Why Local-Only Audio Processing Matters

"Your audio never leaves your device" sounds like a privacy slogan. It is mostly an engineering decision, and it changes what the tool can do.

Published 2026-09-10 · Privacy · Web Audio · Local-first

Most online visualizer tools work the same way: you upload a file, a server renders a video, and you download the result. It is a reasonable design — rendering is expensive and servers are good at it — but it has consequences that show up as queues, length caps, watermarks and a terms-of-service page nobody reads. Doing the work in the browser instead is a different set of trade-offs, and it is worth being clear about which ones.

What "we process your audio" actually means

When a track is uploaded it becomes a file on somebody else's computer. Reasonable services delete it eventually; the question is what happens in between, and the honest answer is that you cannot verify it. For an unreleased master, a client's stems, a demo under NDA or a recording of a live service, that uncertainty is a real cost. It is also usually accompanied by a licence clause granting the service the right to store, copy and process your content — necessary for them to function, and still a thing you agreed to.

How the local version works

The browser has had the pieces for this for years. decodeAudioData turns your MP3, WAV or FLAC into samples. An AnalyserNode hands the page an FFT and a waveform buffer every frame. The scene is drawn to a canvas. MediaRecorder — or, for a cleaner result, an offline pass that renders and encodes frame by frame — turns that canvas plus your audio into an MP4 or WebM. Nothing in that chain requires a server. A file you drop on the page is read directly from disk by your own browser; a microphone or an audio interface is a live stream that is analysed and discarded.

The same is true of the parts that sound like they must be remote. The lyric transcription runs a Whisper model inside the tab: the model is downloaded once, about 60 MB, and cached, and then your audio is transcribed on your own hardware. The phone remote is the only feature that touches a server at all, and what crosses it is control messages — change scene, black out — not audio.

What it buys you

1. No queue. Nothing is waiting behind other people's renders. Export speed depends on your machine and nothing else.
2. No length limit. A four-minute single and a ninety-minute DJ set are the same operation. Server-side tools cap length because length costs them money.
3. No upload wait. A 60 MB WAV is analysed in about a second; uploading it would take longer than that on most connections.
4. It works offline. Once the page is loaded, a plane or a venue with no Wi-Fi is fine.
5. No watermark economics. Watermarks exist to sell the removal of watermarks. When rendering costs the operator nothing per video, there is nothing to meter and no reason to brand your export.
6. Privacy without a promise. The strongest privacy guarantee is not a policy — it is a missing feature. There is no upload endpoint to trust.

What it costs

Being honest about the other side: your machine does the work, so a slow laptop renders slowly, and a very heavy 4K scene will ask more of it than a server farm would. Browsers differ — Chrome and Edge produce MP4 comfortably, Safari may hand you WebM instead. A tab that loses focus gets throttled. And there is no library in the cloud: your presets and projects live in your browser or in a .bloob file you save, so clearing site data clears them.

Those are real limitations. They are also the kind you can see and work around, rather than the kind you discover when a service changes its pricing.

How to check any tool for yourself

You do not have to take a claim like this on faith. Open the browser's developer tools, go to the Network tab, load a track and export a video. If the audio were being uploaded, you would see it: a large request going out, and a long wait that is not your CPU. Watching the network panel stay quiet while a four-minute video renders is more convincing than any privacy page, including this one.

Try it with any tool you are considering. It is a thirty-second test, and it tells you exactly where your music is going.