Desktop AI dubbing across AI providers, with 70+ translation languages, or fully offline and free — no keys, no internet, pay no one. Your video never leaves your computer.
Run it free and offline, or mix in cloud providers when you want more languages and quality. The video file stays on your computer either way.
0+
translation languages
0
cloud providers
0+
on-device models
Fully local
On-device models for every step. Your media never leaves your machine.
30+ on-device models, including Whisper and Omnilingual ASR for transcription, Qwen3.5, Qwen3.6, Gemma 4 and Index-Translate for translation, and Supertonic 3 and Kokoro for speech synthesis. Qwen3.5 0.8B is the default local translator on Windows and Gemma 4 E2B on Mac, unless Apple Foundation Models qualifies; Hy-MT2 is another option. The largest translation models, such as Gemma 4 26B-A4B, need a very powerful computer with lots of memory. No keys, no internet after model installation, nothing to pay per minute. Language coverage is more modest than cloud.
Mix & match
/
Some steps on-device, others in the cloud, even a different vendor per speaker.
Transcribe locally, translate with GPT, voice each speaker with a different vendor, all in one render. Swap any step whenever you like.
Fully online
Bring your own OpenAI, xAI, ElevenLabs or Google keys. Paid directly, no markup.
Best quality, the widest voice and transcription coverage, and all 70+ translation languages. You pay providers at their public rates, with no markup from us. Your video never leaves your computer.
See & hear it
Same clip. New voice.
Flip between the German original and the English dub on the exact same footage. The picture is untouched. Only the voice changes.
DE Original
Press play, then switch between German and English
Switch tracks anytime. Playback keeps its place. Loads only on play.
Private by design
Your video files never leave your computer. Cloud providers only see what you opt to send. Or use on-device models, where all media stays local.
Mix any providers
OpenAI, xAI, ElevenLabs, Google, Inworld, Soniox, your own OpenAI-compatible endpoint, or on-device. The macOS build also includes Apple (on-device) speech recognition and translation, the latter through Apple Intelligence or Apple Translation, on macOS 26+ with supported hardware and languages, and speech with Apple system voices on macOS 15+. Combine providers inside one project. No vendor lock-in.
Flat subscription
Pay only to lift the 10-segment dubbing limit. Cloud API keys go straight to providers. We don't resell tokens.
How it works
Three steps. Real editor underneath.
You get the full pipeline, with a checkpoint at every step instead of one opaque button.
STEP 01
Drop in your video
Pick source & target languages, define speakers and assign voices, then add a glossary so names and jargon stay intact.
STEP 02
AI does the heavy lifting
Separation, transcription, translation and dubbing, segment by segment, synced to the original timing. Lines that run long get speed-adjusted. Track transcription progress. If translation stops early, completed segments remain available, with the reason and retry guidance.
STEP 03
Edit, then export
Review each segment's text, timing and speed badges. Then generate voices and create the final video.
Pros who care about glossary, terminology, speaker control.
Indie studios
Small video producers without enterprise budgets.
Public sector
Where data must stay on the local network.
Educators & NGOs
Global audiences, tight budgets, sensitive content.
Signature feature · interactive
Build your pipeline. Watch it react.
Pick the right tool for each step instead of committing to one stack. Choose one provider for transcription and translation, and assign different voice providers per speaker. Watch your privacy update in real time. It's the same choice you make inside the app.
One provider per step. Voice can use several (a different voice per speaker).
TRANSCRIBESpeech → text
TRANSLATEKeep tone & glossary
Local server here means a server running on your computer. The app also accepts hosted OpenAI-compatible endpoints, which send translation text to their host and may charge for usage.
SEPARATEVocals / background split
Always on-device
VOICEText → speech · pick one or more
Your pipeline
SeparateOn-device
Transcribe
Translate
Voice
PRIVACY
WHAT LEAVES YOUR COMPUTER
WHY THIS MATTERS
Every cloud dubbing tool ships its own voice library, its own translator, its own transcription. If their women's voices sound robotic, you're stuck. If their translator butchers your jargon, you're stuck.
