Generate captions for short clips with AI

What you'll make

A captioned short clip — accurate subtitles timed to the speech, styled for vertical video — ready to post to TikTok, Instagram Reels or YouTube Shorts. You'll have a repeatable process that takes minutes per clip instead of typing every word.

What you need

• A Claude account (or ChatGPT — either works for this)
• A short clip (15–60 seconds) with clear speech
• A free video editor with caption support (CapCut's free tier is the most common choice for clippers)
• 15 minutes

Steps

1. Get the audio transcript of your clip. If your video editor doesn't auto-generate captions, extract the audio and get a transcript:

I have a [length]-second video clip. Here is what is said:
[paste the spoken content]

Format this as a caption file with timing. Break the speech into short phrases (2–5 words each), and estimate timestamps assuming the full clip is [X] seconds long. Distribute the phrases evenly across the clip duration.

Format as:
[00:00] phrase
[00:02] phrase
[00:04] phrase

2. If you already have a transcript from the previous step (finding clip-worthy moments), use that. Just ask Claude to break it into caption-sized chunks:

Break this text into caption segments of 2–5 words each, formatted with timestamps for a [X]-second clip:
[paste the spoken text]

3. Import the captions into your video editor. In CapCut: add text layers manually using the timestamps, or use the auto-caption feature and then correct any errors using Claude's version as a reference.

4. Style the captions for vertical video. Ask Claude for style guidance:

I'm captioning a short-form clip for [TikTok / Reels / Shorts]. The video is about [topic]. Suggest:
1. A caption font style that's common and readable on mobile
2. Where to place captions on screen (avoiding the platform's UI areas)
3. Whether to use a single colour or highlight key words
4. How many words to show on screen at once

5. Export the video and watch it back on your phone with sound off. If you can follow what's being said from the captions alone, it's ready to post.

Check your result

• Captions are accurate — no misheard words or nonsense phrases
• Each caption segment is 2–5 words (not full sentences that are too small to read)
• Captions are placed where they don't overlap with platform UI (bottom third, above the description area)
• The clip makes sense with sound off

If the auto-generated captions from your editor are full of errors, use Claude's version as the source of truth and type corrections manually. It's faster to fix 5 errors than to type 50 words from scratch.