How to write a video brief: parts, template, agent tips

To write a video brief, state on one page what the video must be (length, shape, platform), who it is for, the one message it carries, what must appear in it, the tone, and how you will judge it done. Leave the shot-by-shot decisions to whoever makes it, whether that is an editor, an agency, or an AI video agent.

The template below is editorial advice, not a schema. The notes on handing a brief to an agent come from Sume's Create a run, Format API, and Agent Completions docs, read on 2026-09-28.

What should a video brief include?

A brief answers the questions a maker would otherwise ask you halfway through. Six parts cover most videos:

• Deliverable: running time, aspect ratio (9:16, 1:1, or 16:9), where it will be posted, and how many versions.
• Audience: who watches, what they already know, and why they would stop scrolling.
• One message: a single sentence the viewer should remember, plus the next step you want them to take.
• Must-haves: the product, logo, lines of copy, prices, or legal text that has to appear, and the assets you are supplying.
• Tone and references: two or three adjectives and links to videos that feel right. Say what to avoid, too.
• Done means: who approves it, what gets checked (spelling, claims, music rights), and the deadline or budget.

Is there a video creative brief template?

Copy this and fill each line. Keep it short enough that the maker reads all of it; if a section grows past a few lines, it is probably a script or a shot list, which belong in their own documents. A shot list for AI video is the next step once the brief is agreed.

VIDEO BRIEF
Deliverable:  30 s, 9:16, Instagram Reels and TikTok, 1 version
Audience:     first-time buyers comparing running shoes
Message:      "Light enough to forget you're wearing them."
Next step:    visit the product page
Must show:    the shoe from the side, the logo end card, price $89
Assets:       3 product photos, logo PNG, brand font name
Tone:         upbeat, bright daylight, no stock-office look
Avoid:        slow motion, medical claims
Captions:     burned in, English
Done means:   copy and price checked; approved by the brand lead

How do I give a brief to an AI video agent?

An agent takes the same brief, but its request has separate fields, and each part of the brief fits one of them. On Sume the same split applies to an Agent Completion (POST /v1/agent/completions) and a Format run (POST /v1/formats/{handle}/{slug}/runs). The Format docs describe the call as carrying "the what: product URL, brief, script, prices". Video agent API: brief to finished video shows a minimal request.

How long can a brief be?

Long enough for the facts, but the instruction itself should stay short. On a Format run, instruction accepts 8,000 characters and only the first ~4,000 reach the prompt. input accepts 2 MiB and is carried whole, as a file the agent reads, never truncated. So put the one-paragraph ask in the instruction and the long material (product copy, a script, a list of claims) in input.

• Media URLs inside input share the run's budget with attachments: 30 files in total, at most 30 images, 10 videos, and 10 audio files.
• An attached image over 30 MB, or a set over 500 MB, is refused with 413 attachment_too_large.
• The only attachment type is input_image; the Format docs say to send documents by URL in input instead.

Why cap the spend in the brief?

A person you brief will ask before going over budget. An agent called from a backend has no one to ask. Sume's docs say the spend cap stands in for the chat's spend-approval prompt, and an Agent Completion without generation_spend_cap_usd fails with 400 invalid_request. On a Format run, the cap can go up to the platform maximum of $500.

A brief does not guarantee the agent follows every line, so write the must-haves as checks you will make on the result. For ads, How to make a video ad with AI adds the draft-then-approve step, and How to write an AI video prompt covers the single-shot prompt a brief eventually turns into.