish that the movement or ending is usable.

More from the journal

Film