How to brief an AI avatar video, from script to export

How we checked what a brief has to carry

A video made with an avatar is specified rather than edited. The tool does the camera work, the lighting and the continuity, and the one thing it cannot do is guess, so every decision it has taken over has to be written down before the render starts. That is why a brief for this kind of video is longer than the two paragraphs a filmed shoot usually begins with, and why the extra length is exactly the list of choices the software made automatic.

Six of those choices come up in every project of this kind, which is what the parts below are built from. Five are answerable from a vendor’s plan table alone; the sixth is a judgement about the result rather than a line item, and the last section says what to do about it. The parts are ordered the way the work happens, because a brief written out of order tends to be rewritten at the render.

No avatar was built and no clip was exported for this page. What follows is read from the product documentation and plan tables the companies publish, with the check date at the end, and the figures quoted in the linked reviews come from those same pages.

Six parts of a brief

Part 1: What the video is for

This line decides the tool rather than the render, so it is written first. A clip built to sell something is judged on how quickly it earns attention; a clip built to teach something is judged on whether somebody can follow it twice without asking a question. The two want different pacing, different lengths and often different products, and a brief that leaves this line out reads as generic somewhere in the middle of the video.

Write it as one sentence naming the audience and what they should be able to do afterwards, and every later part of the brief can be checked against it.

Part 2: The script, and what the avatar is asked to say

For this kind of video the script is the shot list. There is no coverage to choose from later, so the words carry the pacing, and the tool renders what is written, including the sentences you would have cut in an edit.

Two practical limits sit on this line. The first is the scene ceiling: one vendor documents a project as up to 150 scenes of five minutes each, capped at four hours in total, with the limit landing when either count is reached. The second is pronunciation. Names, acronyms and product terms are where a synthetic presenter gives itself away, and the fix belongs in the script rather than in a re-render, so spell the awkward ones the way they should be said before the first export.

Part 3: Who appears, and whose likeness it is

This is the part a filmed shoot does not have, and the part with the widest spread between plans. Four routes exist across the tools here: a stock presenter picked from a library, a still photograph animated into a talking head, a personal avatar built from a short clip of a real person, and a bespoke presenter commissioned above every plan as an add-on.

Library sizes are published and differ by tier, and so do the personal-avatar allowances: one tool includes three personal avatars on its first paid tier and five on the next, while another makes photo avatars and voice clones unlimited as soon as you leave its free plan. Whose face it is also matters beyond the plan, because a personal avatar needs the consent of the person it depicts, and a bespoke one is a purchase with a published annual price rather than a tier.

Part 4: The language, and whether one recording serves all of them

Where the video has to run in more than one market, this line decides whether that costs a second production or a second render. The widest published coverage here starts at 160 languages and is available from the free plan upward, with dubbing documented from around 70 languages on an entry tier to more than double that at the top. A second tool opens a much longer list the moment you pay, and a third sits between the two.

Write the brief in the language the script is strongest in. A dubbing flow re-voices a finished video rather than rebuilding it, so translation becomes a step at the end instead of a second script, and the timing of the original survives it.

Part 5: Who signs it off, and what the tool enforces

This is the line that separates a video made by a person from a video made by a company, and it has to be written before the render because the controls behind it live at the top of the published range. Brand enforcement keeps unapproved fonts and colours out of a finished clip, an admin role holds it, folder permissions decide who may view, comment on or edit what, and an approval loop lets one person accept or decline an edit before it goes out.

One vendor documents that machinery in detail and places it at the top of its range, while another bundles a brand kit and an approvals workflow into its highest published tier and charges per seat on top. Below those tiers an account carries one editor, which is worth writing into the brief as a constraint: a sign-off step needs two logins before it can exist at all.

Part 6: The file that leaves

The last line is the deliverable, and it is the one most often left implicit. Write down the resolution, whether a watermark is acceptable, and where the file is going. One tool in this group publishes 1920×1080 as the ceiling on every paid plan, so climbing its range does not buy a sharper picture; another keeps 4K for the tiers above its entry level; and the free plans either watermark their output or, in one case, stop short of an export entirely.

Where the video lands in a training system rather than a feed, the format is worth naming: one of these tools exports a package that drops into an LMS rather than a plain video file, and that single line can decide the purchase by itself.

What the brief cannot decide for you

Six parts settle what the tool is asked to produce. They do not settle whether the result convinces, and it is worth being clear about which of the two the brief is for.

Sample footage belongs to the vendor

The clips that show these avatars off are the frames the company chose to publish, so they demonstrate what the tool does at its best rather than what it does on an average Tuesday. A brief that asks for a person walking while talking, a gesture timed to a particular word, or an unusual name at speed is asking for the cases the samples avoid, and the free plan is where to test one before committing the budget.

One brief is not a house style

A brief is written for a single video. What keeps a series consistent is not the brief but the account behind it: the same avatar, the same voice, the same brand kit and the same approval step carried from one clip to the next. That is a decision about which tier to buy rather than about what to write, and it is the reason the parts above keep arriving at the top of the published range.

Sources

HeyGen (heygen.com), checked 16 September 2026: the plan comparison and help centre for the avatar types, the personal and photo avatar allowances per tier, the video length limits, export resolutions and the voice options included.

Synthesia (synthesia.io), checked 16 September 2026: the documentation for the scene and runtime limits, the pricing page for the personal-avatar allowance and the studio-avatar add-on, and the help centre for brand kits, permissions, the approval role and SCORM export.

Creatify (creatify.ai), checked 16 September 2026: the pricing page for the actor library, the custom-avatar allowance and the video length cap each plan carries.