How to Add Auto-Generated Captions to a Video With AI

The fastest way to add captions to a video is to transcribe its own audio track and burn the words onto the screen, timed to when they're actually spoken — no manual typing, no subtitle file to sync by hand. Most browser-based editors charge for this per minute; on Arcframe's timeline, the captioning step itself is free once the clip is transcribed.

This only works one way, though, and it matters: the captions say what the video says. If the audio is in Hindi, the captions come out in Hindi. If you need the words translated into a different language, that's a separate dubbing step, not a caption setting — more on that below.

What "auto-generated captions" actually means here

There are two different things people call "captions," and mixing them up wastes time:

Feature	What it produces	Typical cost
Transcription-based captions (this post)	On-screen text in the same language the speaker is using, timed word-by-word to the audio	Transcription: ~1 credit per 3 minutes of source audio (free if that clip was already transcribed). Burning the captions onto the timeline: free
Translated subtitles / dubbing	Text or a new voice track in a different language than the original audio	A separate text-to-speech or dubbing pass, priced per generation

This post covers the first one. If you need a video re-voiced in another language entirely, look for a dubbing workflow instead — captioning a video doesn't translate it.

Step 1: Get the video onto a timeline

Start a timeline and add the clip as a media source. There are three ways to bring a video in:

• It's already one of your AI generations — add it by its job reference.
• It's a public file link (a direct .mp4/.mov/.webm URL, including most Google Drive "anyone with the link" shares and Dropbox share links) — import it directly.
• It's a file sitting on your device — get a one-time upload link, push the file to it, then add it the same way.

Place the clip on the main video track. It needs to actually be positioned in the timeline before it can be transcribed — just having it in your media library isn't enough.

Step 2: Transcribe the audio

Transcription runs with word-level timing, which is what lets the captions land exactly when each word is spoken rather than appearing as one static block for the whole sentence. It costs roughly 1 credit per 3 minutes of source audio the first time a given clip is transcribed — if you've transcribed that exact clip before (on any timeline), it's free the second time, because the transcript is already on file.

This step is also what powers filler-word and dead-air trimming elsewhere in the editor, so if you've already cleaned up a talking-head clip, you've likely paid this cost already and captioning it costs nothing extra.

Step 3: Burn the captions onto the timeline

Once a transcript exists, generating captions is free. It places short, one-line captions on their own track, synced to the words as they actually play — which matters if you've already cut filler words or shortened pauses, since the captions follow the edited timing, not the original recording's.

Re-running this step after further edits doesn't stack duplicate captions — it replaces the previous set with one matching the current cut, so you can caption early, keep editing, and recaption at the end without cleanup.

Step 4: Export

Rendering produces a standard 1080p MP4 with the captions burned into the picture — meaning they display on every platform without relying on a separate subtitle file or a platform's own caption toggle. That's the right choice for Instagram Reels, TikTok and LinkedIn video, where a large share of viewers watch muted and a platform-native caption toggle isn't guaranteed to be on.

What this does not do

Being specific about the limits here is more useful than pretending there aren't any:

• No translation. Captions reflect the language actually spoken in the clip. A video narrated in Tamil gets Tamil captions, not an English translation.
• No manual styling per word. The captions are one-line, word-timed text — if you want a specific font, animated pop-in style, or emoji per word, that's a different kind of caption design than this feature is built for.
• Burned in means permanent. Once exported, the captions are part of the video image, not a toggleable overlay. If you need captions you can turn on and off, you'll want a platform that exports a separate subtitle file (.srt/.vtt) instead of burning them in.

If you need the video in a different language

Captioning and dubbing solve different problems, and it's worth knowing which one you actually need before starting:

Your situation	What to use
Video is in English, viewers are English speakers, you just want the words on screen	Transcribe + caption (this post)
Video is in English, but your audience speaks Hindi, Tamil, or another Indian language	Generate a narration track in that language first (Arcframe supports text-to-speech across 11 Indian languages), then caption the new audio
You want the exact same voice, just speaking a different language	Voice cloning plus a dubbing pass, not captioning

Where this fits if you're starting from zero

Arcframe's free tier includes 20 one-time credits with no card required, which is enough to transcribe and caption several short clips before you need to top up. Captioning itself, once a transcript exists, doesn't draw from that balance at all. Current credit and plan pricing is kept up to date on the pricing page rather than restated here, since those numbers change.

If your footage also needs cleanup — filler words trimmed, dead air shortened, background music added — those steps share the same transcript this captioning step uses, so doing them in the same pass before you caption and export is the more efficient order.