// HOW-TO · CAPTIONS

How to edit and translate X video captions (editable captions + Grok, 2026)

Edit X's auto-generated video captions by hand, fix what transcription breaks, and use Grok to correct and translate them — the 2026 editable-captions workflow.

Last verified · 2026-10-06 · by Moe Ameen

X made its video captions editable in early October 2026. Before the update, the captions X auto-generated from your clip's audio were effectively locked — a wrong brand name or a mangled number stayed wrong. Now, inside the video composer, you can edit the generated transcript by hand, ask Grok to correct it, or tell Grok exactly how to change it, including translating it into another language. The announcement came from Allegra Jacchia, who leads X's Creators Product team.

This is the caption-editing task specifically — getting the words right and, if you need it, into a second language — not a tour of the whole composer (for that, see [how to use X's updated video composer](/how-to/use-x-updated-video-composer)). The steps below are the lifecycle of a caption on X: generate it, read it against the audio, fix the predictable errors, hand the hard cases to Grok, translate and verify, then check it reads on a muted timeline before you post. The whole point of editable captions is that there is no longer an excuse to ship the raw transcription — so the real work is the review, and that is what this covers.

The steps

Generate the captions from your clip. Attach your video in the composer and let X auto-generate captions from the audio. This first pass is a transcription, not a finished caption track — treat it as a draft to correct, which is exactly what the update now lets you do. Get the trim and clip order settled first, because editing captions before you have locked the cut means re-checking timing on lines you may delete.
Read the transcript against the audio, line by line. Play the clip with the captions on and read every line while you listen. Auto-transcription is accurate on plain speech and unreliable on exactly the words that matter most: your brand and product names, people's names, numbers and prices, acronyms, and homophones (their/there, to/two, a mis-split company name). These are the errors that make a caption look careless or change the meaning, and they are the ones you are here to catch.
Fix the predictable errors by hand. Edit the wrong lines directly in the composer. Prioritize: correct spellings of names first, then numbers, then homophones, then punctuation that affects how a line is read. A product name spelled wrong is worse than no caption at all, so do not skip it because the rest of the line looks fine. Hand-editing is the right tool for short, specific fixes you already know the answer to.
Hand the messy lines to Grok with a precise instruction. For lines that are garbled rather than one-word-wrong, pass them to Grok. Vague prompts get vague fixes, so be specific: tell it the correct name or term, or describe the change you want ("tighten this to fit two lines," "my company is spelled Kompozy, fix every instance"). Grok's correction is a draft — read what it returns, because an AI cleanup can confidently rewrite a line into something fluent but wrong.
Translate with Grok — then verify a language you cannot read. To caption for a second-language audience, ask Grok to translate the caption track into that language. One thing to be clear about: this translates the on-screen caption text, it does not dub or re-voice the spoken audio — the clip still speaks the original language. Before posting a translation you cannot read, sanity-check names and numbers (they should survive untranslated), and if the audience matters, have a speaker glance at it. A fluent, confidently wrong translation is harder to catch than a garbled one.
Check styling, placement, and the sound-off read. Captions earn their keep in muted feeds, so confirm they are actually readable: a clear band, sitting above X's playback controls and the mobile reply bar, not clipped at the edges. Preview on a phone, not just the desktop composer. If a translated line runs long, tighten it — a two-line cap that fits beats an accurate line that spills off-screen.
Add post text and alt text, then post. The on-video captions, the post copy, and the alt text are three separate jobs. Write post text that frames the video (X still reads it for search and reach), add alt text describing the video for accessibility, then post or schedule. Remember the boundary: these captions are baked into this X upload. Download the file for TikTok, Reels, or Shorts and you re-caption from scratch for each one.

Common gotchas

Editable does not mean edited. The composer makes fixing captions easy but will happily post the raw transcription if you skip the pass — the review is the step, not the feature.
Grok's correction and translation are drafts. It still misreads names, numbers, and idioms; proofread before posting, especially a translation into a language you do not speak.
Grok translates the caption text, not the audio. The clip still speaks the original language — this is subtitling for reach, not dubbing.
The captions are X-native. They stay on this X post; downloading the file to post elsewhere leaves the caption track, crop, and styling behind.
Translated lines run longer than English. Check the sound-off read on a phone — an accurate caption that spills past the frame edge is worse than a tightened one.
Rollouts on X are staged. If a caption or Grok control is not where this describes, it may not have reached your app or account tier yet; update the app and check again.

Where Kompozy fits

Read back through these steps and notice what the task actually is: proofreading a machine's guess at what your audio said. You generate a transcription, then hunt for the names, numbers, and homophones it got wrong, and verify a translation you may not be able to read. X's editable captions make that hunt faster, but the hunt is still the job — because the caption is reverse-engineered from audio, so the errors are baked in before you ever open the composer. [Kompozy](/) attacks the problem from the other end: it removes the thing you are proofreading.

In Kompozy the caption is not a transcription of audio — it is the authored script the video was built from, so the words on screen are the words you wrote, not a model's best guess at them. That flips the error class this tutorial is all about. Your brand and product names are spelled the way the [Persona Brief](/glossary/persona-brief) says to spell them, so "my company is spelled Kompozy, fix every instance" never has to be typed because it was never mis-transcribed. Numbers come from the source copy, not from a model mishearing a price. The captions are then burned into the frame, word-synced, from a reusable style preset in HyperFrames — so placement and the sound-off read are set once instead of re-checked clip by clip. For a genuine second-language version, a HeyGen persona can speak the target language on camera rather than only subtitling it, which is the one thing Grok's translation explicitly does not do.

Honest boundary: to fix the captions on one clip you already filmed for X, use X's composer — it is right there and it is good at it. Kompozy is for when captioned, correctly-spelled, on-brand video has to ship across the eight social platforms plus blog and email on a repeating cadence, and the hand-proofing you just did does not scale to that. Starter ($199/mo for 5,500 credits) fits a solo creator; Pro ($499/mo for 18,000 credits) suits a brand or agency running many accounts; Enterprise is custom.

Frequently asked questions

Can you edit auto-generated captions on X now?

Yes. As of an early-October 2026 update to X's native video composer, you can edit the auto-generated caption transcript by hand before you post, rather than accepting the raw transcription. You can also ask Grok to correct the captions or tell it exactly how to change them. The fix is part of the upload, so do it before posting.

How do I translate X video captions into another language?

In the updated composer, ask Grok to translate the caption track into your target language. This is X's built-in path to subtitling a clip for a non-English audience without a separate tool. It translates the on-screen caption text only — it does not dub or re-voice the spoken audio — so review the result for names and idioms before posting.

Does Grok dub the audio or just translate the captions?

Just the captions. Grok translates the caption text that appears on screen; the video's spoken audio stays in its original language. If you need the voice itself to speak another language, that requires dubbing or re-voicing, which X's caption tools do not do — they subtitle, not dub.

What do auto-captions get wrong most often?

Proper nouns and numbers. Auto-transcription handles plain speech well but reliably stumbles on brand and product names, people's names, prices and figures, acronyms, and homophones (their/there, to/two). Those are the lines to read against the audio and fix first, because they are the errors that change meaning or make a caption look careless.

Do the edited captions carry over if I post the video elsewhere?

No. The captions, trims, and styling you create in X's composer are baked into that X upload only. Download the file and post it to Instagram, TikTok, or YouTube and none of the X-native editing travels — you caption and crop again for each platform. That per-platform repetition is the real cost of relying on each app's native editor.