How to Add Auto Captions to Short Videos with Captions

November 21, 2026

Captions is an AI video editing app built around talking-head clips. Record or import footage, and it transcribes the audio, generates animated captions, adds eye contact correction and background removal, and exports per-platform.

Most of its users are making 30-to-90-second vertical clips. That is the case this covers.

What is the fastest route to a finished clip?

Record inside the app if you can. In-app recording skips the file-upload wait and gives you the tool’s own framing guidance.

If you are importing, use the highest-quality source you have, then: pick a caption preset, adjust the aspect ratio to 9:16, cut silences, and export.

Four steps, ten minutes for a one-minute clip once you have a preset you like. The first clip takes longer because you are choosing a style.

How do you choose a caption style that does not look cheap?

The presets fall into three groups: boxed captions, karaoke-style word highlighting, and minimal single-line text.

Karaoke highlighting gets more attention and looks more like a template. Minimal single-line text reads as more professional and works better for B2B. Boxed captions are the safest middle.

Three rules regardless of preset:

  1. One or two lines maximum. Three lines cover the subject’s face.
  2. Keep captions off the lower 20 percent of the frame — platform UI sits there.
  3. Contrast matters more than colour. White with a thin dark stroke is legible on any background.

How do you fix transcription errors efficiently?

The transcript is editable and edits re-time the captions. Do not read the whole transcript. Instead:

  • Skim for proper nouns — company names, product names, people.
  • Check numbers and units (“fifteen” versus “50”).
  • Check homophones your topic uses (“cache” versus “cash”, “site” versus “sight”).

On a one-minute clip that is a 90-second pass. Skipping it is how a clip ends up saying “hundred percent” when you said nothing of the sort.

Does eye contact correction work?

Mostly, at moderate angles. It re-renders the eyes toward the lens, which reads as attentive on a phone screen and slightly uncanny on a large monitor. Small corrections (you looking 10 degrees off) are invisible. Large corrections (reading a second screen) produce a flat stare.

The background removal is more reliable than it was two years ago; hair edges still show fringing, especially with curly hair against a busy background.

What does it cost?

Free tier for watermarked exports and limited features; paid tiers in the region of $10 to $30 per month depending on export limits and resolution, with higher plans for teams. If you publish a handful of clips a month, the entry paid tier is enough; the jump is mostly about watermark removal and export volume.

Where it does not fit

Long-form editing. This is a clip tool. Multi-cam interview edits with audio mixing belong in a timeline editor.

Non-English audio. Transcription quality varies by language and accent; check a sample before committing a batch.

Anyone who needs burned-in subtitles in SRT form for an accessibility standard without styling — the app is designed to export styled video, so check the export options carefully.

Quick Summary

  • Record in-app or import, apply a caption preset, set 9:16, cut silences, export — about ten minutes per minute of finished clip.
  • Prefer one or two lines, keep captions out of the bottom 20 percent of the frame, and prioritise contrast.
  • Fix only proper nouns, numbers and homophones in the transcript; that takes 90 seconds per clip.
  • Eye contact correction is fine for small angles and uncanny for large ones; hair edges still fringe.
  • Free tier is watermarked; paid tiers run roughly $10–$30 per month.