Word-by-word auto captions for interviews, talking heads, Shorts and Reels — the kind you know, minus the subscription. Plus something no other Premiere Pro plugin does: on music, it separates the vocals from the instrumental first. Then every word lands.
€0 for Shorts & Reels · Pro €79 one-time — no subscription
The finished mix: fragments, wrong words, gaps.
Vocals only: every line, word by word, on the beat.
Measured on one and the same 20-second excerpt of a real music video, both runs on the same machine with the same Whisper model (medium). Not a lab-conditions number.
Music mode is the specialty — underneath sits a full auto-caption workflow for everything spoken. Podcast clips, interviews, tutorials, vlogs: the everyday work you’d otherwise open Submagic or CapCut for.
The same models the big web services run on — just on your machine. 29 languages, auto-detected, full quality even on the free plan.
Uncertain words are highlighted. Enter confirms, Tab jumps to the next — a 60-second Reel is clean in under a minute.
Preview and render share the exact same drawing routine. What you see in the panel is what ends up as a transparent clip over your edit.
Logos, stickers, lower thirds, screenshots — select the clip, pick one of 19 animations, done. Real easing curves, true motion blur via Premiere's own shutter, zero keyframing, and everything stays editable. See the animations in action →
Speech recognition is trained on spoken language. Put a beat under the voice and it hears drums instead of consonants — and starts guessing. That’s why every guide repeats the same advice: for song lyrics, type them in by hand.
CapBeat adds a step no other Premiere Pro plugin has: a second neural network (Demucs, open-sourced by Meta) splits the mix into vocals and instrumental — locally, on your CPU. Recognition only ever hears the voice. Guessing becomes reading.
Web services exist that can do this — in exchange for uploads, subscriptions and a queue. On the Premiere timeline, with nothing leaving your machine, this plugin stands alone.
The plugin runs as a panel inside Premiere. No export, no browser tab, no reimport.
Language is detected automatically — 29 languages. For music: tick “Music / vocals” and the plugin handles the rest.
Your fonts, your colors, word-by-word animation. The preview is pixel-identical to the final render. Every word is editable with a click.
Captions land as a transparent clip above your edit. Your footage is never re-rendered or touched.
Every preset is fully adjustable: font, size, colors, position, words per cue, background box.
Everything computes on your machine — we have zero server cost per video. Charging a subscription for that would be rude.
| Tool | Model | 3 years | Lyrics from a mix | Runs without upload |
|---|---|---|---|---|
| CapBeat Pro | €79 one-time | €79 | yes | yes |
| PremiereCopilot Pro+ | $7.99/mo | $287* | no | yes |
| Captioneer | $15/mo or $150 lifetime | $150–540 | no | yes |
| Submagic | subscription | from $360* | no | web upload |
Prices as listed on each vendor’s website, July 31, 2026. *Annual figures taken from PremiereCopilot’s own comparison table — verify before publishing this page.
Transcription, vocal separation, rendering — all of it runs on your machine. No account, no cloud, no queue. It needs the internet exactly once: to download the AI models on first start. For NDA footage that’s not a footnote, it’s the requirement.
Right-to-left included: Arabic and Hebrew captions are laid out correctly — in reading direction, word by word.
We don’t promise features that don’t exist. This is the honest state of things.
Transcription, live preview, corrections, rendering onto the timeline.
Local vocal separation, lyrics from a finished mix.
Follows your Premiere language automatically.
Detect dead air and cut it from the sequence on request.
Names, punctuation, jargon — checked locally, never rewritten.
19 fly-ins, pops and bounces with real easing and true motion blur — see them in action.
Word-timed captions as a standard .srt file — hand them to any workflow (Pro).
Bring finished transcripts in instead of re-transcribing.
The panel itself is already cross-platform; the bundled AI runtime and the installer are being rebuilt for Apple Silicon. Email us and we'll tell you the day it ships.
Windows shows a SmartScreen warning for any new installer that isn’t widely downloaded yet. Click “More info” → “Run anyway”. A signed build is in the works — the certificate is on order.
Not yet — the download is a Windows installer and will not run on macOS. CapBeat currently needs Windows 10/11 with Premiere Pro 2023 or newer. A Mac build is actively being worked on: the panel is already cross-platform, but the bundled speech runtime and the installer have to be rebuilt and notarized for Apple Silicon. We won’t name a date we can’t keep — email us and you’ll hear the day it’s out.
On our reference clip: 14 recognized words without separation, 47 with — the complete lyric. It shines on clearly sung pop, rap and singer-songwriter material. Extreme metal vocals or very dense mixes remain hard even after separation. So: test your material with the free version first, then buy.
Your version keeps running unchanged, forever. You lose nothing — you just stop receiving new features until you decide, voluntarily, to upgrade to a future major version (discounted for existing customers).
14 days, no questions asked, by email. We don’t want customers who feel burned.
Right after checkout, Polar (our payment provider) emails you a license key. In Premiere, open the CapBeat panel, scroll to the License card, paste the key and click Activate. Done — music mode, all styles and unlimited length are unlocked. Pro covers 3 machines; the panel has a deactivate button that frees up a seat if you switch computers. Lost the key? Open your purchase at polar.sh/purchases — it's listed there.
No. There is no server they could be uploaded to. The only internet connection is the one-time download of the AI models on first start (~1.5 GB speech, ~80 MB separation).
Premiere’s transcription is fine for spoken language and plain subtitles. It can’t do word-by-word social-style animations, and it fails on music like everything else. CapBeat fills exactly those two gaps.