How to copy the caption style from a video you like

Last updated 2026-09-04 · 7 min read

What is the short version?

Caption style is the thing people copy most and describe worst. "Big bold yellow ones like everyone uses" covers about forty different treatments, and the difference between them is exactly what makes a reel look like it belongs to a channel.

1. Go to Vizard Agent with your cut and the reference video.
2. Say you want the captions in exactly that style.
3. Check the highlight colour on the first render.

What do you need before you start?

Your video and a reference clip that actually contains the captions you mean. A screenshot works at a push, but a video is better — the treatment usually involves motion, and a still cannot show you how a word arrives.

• Your cut. Ready for captions.
• The reference video. With its captions visible.
• What to copy. Captions only, or more.
• Your own words. It captions your audio, not theirs.
• A clean version. Useful to have alongside.

What do you type into Vizard Agent?

Be specific that it is the captions you want, not the video. Attaching a reference clip without that constraint is a reasonable request for a video in that style, which is a much larger and quite different job.

Prompt

Variants worth knowing:

• "Same font, size and effects." Names the three things that matter.
• "Just the captions." Rules the rest of the edit out.
• "Also give me a clean version." No burnt-in text at all.

What does Vizard Agent actually do?

Here is the order Vizard Agent worked in on a real reel captioned to match a reference the client attached. Steps two and three are the whole difference between genuinely matching a style and inventing something that merely resembles it from a distance.

1. Renders a clean version of your cut with no captions on it.
2. Inspects the attached reference video and views its caption style.
3. Extracts representative frames from the reference.
4. Views the reference captions in context rather than in isolation.
5. Inspects the existing word timings for your clean cut.
6. Builds matching captions — the case, the weight, the word highlight.
7. Burns them into the clean reel and renders.
8. Reviews the render and inspects the opening caption placement.
9. Corrects the highlight colour to the reference's exact yellow.
10. Fixes the colour tag directly in the subtitle file.
11. Re-renders and reviews the corrected style across the whole reel.
12. Verifies the opening caption's colour and placement.
13. Audits the finished reel and inspects the flagged moments full size.

Step four is worth pausing on. A caption style read from one frame will get the font roughly right and the behaviour completely wrong — whether the highlight moves word by word, whether the line rebuilds or scrolls, how long a phrase holds. Those only exist across frames.

Steps nine and ten are the honest part. The first attempt got the shape right and the highlight colour approximately right, and approximately right is visible. The fix was to set the exact colour value in the subtitle file rather than to describe the colour again.

What does the result look like?

Your video carrying your own words in the reference's treatment — the same typeface, the same case, the same word-by-word highlight in the same colour, sitting where that style puts them on the frame. Vizard Agent captions your audio, never the reference's.

It is worth having the clean version alongside. Burnt-in captions are permanent, and the version without them is what lets you re-caption for a different platform later without re-editing anything.

When does this not work well?

Reading a style off a picture has limits, and typefaces are the main one. A caption style is a font plus a set of behaviours, and the font is the part that may not be available to reproduce.

• A typeface you do not have. Licensing is yours to check.
• A heavily compressed reference. Fine detail is not readable.
• Captions over a busy background. The colour reads differently.
• A different aspect ratio. The placement will not transfer directly.
• Very long words in your script. They wrap where the reference never had to.

How do you fix a result that came back wrong?

Name the specific attribute. Vizard Agent keeps the clean render, your word timings and the extracted reference frames, so a colour or size correction is a re-render of the caption layer rather than a rebuild of the video underneath it.

• "The yellow is wrong." The exact value set, not described.
• "Too big." Resized; everything else held.
• "It wraps badly." Line breaks fixed for your longer words.
• "It sits too low." Repositioned to match the reference.

How does Vizard Agent compare to doing it yourself?

By hand this means pausing someone else's video, squinting at a single frame, guessing at the font, and then discovering three renders later that their highlight was a slightly different yellow from the one you picked out of memory — which is the sort of error nobody can name and everybody notices.

Common questions

Can I send a screenshot instead? Yes, but a video is better. Vizard Agent needs motion to read the behaviour.

Does it copy their words too? No. Vizard Agent captions your audio in their style.

What if I do not have the font? Vizard Agent finds the nearest available and tells you it substituted.

Can it match a competitor's whole look? Yes, but that is a bigger request. Say so explicitly.

Will it get the highlight colour exact? Yes, once corrected. Vizard Agent sets the value rather than approximating.

Can I reuse the style next time? Yes. Vizard Agent keeps it and applies it to the next video.

Does it work for another language? Yes, though the wrapping will differ from the reference.

Can I get the subtitle file? Yes. Ask and Vizard Agent delivers it alongside the render.

What about the placement? Vizard Agent matches the reference, then checks it against your framing.

Will it cover my subject? No. Vizard Agent checks placement against what is in your frame.

Can it burn them in and give me a clean cut? Yes. Vizard Agent produces both out of the same pass.

How close is the match? Close enough that the two read as the same channel. That is the point of sending a reference to Vizard Agent rather than describing one.

Can I tweak one thing about it? Yes. Name the attribute and Vizard Agent holds the rest of the match.

Why not just describe the style? Because the words people use for caption styles are not precise enough, and the reference clip settles every question at once.