We don't build voices or translators. We wire the best ones together, so you always pick the right tool for the job, swap freely, and never wait on one vendor to fix a model.
EVERY PROVIDER, ONE WORKBENCH
8 providers on Windows · 70+ translation languages · 50+ on-device models
Apple adds a ninth provider in the macOS build: on-device speech recognition (with speaker diarization) and translation on macOS 26+ with supported hardware and languages, and speech with Apple system voices already installed on the Mac (macOS 15+). Apple Intelligence translation needs Apple Intelligence to be available and turned on; Apple Translation uses language packs you install in macOS.
Mac only · Ready for macOS 27
Apple's on-device AI, built in.
On a Mac, transcription, translation and voices can come from the models that ship with macOS. The app uses whatever version the system provides, so on macOS 27 you work with Apple's macOS 27 models. They run on your Mac, need no API key and add no provider bill.
Each Apple model is an ordinary choice in the project settings and the provider tests, so you can mix it with downloadable on-device models and cloud services in the same project.
macOS 26+
Speech to text
Apple Speech, with speaker labels
Apple Speech Transcriber turns the audio into timed segments with confidence scores. Switch on speaker detection in the project and the app labels each speaker with its own on-device speaker models. Missing speech assets for a language are installed by macOS after you approve. Choose the source language yourself; automatic detection isn't available here.
macOS 26+
Translation
Apple Translation
Translates with the languages installed in System Settings → General → Language & Region → Translation Languages. If one is missing, the app says where to add it. It doesn't need Apple Intelligence. Project context and the glossary aren't used by this model.
Apple Intelligence
Translation
Apple Foundation Models
The on-device language model behind Apple Intelligence translates your segments in batches and is given your project context and glossary. It needs macOS 26 or later, Apple Intelligence turned on and ready, and both languages supported by the model. New Mac projects pick it by default when they qualify, otherwise Gemma 4 E2B.
macOS 15+
Text to speech
Apple system voices
Give a speaker one of the Apple voices already installed on your Mac, in whichever default, enhanced or premium versions you have. The app never downloads voices itself; macOS manages them. Personal Voice is never used.
Apple Intelligence
In-app Help
Answers from the help, on your Mac
Ask the Help panel a question and Apple Foundation Models writes a short answer from the bundled English help pages, with numbered sources you can open. Without Apple Intelligence, local help search still works.
Siri Help · macOS 27
Siri & Shortcuts
Search the help from anywhere
The Shortcuts action “Search AI Video Translator help” finds the matching help passages on your Mac and hands them to your shortcut. On macOS 27, ask Siri and the app's Help panel opens with your question.
What it needs, and what stays on your Mac
The Mac app runs on Apple Silicon with macOS 15 or later. Apple speech recognition, Apple Translation and Apple Foundation Models need macOS 26 or later on a supported Mac; Foundation Models also need Apple Intelligence turned on. Asking Siri to search Help needs macOS 27. Which languages each Apple model covers is decided by macOS on your Mac, and the app checks your languages before it runs. Audio, text and help questions are processed on your Mac and never sent to us or to Apple through these models; the app doesn't use Private Cloud Compute. Details are in the privacy policy.
External control
Start a translation from Shortcuts or your AI agent.
The app does the translation. Shortcuts and AI agents request it: the app owns the job, keeps it running on its own, and applies the same file access, consent and free-tier rules as when you click.
Mac only
Shortcuts
Translate a video from Shortcuts
In Shortcuts, give Translate video a video or audio file and a target language. Use Translate video and wait to get the finished file in one action. You don't need to select a project. Closing the window while processing asks you to confirm quitting; choose “Keep processing” to keep the job running.
▪Translate video starts the job and returns it at once, with a dialog confirming acceptance. Check translation, Get translated video and Cancel translation work on that job.
▪Translate video and wait returns the finished file and shows its path in the success dialog. Cancelling the wait leaves the job running; use Cancel translation to stop it.
▪Choose missing settings before running. The app uses its new-project defaults unless you pick a provider, model or voice. A missing choice fails immediately with a clear error; choose it in the Shortcut and run again. There are no runtime questions.
