Open source · View on GitLab · OpenDub at opendub.app keeps its code and releases at gitlab.com/open-dubapp/OpenDub; other projects that use the name OpenDub are separate from it.
What local AI dubbing means here
With local AI dubbing, nothing about your video goes to a dubbing service. In the OpenDub app with the free voice, every step runs on your computer:
- Separate. Demucs splits the voice from the music and effects, so the soundtrack survives.
- Find speech. Silero VAD marks where someone is talking.
- Transcribe. Whisper writes the words and their timing.
- Translate. A local translation model (opus-mt) translates each sentence.
- Speak. OmniVoice speaks each line in a clone of the speaker's voice, taken from a few seconds of clear sentences in the video itself.
- Mix and render. The new voice goes over the music, and ffmpeg writes the dubbed MP4 with subtitles.
The app answers only on your own computer, at 127.0.0.1. When you dub with the free voice, the audio goes from the page to the app on the same computer and never onto the internet. The privacy page lists what each part of the app receives and keeps.
What you download for local AI dubbing
- The app: downloads for Mac (Apple Silicon) and Windows carry Python, ffmpeg and everything else inside. On macOS and Linux there is also a one-line install.
- The models, once: about 3.6 GB on a Mac and 6.5 GB on Linux. 3.3 GB of that is the free voice's model, which comes down on the first dub.
The models come from Hugging Face. Nothing of yours is sent to download them, though Hugging Face sees the request, like any website you visit. After that they stay on your computer.
How fast local AI dubbing is
OmniVoice runs on the CPU at about 15–35 seconds a line on a laptop. A short clip takes minutes; a long video takes much longer, and the app has no length limit, so you can leave it running. Separating the voice is faster in the app than in a browser tab. In the app you can then change any line's translation, tone or timing and re-dub just that line.
How to start local AI dubbing
- Install the app from opendub.app on your Mac, Windows or Linux computer.
- Choose the video. An MP4, MOV or WebM file on your computer. Its language is detected automatically.
- Choose the language to dub into. One of the 17 the app dubs into.
- Choose OmniVoice as the voice model.
- Choose what happens to the original sound. Replace the voice keeps the music and effects and swaps only the speech; Voice-over leaves the original faintly underneath.
- Start the dub. The first dub downloads the voice model, once.
The app gives you the dubbed MP4, the dub audio as WAV, subtitles in both languages as SRT, and the clip the voice was cloned from. The desktop app needs no account.
Local AI dubbing in the browser, with nothing to install
If you cannot install anything, the card at the top of opendub.app also dubs on your device, in the browser on a computer or an Android phone: with the free voice the words, the translation and the voice are made in the tab, and the video is not uploaded. That voice is a stock voice rather than a clone of the speaker, it speaks fewer languages, and it stops at five minutes per video. See Translate video to Spanish or Translate video to Hindi for how it works.
What is not local
- With a Higgs Audio or ElevenLabs key, speech clips and text go to that provider, and you pay it directly. The video file itself is never sent.
- The command-line route (
python -m opendub.pipeline) uses a Higgs Audio key. - The free translation models are mostly to and from English. For some pairs the free route has no translator, and then a Higgs Audio key does the translation.
Limits of local AI dubbing with OmniVoice
- OmniVoice's model weights are licensed for non-commercial use only. For commercial work, use your own Higgs Audio or ElevenLabs key.
- It is slow on a laptop CPU: about 15–35 seconds a line.
- It does not change the lips in the picture.
- It works on a video file, not on a link to a video online.
- The Mac download is for Apple Silicon.
Open source, so you can check it
OpenDub (opendub.app) is open source under AGPL-3.0-or-later. Every step of local AI dubbing above is plain code in the repository on GitLab, so you can read what runs on your computer before you run it. Other projects that use the name OpenDub are separate from it.
Questions
Can I do AI dubbing locally without uploading my video?
Yes. In the OpenDub app on your computer with the free OmniVoice voice, separation, transcription, translation, the voice and rendering all run on your computer, and your video and audio never leave it.
Is local AI dubbing with OpenDub free?
Yes. The app and OmniVoice are free. OmniVoice's model weights are licensed for non-commercial use only.
Does the voice sound like the original speaker?
Yes. OmniVoice clones the speaker from a few seconds of clear sentences in the video itself. Only clone voices you have the rights to use.
Which computers does it run on?
Mac (Apple Silicon) and Windows have downloads; macOS and Linux also have a one-line install. OmniVoice runs on the CPU.
Is there a length limit?
Not in the app on your computer. The browser version stops at five minutes per video.
Do I need an account?
Not for the app on your computer. On opendub.app, downloading a dub takes a free sign-in.
Is there a watermark on the dubbed video?
No. OpenDub adds no watermark and no logo to the dubbed video. The picture is left as it is; the only thing drawn on it is the subtitles in the new language.