How to Write an AI Video Shot Brief for Your Team

Before generating an AI video shot, write down what the audience must understand, what moves, how the shot connects to its neighbours and what would make a take usable. Put those decisions in a short shared brief. Then translate the relevant parts into the model's prompt and controls.

A shot brief gives a teammate enough information to produce material the editor can use. It also gives the team a reason to accept a take, revise one specific problem or change the plan. The template below is an editorial workflow for collaborative filmmaking, with an invented example; it adds no rules to a Prompt2Screen challenge.

Give each shot one job in the story

Start with an audience-facing purpose: “We realise the driver has recognised the passenger” is more useful than “cinematic close-up.” The first statement explains why the shot belongs in the film. The second describes a possible treatment.

Name the information that must reach the viewer. It might be a reaction, a location, a change in an object or a pause that creates uncertainty. If a shot has several jobs, decide which one matters most. That priority will help when a generated take achieves some of your intentions but misses others.

Test the purpose in a rough edit first. Our animatic guide explains how to use temporary frames to check whether the sequence communicates before making its finished images.

Use a compact brief that another person can act on

Here is a fictional exercise about a driver finding an old photograph in a parked car. It is not a description of a Torino competition entry.

• Shot ID and owner: S04; one named teammate prepares the take and sends it to the editor.
• Story purpose: the viewer notices that the photograph matters before seeing whose face is in it.
• Starting view: medium close view from the passenger side; driver seated on the left, photograph held below the steering wheel.
• Action: the driver lifts the photograph, stops and holds it in view.
• Camera: fixed position; the movement comes from the driver.
• Continuity anchors: same coat, seat fabric and direction of daylight as the preceding shot.
• Edit target: roughly three seconds of usable action, with a short pause before and after it.
• Acceptance check: the lift and hesitation read clearly; the photograph remains one object; the driver's position stays legible.
• Fallback: a separate insert of the photograph followed by a reaction shot.

The edit target describes the material you want in the sequence. It does not prescribe an available generation duration. Choose settings supported by the tool and account you are actually using, allowing room to trim where practical.

Keep the production brief separate from the prompt

The brief contains information for people: ownership, story function, references, delivery and acceptance. The model prompt needs the instructions relevant to generating the image or movement. Pasting every production note into the prompt can obscure the action you want.

For an approved starting image in the fictional exercise, a first motion prompt might be: “The seated driver slowly raises the photograph, pauses and holds it steady. The camera remains still.” This is an original example to test and adapt, not a guaranteed recipe.

Runway's image-to-video prompting guide explains that the input image establishes the visual scene, while text directs movement and its progression. It recommends beginning with the essential motion and refining the instruction. That is guidance for the workflow described by Runway; check the documentation for the particular model you choose.

Keep the acceptance check beside the generated results. A clip can follow the prompt yet still fail its story purpose: the photograph may rise beautifully without the hesitation being readable.

Check the starting image before requesting movement

Inspect the reference at the size you will use in the edit. Confirm that the pose can lead into the action, the important object is visible and the framing leaves room for the movement. Repair or replace a reference that already contradicts the brief.

In the example, a photograph hidden behind the wheel gives the generator a different problem from a photograph already held clearly in the driver's hand. Decide which starting state serves the shot, then supply it consistently to the person generating the take.

For sequences where the location itself carries the story, the Milan AI Festival guide to consistent architectural spaces offers a more detailed spatial workflow. Use that when a doorway, room layout or route needs to remain understandable across several views.

Review the take in its actual cut

Place the candidate between the preceding and following shots before judging it. Watch at normal speed. Does the viewer notice the intended information? Does the movement give the editor a clear place to enter and leave? Do the surrounding shots make the reaction understandable?

Review image defects separately from story clarity. A small background inconsistency may be less disruptive than an action that reads as dropping the photograph instead of recognising it. Name the issue precisely so the next attempt has a defined purpose.

Change one major variable per deliberate test where possible. For example, keep the image and camera instruction while simplifying the hand action. Record what changed and what happened; a different result alone does not establish that you have found a reliable fix.

Agree when to use the fallback

Before repeated attempts consume the remaining production time, agree on a review point. If the shot still fails its essential acceptance check then, try the fallback in the edit. An insert and reaction may communicate the same beat with a different visual rhythm.

Save the approved take under its shot ID, mark the chosen version and tell the editor which section is usable. Keep the brief and decision note with the project; our asset organisation guide covers the wider handoff.

A useful brief ends in a decision: this take communicates the beat, this revision addresses a named problem, or this alternative serves the sequence better. That makes generation part of directing and editing a film, with each attempt connected to a purpose.

Sources & further reading

• Runway — Image to Video Prompting Guide ↗
• Milan AI Festival — How to make an AI architecture film with consistent spaces ↗

Updated 3 November 2026. Event availability and final details are maintained on the linked official event pages.