Tutorials#Spokesperson video creation

How to Create Avatar Ads Without Filming (Step by Step)

A step-by-step production run for avatar ads with no camera: what to prepare, what to generate, what to fix by hand, and what to check before spend.

Hero image for How to Create Avatar Ads Without Filming
Rando TkatsenkoAuthorRando TkatsenkoMarch 17, 2026Updated September 20, 20266 min read

Before you start: what you need in front of you

This is the production run, not the strategy. For when avatar ads are the right call at all, which platforms suit them and how they compare to filmed talent, start with AI avatar videos and avatar ads.

Have four things ready before you open anything:

  • A presenter image you are allowed to use. A stock presenter from your tool, or a photo of someone who has given documented permission for this specific use.
  • An approved script, 15–30 seconds, hook in the first sentence.
  • A brand kit: logo, colours, caption style, and any legally required copy.
  • The placement list — which ratios and which platforms this ad has to exist in.

Skipping the fourth is the most common cause of rework later.

The run, start to finish

1. Cut the script to spoken length. Read it aloud and time it. Written copy that looks like 20 seconds is usually 30 when spoken. Trim from the middle, never the hook.

2. Mark the hook visually. Decide now what text appears on screen in the first second, because a large share of viewers will read it before they hear anything.

3. Generate the presenter pass. Script plus presenter image, or upload your own narration if you need a specific voice. Generate two or three voice and tone options rather than one — they cost almost nothing at this stage and voice is frequently what changes the result.

4. Listen before you look. Play the audio with your eyes closed. Wrong emphasis, rushed numbers and mispronounced product names are obvious this way and easy to miss when you are watching the picture.

5. Caption, then correct. Auto-generate captions, then fix them by hand — brand names and product terms are exactly what automatic captioning gets wrong, and a misspelled product name in your own ad is worse than no captions.

6. Add the hook overlay and brand frame. On-screen text for the first beat, logo placement, and any required legal line.

7. Balance the audio. Narration level consistent, music well under it, no clipping. Sound-off viewers will not notice; the ones with sound on will notice immediately if it is wrong.

8. Check every ratio inside the project. Square, vertical, landscape. Look specifically at whether captions and hook text clip at the edges. Fixing this now costs a minute; fixing it after approval costs a rebuild.

9. Export the set together. All ratios, all voice variants, named so you can tell them apart in the ad account three weeks from now.

What to check before you put spend behind it

A short pre-flight list, in the order things actually go wrong:

  • Does the first second work with no sound?
  • Is every claim in the script on your approved list?
  • Do the captions match the audio word for word?
  • Does the vertical cut clip any text?
  • Is the presenter's usage permission documented?
  • Does this placement require synthetic-media disclosure?
  • Is the file named so a colleague could identify the variant without opening it?

Where this usually goes wrong

  • Generating before the script is approved. Every review comment after generation costs a re-render; before it, it costs a keystroke.
  • One voice, one take. The cheapest variable to test, routinely skipped.
  • Treating captions as accessibility rather than as the hook. They are the hook for the muted majority.
  • Producing the vertical cut as a separate job. It is the same project; treat it that way.
  • No naming convention. Six variants named "final_v2" is a problem you create for yourself two weeks later.

FAQ

Q: Do I need a real person's photo at all? A: No — stock presenters work. If you do use a real person, get documented permission for this specific commercial use.

Q: How long should an avatar ad be? A: Most paid-social avatar ads land between 15 and 30 seconds. The constraint is the script, not the format.

Q: Can I reuse one presenter across a whole campaign? A: Yes, and you generally should. Consistency across a variant set is what makes the ads read as reinforcement rather than noise.

Q: What is the minimum viable finishing pass? A: Captions, an on-screen hook, and a level audio mix. Below that you are testing whether your audience tolerates rough creative, not whether your message works.

Q: How many variants should I produce in one run? A: Enough to test one hypothesis — usually five to ten, produced together and retired together.

Next

Choosing a tool for this: best AI avatar video tools. The wider strategy, platforms and use cases: AI avatar videos and avatar ads.

Producing them: Avatar video ads with Shorz