←Back to contents·The Bitstream·Vol.1GitHub

Inspect everything.

Every media file carries its own dossier — duration, codecs, dimensions, rotation, bitrate, cover art, even the GPS point your phone quietly wrote into it. Mediabunny reads it all by touching only the bytes that matter, so a 5 GB file answers as fast as a 5 MB one. Drop one in and watch the byte counter; nothing is ever uploaded.

Fig. 03 — The Lab ReportReady

Source— drop a file —

EngineMediabunny

Bytes read—

Drop a file — or click

MP4 · MOV · MKV · WEBM · MPEG-TS · MP3 · WAV · OGG · FLAC — stays on your device

Bring your biggest file. 5 GB inspects as fast as 5 MB.

Duration, codecs, bitrate, rotation, cover art, VFR, GPS leaks, truncation — read from the header bytes alone, even on a 5 GB file.

Fig. 04 — Why Mediabunnyvs the usual alternatives

Live — measured on your file, just now

Inspect a file above and these fill with real numbers from your own device — bytes read, percentage of the file touched, and time to the full report. No synthetic benchmark.

Mediabunny vs the usual ways to read metadata

	Mediabunny	mediainfo.js	Server ffprobe	<video> element
Where it runs	Your browser · zero WASM	Your browser · WASM	Your server	Your browser
Download to start	~156 KB gz ¹	~2.6 MB wasm ²	—	0 · built-in
Uploads your file?	No	No	Yes — the whole file	No
A 5 GB file	ms — reads only header bytes	OK — chunked reads	Upload first (~13 min ³)	Loads, but tells you little
What you get	Duration, codecs, fps + VFR, bitrate, rotation, HDR, tags, cover art, GPS check, truncation probe	Very deep field list	Everything ffprobe knows	Duration + dimensions only
Can-decode check	Yes — and it decodes a real frame to prove it	No	No	canPlayType() guesses
Cost	Free	Free	$ server time	Free

¹ Full library, gzipped, before tree-shaking — same measurement as the Convert demo; a metadata-only import is far smaller. ² MediaInfoModule.wasm 0.3.7 as served by unpkg, 2,565,723 bytes, measured 2026-06-11. ³ Uploading 5 GB at 50 Mb/s — arithmetic, your connection may differ. Remaining cells are qualitative.

The receipt is live: the counter and the strip above are wired to the exact byte ranges Mediabunny requested from your file.

The source — the entire inspector, nothing hidden

Fig. 03a — Files

inspect-demoread-only

Fig. 03b — CodeTS

inspect-demo/src/inspect.ts

1import { Input, ALL_FORMATS, BlobSource } from 'mediabunny';
2import { describeTrack } from './tracks';
3import { tailProbe, privacyScan } from './analysis';
4import { makePoster } from './poster';
5 
6// Inspect any media file — reading only the bytes it needs.
7export async function inspect(file) {
8  let bytesRead = 0; // the receipt — watch how little gets read
9  const source = new BlobSource(file);
10  source.onread = (start, end) => { bytesRead += end - start; };
11 
12  const input = new Input({ source, formats: ALL_FORMATS });
13  try {
14    const tracks = await input.getTracks();
15    const report = {
16      format: (await input.getFormat()).name,    // container
17      mimeType: await input.getMimeType(),       // full codec string
18      duration: await input.computeDuration(),   // seconds
19      tags: await input.getMetadataTags(),       // title, cover art…
20      tracks: await Promise.all(tracks.map(describeTrack)),
21    };
22    report.integrity = await tailProbe(input, tracks, report.duration);
23    report.poster = await makePoster(tracks);  // decode before dispose
24    report.privacy = privacyScan(report.tags); // GPS, device, dates
25    report.bytesRead = bytesRead;              // vs. file.size
26    return report;
27  } finally {
28    input.dispose(); // frees demuxer state + pending reads
29  }
30}

The real source — click any file on the left. This exact code produced the report above. No WASM, no server, no upload.