How to add captions to video with AI (and when a free tool is enough)

Adding captions to a video with AI is a one-click job in 2026 — every mainstream editor ships auto-captions — but accurate captions are not. Speech models still misread technical vocabulary, names, numbers, and sung or accented speech, so the real workflow is four steps: generate, correct, style, export. And be clear-eyed about tooling: for caption-only jobs, free options are genuinely enough. Microsoft's Clipchamp auto-captions and exports 1080p with no watermark at $0, as of July 2026 — you do not need a subscription to caption a finished video.

How accurate are AI auto-captions in 2026?

Very good on clean, conversational speech; unreliable at the edges. The failure modes are consistent across tools: domain jargon, proper nouns, numbers and units, overlapping speakers, heavy accents, and lyrics. Descript — the transcript-first editor whose entire editing model rests on transcription quality — is the clearest case study: its reviewers report accuracy dropping sharply on technical vocabulary, with some users in niche fields describing two to three hours of correction per hour of audio. Descript's own tiering tells the same story: the free plan includes 60 minutes of transcription a month with 720p watermarked exports, and the $16/month Hobbyist plan (billed annually) lifts that to 10 hours and clean 1080p, as of July 2026 (descript.com/pricing).

The practical conclusion: treat auto-captions as a fast first draft. Never ship them uncorrected, and budget correction time proportional to how technical the speech is.

Step 1: generate the captions

Run the auto-caption pass in whatever editor holds your timeline. One decision to make now, not later: whether captions will be burned into the picture or delivered as a separate subtitle file. Social feeds favor burned-in captions with deliberate styling; long-form platforms and accessibility workflows favor subtitle files viewers can toggle. That choice determines which export settings matter in step 4.

Step 2: correct the transcript

Do a single disciplined pass with the audio playing. Fix proper nouns first — product names, people, places — then numbers and units, then homophones the model guessed politely wrong. Punctuation earns attention too: caption lines read one at a time, so a period in the wrong place changes meaning more on screen than on paper. If the video is technical, keep a list of your domain's recurring terms; you will retype the same corrections on every video, and knowing them makes the pass mechanical.

Step 3: style for readability and safe zones

Readability rules are old broadcast craft, and they still hold: at most two lines per caption, roughly 40 characters per line, high contrast (light text over a dark backing bar survives every background), and no styling that outshouts the video. Sentence case reads faster than all-caps.

On 9:16 vertical video, placement is the step most people get wrong. The bottom band of the frame is covered by platform UI — progress bars, captions of the platform's own, engagement buttons along the right edge — so captions parked at the bottom get overlapped exactly where viewers look. Keep captions in the central band, just below the vertical midpoint, clear of the right rail, and never across a speaker's mouth. On 16:9 the classic lower-third placement remains correct.

Step 4: export

Burned-in captions travel with the file and look identical everywhere; subtitle files stay editable and toggleable but depend on the player. For social clips, burn them in. For an archive or a multi-language workflow, keep the subtitle file as the source of truth. Either way, export at the resolution the video deserves — captions are the one overlay viewers stare at, and compression artifacts on text are the first thing they notice.

Which captioning tool fits your job?

Tool	Auto-captions	Cheapest captioned path (July 2026)	Free-tier catch	Best when
PrismPoster	In the timeline editor	Creator, €29/mo	Timeline (and its captions) not on the Free plan	The video was generated here; long exports
Descript	Transcript-first editing	Free, 60 min/mo; $16/mo annual for 1080p	720p, watermarked exports	Talky long-form and podcasts
CapCut	Yes, strong	Free tier	Feature matrix unpublished since the early-2026 restructure	Quick vertical clips
Clipchamp	Yes	Free	None serious — 1080p, no watermark	Caption-only jobs on existing footage
Kapwing	Yes — about 1,000 subtitle minutes/mo on Pro	Pro, $16/mo annual	Free exports capped at 1 minute	Team caption workflows
VEED	One-click subtitles	Creator, $10/mo annual	Free is 720p, watermarked, ~10-minute cap	Styled social clips

The free honesty, spelled out: Clipchamp captions and exports at 1080p with no watermark on its free tier, as of July 2026 (clipchamp.com), which makes it the default answer for captioning footage you already have. CapCut's auto-captions remain among the best, but its early-2026 price restructure moved previously free features behind paywalls and the current free matrix is not clearly published — verify in-app before relying on it. Kapwing's Pro tier includes roughly 1,000 subtitle minutes a month (kapwing.com/pricing), while its free tier's 1-minute export cap makes it a demo. Deeper tool-by-tool comparisons live in Descript alternatives and CapCut alternatives.

Where PrismPoster fits

PrismPoster's auto-captions live inside its browser Video Editor, which comes with the Creator plan at €29/month — stated plainly: captioning here is part of a paid editing workflow, not a free utility. The editor exists for media you generate in PrismPoster's studios, and the caption flow is built around that pipeline: caption the cut on the multi-track timeline, style it, and the captions render into a server-side export with no duration cap — an hour-long recording captions and exports without your laptop doing the rendering, which is where podcasters and long-form creators feel the difference (why that matters for browser editors, and what it means for podcast workflows).

Two honest boundaries. If all you need is captions on camera footage, use Clipchamp or CapCut — free tools cover that job completely. And every PrismPoster export carries embedded AI-provenance metadata on every plan; that is a provenance disclosure, not a branding watermark you pay to remove.

Frequently Asked Questions

How accurate are AI auto-captions?

Strong on clean conversational speech, weak on jargon, names, numbers, accents, and lyrics. Users of transcript-based editors report multi-hour correction passes on technical content, so treat auto-captions as a draft and always run a correction pass.

What is the best free way to add captions to a video?

Clipchamp: auto-captions with 1080p watermark-free exports on the free tier, as of July 2026. CapCut is also strong, though its free-tier limits have been unclear since its early-2026 restructure.

Where should captions sit on a 9:16 video?

In the central band just below the midpoint of the frame — clear of the bottom band and right edge, where platform progress bars and buttons overlap the picture, and never across a speaker's mouth.

Should captions be burned in or exported as a subtitle file?

Burn them in for social feeds, where styling and guaranteed visibility matter. Use a subtitle file when viewers should be able to toggle captions or when you maintain multiple languages; keep it as the editable source of truth.