▪The file lands in Downloads. The result is saved there once, named with “_translated_”, the target language and the actual exported container extension, and a notification appears if notifications are allowed for the app.
▪The free tier works as in the app. Export keeps the full video duration with the first-10-segment dubbing allowance. Switch on “Full dub required” to refuse an over-limit run before translation or voice generation.
▪Consent stays in the app. Give consent for external AI services in the app before running the Shortcut.
Mac & Windows
MCP for AI agents
Connect your AI agent
MCP is off by default. Turn it on in Settings → MCP first, then copy your host connection setup in that tab. Your agent host launches the app with --mcp and talks to it over stdio. Setup is documented for Claude Code, Claude Desktop, Cursor and VS Code with Copilot.
▪Discover, prepare, start. The agent can inspect transcripts, settings, speaker choices and source frames, apply free text/timing/settings edits, and use shared undo. Explicitly requested processing can start app-owned translation, voice, retranscription or export jobs; cloud providers may charge and results are not undoable.
▪Follow and cancel. It can wait for progress, cancel the job, and search the app's Help when something fails.
▪Keys never leave the app. Provider API keys stay in the app and are never part of the agent's configuration or chat.
▪A path from chat is not a file grant. If the app can't reach a file or folder, you choose it with Browse in the app, then the agent tries again.
▪Workflow built in. The server hands the agent the translation workflow on connection.
Claude Code · macOS
claude mcp add --transport stdio --scope user ai-video-translator -- "/Applications/AI Video Translator.app/Contents/MacOS/AI Video Translator" --mcp
Claude Code · Windows PowerShell
claude mcp add --transport stdio --scope user ai-video-translator -- "$env:LOCALAPPDATA\Microsoft\WindowsApps\videotranslator.exe" --mcp
Open the app and finish setup first, then check the connection with /mcp. Other hosts that can run a local stdio server may work too, but compatibility isn't promised for every client.
What the requesting assistant receives
Siri receives help passages when you search Help. Shortcuts or an MCP host receives help passages, your choices, job status, file paths and the location of the result file. MCP hosts can also receive full transcripts, project settings, speaker choices, timing, audio status and extracted source frames as JPEG images. Enabling MCP permits reads, free edits and explicitly requested processing on all registered accessible projects, with no per-project permission. Paid operations need your explicit request and must never be retried automatically. An MCP result carries the path, type and size, never video bytes, and reports warnings such as the free-tier limit. Local translation doesn't mean the assistant's own processing is local: Apple handles Siri and Shortcuts under Apple's terms, and an MCP host and its assistant under their own terms and privacy policies. We receive neither the requests nor the results. See the MCP setup guide, the Shortcuts and Help guide and the privacy policy.
Real editor
You edit every segment.
The Café conversation timeline shows English source text beside German translations, with a different provider voice on each visible segment. Three voiced segments show green play buttons, durations and speed badges.
▪Confidence per segment. Low-confidence parts get highlighted, so you review only what matters.
▪Speed warnings before render. Catch >130% speed-ups when text won't fit the slot.
▪Per-speaker provider and voice. Each speaker gets its own provider, voice and speed. Set it once and it applies across every segment.
▪Glossary & project context. Tell the AI what your project is about. Names stay names. Jargon stays jargon.
▪Voice emotion tags. [whispers], [laughs], [slow]: studio-grade direction where the model supports it.
▪Background-mix slider. Don't kill the room tone. Bring it back in.
Try before you trust
Pick the right tool for the job.
Every video raises the same questions. Which translator keeps my jargon? Which transcriber handles a noisy interview? Which voice sounds right for this character? Which separator gives the cleanest vocals?
Most dubbing apps ask you to commit before you find out. We don't. Four provider comparison tests let you A/B providers on your actual source material, side by side, with your own keys (no markup) or fully offline. A CPU / GPU Test also compares local translation on Windows and in the macOS build. Tap any preview to enlarge.
