Skip to main content
CutXflow
Try free

How CutXflow works

Last updated

This page explains what CutXflow does with your video: what stays on your device, the one kind of media that is sent to a server, where the limits are, and what CutXflow does not do. It describes the current version of the product.

Is my video uploaded?

No. CutXflow reads your video file inside your browser and does not upload it. The only media sent to a server is the video’s audio track, compressed to a small file on your device and forwarded by CutXflow’s server to a third-party speech-to-text service to produce the transcript. Caption rendering and MP4 export also run on your device.

What stays on your device

  • Your video file. Your browser reads it from disk; CutXflow never sends it anywhere.
  • Audio extraction, the waveform, the caption preview and every edit you make.
  • Caption rendering, MP4 encoding and the SRT file.
  • A local copy of the video in your browser’s storage (IndexedDB), so a saved project can be reopened without choosing the file again. It does not leave your device, and your browser may clear it.

What is sent to CutXflow’s servers

  • The video’s audio track, compressed to 32 kbps Opus: about 240 KB per minute of video, typically a few hundred kilobytes for a short reel.
  • With it: the language you chose, the clip’s length in seconds, your sign-in token so the right account is billed, and a one-way fingerprint (SHA-256) of the audio and language so an accidental resubmit is not charged twice.
  • Account and billing traffic: sign-in through Firebase Authentication, payments through Cashfree and, if you are signed in, your caption text, timings and style settings, saved to your account. The Privacy Policy lists what is collected.

Is the audio kept after it is sent?

CutXflow’s server does not save the audio, but the speech-to-text provider it is forwarded to, ElevenLabs, applies its own retention policy and may keep a copy. CutXflow returns the transcript to your browser and records only usage figures for the request (seconds transcribed, credits charged, language). The finished transcript can stay in the server’s memory for up to an hour so a repeat request is not billed twice.

What the speech-to-text provider does with audio it receives is governed by that provider. CutXflow uses ElevenLabs, whose privacy policy is public. Our own Privacy Policy states that CutXflow does not save the audio, that ElevenLabs may keep a copy under its own policy, and that your content is not used to train machine-learning models.

Do I need an account?

Yes. The caption studio asks you to sign in with Google or an email address, and the server rejects a transcription request that does not come from a signed-in account; email sign-ups must verify their address first. The account tracks your transcription credits and saves your projects. The free plan includes 3 min of transcription a month.

How does CutXflow turn a video into captions?

Five steps take a video file to finished captions. Steps 2 and 3 run automatically; the rest are yours.

  1. Sign in and drop in a video. Sign in with Google or an email address, then choose an MP4, MOV, WebM, MKV, M4V, AVI or 3GP file. Your browser checks the file type, size and duration on your device; nothing is uploaded yet.
  2. Extract the audio on your device. CutXflow drops the video track and compresses the soundtrack to a small 32 kbps Opus file, about 240 KB per minute of video. Where the browser supports it this uses WebCodecs; otherwise a bundled ffmpeg.wasm audio engine, about 31 MB downloaded once per session, does the job.
  3. Transcribe with word-level timestamps. The audio, your chosen language and the clip length are sent to CutXflow's server, which forwards the audio to ElevenLabs' Scribe speech-to-text model. The transcript comes back with a start and end time for every word. For Hinglish, CutXflow then rewrites the Hindi transcript in Latin letters word by word, leaving every timestamp unchanged.
  4. Review and edit. Pick a caption style and preview it in the browser. Fix any word's spelling, timing or colour, split or merge lines and move the caption block. Words the model was least sure about are flagged so you know what to check first.
  5. Export a video or an SRT file. Export draws each caption onto every frame of your video in the browser and encodes an H.264 MP4 with WebCodecs, keeping the source frame rate. Or download an SRT subtitle file built from the same caption lines. Neither sends your video anywhere.

How accurate are the captions?

