How to make ten versions of one ad with a different setting in each

Last updated 2026-08-31 · 8 min read

What is the short version?

Paid social rewards volume of creative, not volume of recuts. Ten versions of one video with different captions is one creative; ten different people saying different things in different rooms is ten, and that is what a test needs to find a winner.

1. Go to Vizard Agent and give it the product and the claims you may make.
2. List the settings you want — kitchen, coffee shop, car, park, library.
3. Say how many variations, and whether you also need a shorter cut.

What do you need before you start?

The product, the list of settings, and above all the boundary of what you are permitted to claim about it. Vizard Agent writes ten different scripts from that brief, and ten separate chances to overstate something is exactly ten times the risk that one script carries.

• The product page. For features and positioning.
• The claim boundary. What must be said, what must not.
• The settings list. One per variation.
• The voice. Pinned, so every version matches.
• The cuts. Thirty seconds plus a fifteen.

What do you type into Vizard Agent?

Give the compliance rule as an explicit instruction, not as background. Vizard Agent writes the scripts, so the sentence about what may not be claimed does more work here than any amount of description about tone or audience.

Prompt

Variants worth knowing:

• "Distinct, not rewordings." Otherwise they converge.
• Pin the voice by name. Ten voices reads as ten products.
• Ask for the short cut too. Built from the same timeline.

What does Vizard Agent actually do?

Here is the order Vizard Agent worked in on a real ten-variation campaign. Notice that the variations are generated in parallel at step nine but assembled from one timeline at the end — the settings differ, the structure does not.

1. Downloads every supplied image and builds an overview sheet of all of them.
2. Reviews the scenes in groups, close up, including the end card and creator references.
3. Checks the avatar, voiceover and music tools and casts the voice on a first line.
4. Generates all voiceover lines with the pinned voice and fetches a music bed.
5. Trims silence and gently tightens the tempo of the recorded lines.
6. Animates the creator so she speaks her lines, then checks the frames.
7. Renders the timeline segments, assembles the base cut and mixes the audio.
8. Burns in the captions and produces the first full versions for review.
9. Generates the ten creator looks and two over-shoulder plates in parallel.
10. Composites real product screenshots onto the phone screens in the over-shoulder plates.
11. Re-composites with full coverage and de-spill after checking the first attempt.
12. Renders the punch-in segments and the full-screen product beats.
13. Cuts the avatar segments, mixes, writes captions and renders all eleven videos.
14. Builds the fifteen-second cut, captioned and clean, and extracts review frames from every version.

Step ten is the detail that separates these from stock ads. Vizard Agent composites your actual product screenshots onto the phone in the shot, so the thing being recommended is visibly the thing you sell.

What does the result look like?

From the run this page is written from: eleven 30-second vertical videos at 1080x1920, one per setting, plus a 15-second cut in captioned and clean versions. One voice across all of them; animated word-highlight captions in the brand's green.

Each variation opens differently and makes its case differently, which is the entire point. A test that launches with ten genuinely different hooks tells you something after a day; ten recuts of one hook tells you nothing.

When does this not work well?

Generating creative at volume multiplies whatever is already wrong with the brief, and the compliance risk scales with the count. Vizard Agent applies your stated claim boundary to every single script it writes, but the boundary itself has to be correct before any of that helps.

• A vague claim boundary. Ten scripts, ten interpretations.
• Near-identical settings. They stop being distinct creatives.
• Generated people. Disclose them where the platform requires it.
• Avatar generation fails. Some segments need retrying.
• Volume is not a strategy. Ten weak hooks stay weak.

How do you fix a result that came back wrong?

Name the variation by number and say what is wrong with it. Vizard Agent keeps every script, the recorded voice lines, the generated looks and the shared timeline, so one version can be rewritten or re-rendered on its own without the other nine being touched.

• "Variation four overclaims." Rewritten and re-recorded alone.
• "That setting looks wrong." Regenerated from the same script.
• "Captions collide with the logo." Repositioned across all versions.

How does Vizard Agent compare to doing it yourself?

By hand, ten genuinely different creatives means ten scripts, ten shoots or ten stock licences, and ten edits. Most teams give up and export ten caption variants instead, which is why so many tests measure nothing. Vizard Agent builds ten and reuses one timeline.

Common questions

How many variations are useful? Ten is a reasonable test. Vizard Agent produces them from one brief rather than one at a time.

Will the scripts actually differ? If you ask for it explicitly. Vizard Agent varies the hook, the angle and the phrasing rather than the wording alone.

Should the voice change per version? Usually not. Vizard Agent pins one voice so the versions read as one product.

Can it use my product screenshots? Yes. Vizard Agent composites them onto the phone screens in the over-shoulder shots.

What about advertising compliance? State the boundary and it holds. Vizard Agent applies your claim rules to every script it writes.

Do I need to disclose generated people? Check your platform's rules. Vizard Agent builds the ads; the disclosure obligation stays with you.

Can I get a shorter cut? Yes. Vizard Agent builds the fifteen from the same timeline, captioned and clean.

What if one avatar segment fails? It retries. Vizard Agent re-runs failed segments in smaller batches rather than abandoning the version.

Can captions match my brand colour? Yes. Vizard Agent takes a hex value for the active-word highlight.

How long does a batch take? Longer than one video, much less than ten. Vizard Agent generates the looks and voices in parallel.

Is this the same as re-uploading one video ten times? No, and the difference matters. Those are one creative wearing ten hats; Vizard Agent produces ten genuinely different videos, which is what a paid test is actually measuring.

Can I add more settings later? Yes. Vizard Agent reuses the pinned voice and the timeline, so an eleventh setting is a small job rather than a new one.

What do I do with the results? Run them, kill the losers, and come back. Vizard Agent builds the next batch around whichever hook won.