In the macOS build, every on-device translation model (Qwen3.5, Qwen3.6, Gemma 4, Hy-MT2 and Index-Translate) uses Metal by default, with CPU as an alternative and fallback. GPU acceleration here applies to these translation models; other on-device stages use their own engines.
Tests run with your own API keys for cloud providers, or fully offline for on-device models. No subscription required. The 10-segment dubbing allowance applies only when saving the final video.
Pricing
Two plans. No tricks.
All features available on both. The only thing the subscription lifts is the 10-segment dubbed-voice limit.
Free
Free to download
All features. Full-length video and background audio, with dubbed voice for the first 10 speech segments.
✓ Every provider, every voice, every language
✓ On-device models, fully offline and free
✓ Multi-speaker, source separation, glossary
✓ Commercial use allowed
✓ No watermark
— Dubbed voice limited to the first 10 segments
Mac App Store · Coming soon
Premium
Premium
Subscription
Everything in Free, with dubbed voice for every speech segment. That's the only difference.
✓ Dubbed voice for every segment
✓ All features, all providers
✓ On-device fully included
✓ Cloud providers via your own API keys
✓ No token markup, ever
✓ Cancel anytime
The subscription is purchased in the store you download the app from, and the price is shown there.
Mac App Store · Coming soon
Enterprise
More than the app? Let's build it.
The desktop app handles most dubbing work out of the box, and Free or Premium is all most people ever need. Enterprise is custom work we do with you when you're running translation at scale, have strict data-handling rules, or need models and pipelines that don't exist yet. There's no fixed package. We scope it to what your project actually requires.
▪Custom development. New features, file formats, or pipeline steps built around your workflow.
▪Private & on-device models. Wire in additional offline, fine-tuned, or domain-specific models, including your own.
▪Translation infrastructure. Help designing and standing up dubbing & localization pipelines for a team or platform.
▪Batch & automation. High-volume, scripted, and headless workflows beyond one video at a time.
▪On-prem & air-gapped. Fully local deployments for sensitive content that can never touch the cloud.
▪Priority support & training. A direct line to the people who build it, plus onboarding for your team.
Tell us what you're trying to do, even if you're not sure it's possible yet. We'll give you a straight answer.
How AI provider costs work
If you pick cloud providers (OpenAI, xAI, ElevenLabs, Google, Inworld, Soniox), you bring your own API key and pay them directly, at their public rates, with no markup, no orchestration fee, and no per-minute charge from us. Provider fees depend on the models and usage you choose. Stay on-device and you pay no one: the models run on your machine for free.
What this is
A desktop tool with deliberate trade-offs.
Every product makes choices. Here are the ones behind this one, so you know what you're getting before you install it.
WHAT IT IS
✓A desktop app, not a web service. Your video files live and stay on your machine. We never upload them anywhere.
✓Provider-agnostic. Use OpenAI, xAI, ElevenLabs, Google, Inworld, Soniox, your own OpenAI-compatible endpoint, or fully offline models. Mix them freely inside one project.
✓Your AI bill, your AI keys. Cloud providers are paid directly with your own API keys. We don't resell tokens or add markup.
✓A real segment editor. Waveform, timecodes, confidence scores, speed warnings, draggable segments. Built for review, not one-click magic.
✓Flat subscription. Lifts the 10-segment dubbing limit. No per-minute charges from us. Bought in the store, where the price is shown.
✓Free tier is generous on purpose. Every feature, every provider, every voice. No watermark. Full-length exports with dubbed voice on the first 10 segments.
WHAT IT ISN'T
✕Not a voice cloning tool. We don't clone voices from your source. You can use clones already created on your provider's account, but the app itself doesn't do cloning.
✕Not a lip-sync tool. We dub the audio track. The video frames stay untouched, so mouth movements remain as filmed.
✕Not a subtitle tool. Dubbing only: no subtitle export, import, or embedding. A possible future addition.
✕Not for music or songs. Dialogue, narration, podcasts, interviews and tutorials work well. Sung performances aren't what AI dubbing handles today.
FAQ
Common questions.
Can I use the free version for commercial videos?