CutXflow’s caption accuracy depends on the speech-to-text model and on your recording. It varies with audio quality, accent and background noise, so no single accuracy figure holds for every video and this page quotes none. Every word can be corrected in the editor before you export, and the words the model was least sure about are flagged for review.
  • Where the words come from. The transcript is produced by ElevenLabs’ Scribe model (currently scribe_v2). CutXflow does not run or train a speech-recognition model of its own, so how well your recording is recognised is down to the provider’s model.
  • What CutXflow does to the transcript. It keeps only spoken words — spacing, sound-event labels and stray punctuation tokens are dropped — strips punctuation from each word so animations stay clean, rounds each timing to a whole millisecond and trims overlaps so two words are never highlighted at once. Timings are otherwise the model’s own; drag any word against the waveform to correct one.
  • Hinglish. The speech model has no Hinglish mode, so for Hinglish CutXflow requests Hindi and then rewrites each Devanagari word in Latin letters with its own rule-based converter, including Hindi schwa deletion. Timings are not touched. Spellings follow how Hindi is usually typed in Latin (kya, bahut, ladka) but can differ from yours, and the converter cannot fix a word the model heard wrongly.
  • Languages. The studio offers 15 Indian languages, Hinglish, 36 other languages and automatic language detection.
  • Test before you pay. The free plan includes 3 min of transcription a month, enough to try your own footage. Used transcription minutes are not refundable (see the Refund Policy), so it is worth trying first; plans and allowances are on the Pricing page.

Which browsers can export?

CutXflow exports video with WebCodecs, so the browser must provide it. The requirement CutXflow documents is a recent Chrome or Edge on desktop, and 2K and 4K export are desktop-only. The export checks for WebCodecs before it starts and says so if it is missing. SRT download does not use WebCodecs.
  • The documented requirement. A recent Chrome or Edge on desktop, because export uses WebCodecs. The FAQ, the Pricing page and the Terms say the same.
  • Other browsers. WebCodecs is not in every browser and support keeps changing; caniuse has the current table and MDN documents the API. If a browser provides it, export will try to run, but CutXflow does not promise it will finish outside the documented requirement.
  • Phones. Phones are outside the documented requirement. The export includes fallbacks for them (lower bitrate limits, software decoding, a 720p retry), but there is no guarantee, and 2K and 4K are desktop-only.
  • Resolution. The export panel only offers resolutions the device can encode: 2K and 4K need a desktop with enough memory, and a resolution is hidden when the device reports it cannot encode it reliably at your clip’s frame rate.
  • What comes out. An MP4 with H.264 video and AAC audio. The frame rate of your source is kept, the picture is never upscaled, and the audio is copied unchanged where the container allows it or re-encoded to AAC at 128 kbps where it does not.
  • When an export fails. CutXflow retries with different codec settings, up to 5 attempts, and tells you if it had to drop to a lower resolution to finish. If the device cannot decode your source codec (for example HEVC on some phones), it can convert the clip to H.264 in the browser first, for files under a size limit; if the audio cannot be carried over, it says the export will be silent.
  • Time and focus. Encoding runs on your device at the speed of your hardware and there is no cloud queue, so a long or high-resolution clip takes longer on a slow machine. Keep the tab open and in front: a backgrounded tab can interrupt an export.

What does CutXflow not do?

  • It does not upload, store or view your video file; its servers never receive it.
  • It does not translate captions. They are written in the language spoken, or in Latin letters for Hinglish.
  • It does not label speakers or sounds. Speaker separation and sound-event tags such as (laughter) are switched off in the transcription request.
  • It does not run its own speech-recognition model, and it does not guarantee the transcript is correct: the Terms leave reviewing captions before you publish them to you.
  • It has no cloud render queue and does not host or share your exports; files download straight to your device.
  • It does not transcribe clips of any length. Transcription currently handles clips up to about 17 minutes, because the compressed audio has to fit in a single request.

That makes CutXflow a poor fit if you need to export from a browser without WebCodecs, if your clips run longer than about 17 minutes, or if you want a service that renders video on a server.

Where to read more