Ship every video in every language.
Upload a video or paste a link. VTranslate transcribes it, translates every line together with the lines around it, and hands back subtitles, or the video with captions burned in. Your transcript only goes to models that promise not to keep it.
NewTranslate one video right now, no account neededSign up in under a minute. Pay in Monero, no card details needed.
Three steps from raw footage to a localized release.
- 1
Bring the video
Upload a file from your device, or paste a link from YouTube, Vimeo or any other public platform. One video per job. The original filename stays metadata and never becomes a storage path.
- 2
Choose languages and a model
List every target language at once, then pick the model that translates them. Private models are ranked first and marked recommended.
- 3
Watch it run, then download
Queue position and progress update live. When the job finishes, download a subtitle file per language, or the video with captions burned in exactly as the preview showed them.
Every line is translated with its neighbours.
A caption translated on its own forgets what "it" refers to, who "you" is, and the tone of the sentence it came from. VTranslate translates in overlapping windows, so names, pronouns and register carry across cuts, then puts each line back on its original timing.
"It's ready." → "Está pronto."
Alone, the model guesses a masculine subject."It's ready." → "Está pronta."
The line before was about the new camera, a câmera, so the adjective agrees.The new data brokers are AI companies.
Translating a video means handing a provider the entire transcript, and that is often unreleased work: a course that has not launched, an interview under embargo, an internal briefing. What happens to it next is decided by a retention policy. VTranslate shows you which guarantee every model carries.
Not retained, not trained on
The provider commits to discarding the prompt after the response and never training on it. This is the floor, and it is our default.
Runs inside an enclave
Inference happens in hardware the operator cannot read into, and the enclave signs an attestation you can check.
Encrypted end to end
The request is encrypted to that enclave, so the plaintext never exists on the provider's disks at all.
Bring your own key, or use ours.
Install your own provider key and the account, the bill and the choice of model are yours. Without one, jobs run on the platform's key, which is only ever spent on models that cannot trace a request back to its owner.
Everything in the dashboard is an API call.
Mint a key that carries only the scopes it needs, with reads and writes kept separate, and drive jobs from your own pipeline. The secret is shown once and the server keeps only a peppered digest of it. Job progress streams over a live connection, so there is nothing to poll.
$ curl -H "X-API-Key: vt_live_…" \
https://vtranslate.cc/api/v1/jobsPay for output, not for attempts.
Before a job starts, a hold sized to the video is placed on your balance. When it finishes you pay what the model actually cost plus a platform fee, and the rest of the hold comes back. A job that fails releases its hold and charges nothing.
Prepaid credit
Top up in Monero. No card, no bank, no billing address. You always see your balance, what running jobs are holding, and a warning before you run low.
Monthly plans
For creators publishing every week and teams running high-volume channels: more included videos and a lower fee on every success. Current plans are listed inside the app.
Your own key
Bring a provider key and the provider bills you for model usage directly. Every model becomes selectable, labelled for exactly what it is.
Great work, generously shared.
We are delighted to use Sintel and Big Buck Bunny in our test corpus. Thank you to the Blender Foundation and the film teams, Wikimedia subtitle contributors, and Tatoeba's community for sharing the films, words, and voices that help us test VTranslate.
Your next video, in every language your audience speaks.
Sign in, drop in a file or a link, and pick your languages.