Yes. The free tier has no commercial-use restrictions. Every export keeps the full video and background audio; dubbed voice is limited to the first 10 speech segments. Generated audio, tests and comparisons are all usable in monetized content.
Do you put a watermark on exported videos?
No. We never watermark output, on any tier, free or paid.
Do you store or see my video files?
No. The app runs on your computer. Video files never leave your machine. If you choose cloud providers, audio for transcription and text for translation or speech synthesis are sent directly to them, never the video file itself. With on-device models, all media stays local; optional pseudonymous usage statistics are sent only after explicit consent.
What does the subscription unlock?
Dubbed voice for every speech segment in the final export, instead of only the first 10. That is the only difference between Free and Premium. Both tiers export the full video and background audio, and all features, providers and languages are available on the free tier. The subscription is purchased in the store you downloaded the app from (Mac App Store or Microsoft Store), and the price is shown there.
Do I need to pay AI providers separately?
Cloud providers (OpenAI, xAI, ElevenLabs, Google, Inworld, Soniox) are billed directly through your own API keys, with no markup from us. Downloadable on-device models have no API charge. Apple speech recognition, translation and system voices in the macOS build also require no API key. A server you run locally, such as LM Studio or Ollama, has no provider API charge; hosted endpoints may charge under their own terms.
Is there a Mac version?
Yes. It is coming soon to the Mac App Store; the Windows version is on the Microsoft Store. The Mac version is built for Apple Silicon and requires macOS 15 or later. On macOS 26+ with supported hardware and languages, Apple on-device speech recognition (with speaker diarization), translation through Apple Intelligence where it is available and turned on, and Apple Translation with language packs you install in macOS are also available. On macOS 15+, Apple system voices already installed on the Mac can speak the dub. The in-app Help can answer questions on-device, and can be searched from Shortcuts and, on macOS 27+, by asking Siri. Linux, mobile, and web are not on the roadmap by design. Large video files and AI inference belong on a desktop, not a browser tab.
Does the Mac app use Apple Intelligence and the macOS 27 models?
Yes, where your Mac supports them. The app uses the models built into macOS, so on macOS 27 it works with Apple's macOS 27 models. On macOS 26 or later with a supported Mac you can choose Apple speech recognition, with optional speaker labels; Apple Translation, with translation languages you install in System Settings; and Apple Foundation Models, which needs Apple Intelligence turned on, for translation and for on-device answers in the in-app Help. Apple system voices already installed on the Mac can speak the dub on macOS 15 or later. Help can be searched from Shortcuts and, on macOS 27, by asking Siri. All of this runs on your Mac with no API key, and each model covers the languages macOS provides for it.
Can Siri translate a video?
Asking Siri to translate a video will follow once Siri supports it. Today, use the Shortcuts app: Translate video starts an app-owned job and confirms acceptance; Translate video and wait returns the finished file and shows its path. Choose any missing provider, model or voice in the Shortcut before running; there are no runtime questions. The file is saved once in Downloads, with a completion notification if notifications are allowed. Free exports dub only the first 10 speech segments; “Full dub required” refuses an over-limit run before translation or voice generation. Consent for external AI services is given only in the app.
Can I control the app from Claude or another AI agent?
Yes, on Mac and Windows. MCP is off by default: turn it on in Settings → MCP first and copy your host setup there. The app has a local MCP server: your AI host launches the app with the --mcp argument and talks to it over stdio. Setup is documented for Claude Code, Claude Desktop, Cursor and VS Code with Copilot. An agent can list languages, providers, models and voices, inspect full project transcripts, settings, timing and audio status, see source-video frames as JPEG images, apply free text, timing, settings and speaker edits, and use shared undo history. When you explicitly ask, it can start translation, voice generation, retranscription or export jobs, monitor or cancel them, and search Help. These processing calls can cost money with cloud providers, start without a GUI confirmation and are not undoable; assistants must never retry paid work automatically. The app does the translation itself and keeps the job running independently of the agent. Compatibility with every MCP client is not promised.
Does an AI agent see my API keys?
No. Provider API keys stay in the app and never cross the MCP or Siri and Shortcuts channel. A path typed into chat is not a file grant: if the app cannot reach a file or folder, you choose it with Browse in the app. External-AI consent and the free-tier export rules still apply. The agent can receive full project transcripts, context, settings, speaker voice choices, timing and audio status, as well as help passages, choices, job status, file paths and the result file location. While MCP is on, connected hosts can inspect and edit all registered projects accessible to the app, including unopened projects; there is no separate permission per project or host. The get_frame tool also returns source-video frames as JPEG image bytes. Audio and exported video remain file-backed; the assistant's own terms and privacy policy govern returned content.
How many languages are supported?
70+ for translation — that is the list the app offers as a source or a target language. Transcription and text-to-speech coverage depend on the provider, model and voice. Apple speech recognition, Apple Translation and Apple Foundation Models in the macOS build each support a narrower set of languages, decided by macOS on your Mac.
Why no mobile or web version?
We focus on desktop so large video files and supported on-device models can be processed locally. Linux, mobile and web versions are not on the roadmap.
Can it translate music or songs?
Realistically, no. The app is built for spoken-word audio: dialogues, narration, podcasts, interviews, tutorials. Musical performances have rhythm, melody, and emotional delivery that current AI dubbing pipelines aren't designed to preserve. Source separation can isolate vocals from a backing track, but the result won't replace a song. If your video has singing as a key element, expect to handle it separately.
Does it support voice cloning?
Not directly. The app itself does not clone voices from your source audio. If you've already created a voice clone with a provider that supports it (for example, an ElevenLabs custom voice on your account), that clone shows up as a selectable voice when you use that provider. So cloning is possible through your provider's account, not as a feature inside the app.
Does it lip-sync the dubbed video?
No. The app dubs the audio track but never modifies the video itself. The speaker's original mouth movements stay as they were filmed. For tutorials, podcasts, narration, training videos, and most voiceover work this is a non-issue. If pixel-perfect lip-sync to dubbed audio is essential, a separate AI lip-sync tool can be layered on top of the rendered output.
Does it generate or import subtitles?
Not today. The app is focused on dubbing: replacing the audio track in another language. Subtitle export, import, and embedding are not in scope in the current version. They are a candidate for a future release but with no committed timeline. For subtitle-centric workflows, a dedicated subtitling tool or a CAT tool with subtitling support is currently a better fit.
Are there alternatives?
Yes. Cloud dubbing services include HeyGen, Rask AI, ElevenLabs Dubbing, Synthesia. Offline desktop apps include AutoDub (Mac). Open-source DIY options include VideoLingo, pyVideoTrans, KrillinAI, Bluez-Dubbing. Professional CAT tools for text translation (not dubbing) include Trados, memoQ, Phrase. Each takes a different angle on price, hosting, control, and use case.
Roadmap
Where it's headed.
We ship deliberately. Here's what's live, what's next, and what we're choosing not to build.
SHIPPEDNow
Windows app
Live on the Microsoft Store. All Windows features, with eight providers, on-device or cloud, and never a watermark.
SHIPPEDNow
macOS app
Coming soon to the Mac App Store, for Apple Silicon and macOS 15 or later. The same private, mix-and-match dubbing workflow you get on Windows, plus Apple on-device speech recognition, translation and system voices.
EXPLORINGLater
Subtitles
Subtitle export, import and embedding are a candidate for a future release. On the table, no committed date.
Not on the roadmap, by design
No Linux, mobile, or web version. Multi-gigabyte video files and on-device AI inference belong on a desktop, not uploaded to someone else's server or squeezed into a browser tab. Staying desktop-first is what makes the privacy guarantees real.
Ready to dub privately?
Free to download. Try every feature, every provider, every language. Subscribe in the store only when you need more than the first 10 segments dubbed.
Mac App Store · Coming soon
macOS · Windows
Cookies & analytics
We use privacy-respecting analytics to see how the site is used, and tags for marketing measurement. You choose what runs. Read our Privacy Policy.
