🎉 Mediabunny has hit 2M monthly downloads!! Next goal: 1M weekly.

API design is so freaking hard. Over the course of adding HLS support, I've redone the new API about 5 or 6 times by now. Sometimes you literally first need to see how Variant A feels before you know Variant B was better all along. And knowing you can't change it after release!

You follow the news. I follow the news using my own media player. We are not the same

The long road to full HLS read/write support in Mediabunny is finally coming to an end. Last step for me is the documentation stage; write insightful docs for new symbol exported by the lib. Always takes a while to do, but it's also quite fun!

I mostly maintain a "all bugs have infinite priority" policy for Mediabunny, which definitely feels like a double-edged sword to me. Yes, it means that the library is quite stable, but it can be annoying as it can rip me out of feature development.

Thanks to @techconnectify.bsky.social I now have unwavering confidence that my dishwasher will be able to clean this without a problem

This is a good example of why indirection leads to worse performance; both of these functions do the exact same loop, but this fact is hidden by the function call indirection. Had these been inlined, the repeated logic would've been immediately obvious.

Tried this out on a work project today instead of ffmpeg-wasm and the speed improvement was insane. If you need to convert/splice video on the web, don’t sleep on this library!

Here's Mediabunny capturing 5 seconds of live TV and transmuxing it into an MP4 Total bundle size to do this is around ~50 kB

Mediabunny on the left playing German live TV (with WebCodecs) compared to the official stream on the right. I'm even ahead by a few seconds!! The coolest thing is that I didn't even have to change the media player code; it just thinks it's playing a normal video file.

The upcoming Mediabunny version can play live streams!! Here I'm watching German live TV.

It's good that GPT 5.4 is so bad that it doesn't even hurt cancelling the subscription

The clanker is now a Mediabunny "contributor"😩

Any project I do eventually needs this utility function. So stupid that you need to hack it with a Worker.

Today I learned you can do this

Mediabunny v1.37.0 can now encode FLAC directly in your browser! 🚀 Huge for exporting audio losslessly. Just add this new extension package, 83 kB gzipped. www.npmjs.com/package/@med...

I've released @mediabunny/aac-encoder, a new extension package that polyfills AAC encoding support for environments that don't have them. Firefox for example! Or anything on the server, like Node or Bun! 🚀 Around ~260 kB gzipped in size.

The creators of Goat Simulator should make a factory automation game, I'm sure it'd be decent

it always hits different whenever this digit of the unix timestamp increments

Mediabunny will support recursive .m3u8 files. Cursed? Yes. Violates the spec? Yes. Useless? Yes. Will it work? Also yes.

Which naming do you prefer? (JavaScript context) onthing: (thing: Thing) => void; onThing: (thing: Thing) => void; I find the second more readable, but the first mirrors the practice used by all builtin JS and Web APIs.

Mediabunny is now 20x smaller!! 🚀 (due to a bug in Bundlephobia)

If I didn't scrutinize every line written by Opus 4.6 or GPT-5.3 Codex, Mediabunny's codebase would quickly rot into a tangled mess of unconfident, unjustified and overcomplicated code. "Shipping at inference speed" would literally require that I give up care for the project.

Fucking TLDRAW uses Mediabunny!!

Microsoft's Clipchamp is using Mediabunny!! 🔥

Mediabunny v1.35.0 adds support for reading AND writing videos with non-square pixels 👀 Sounds niche, but it's actually fully supported by WebCodecs and the lack of Mediabunny's support for it was a blocker for some. Well, not anymore! 🚀

I could never structure my software like this. Each of these files is like 6 lines long. I can see why you would *think* this is a good (or at least natural) structure, but to me, working with this would be a nightmare.

It's interesting that LLMs routinely get this wrong despite being otherwise perfect at grammar

Wouldn't it be possible for search engines to completely remove the idea of "SEO" by using new AI tech? As in, to finally realize the true vision of search engines: the *best* page ranks first. Not the most optimized or keyword spammy, but simply the best.

Mediabunny's codec support is growing beyond what is inclued in WebCodecs. To make sure that the expected formats of all of the codecs are explicitly defined, I recently created the Mediabunny Codec Registry which does just that! Over time, it'll grow as more codecs are added.

If I ever make a programming language I won't include > and >= operators

The upcoming HLS support in Mediabunny requires streaming AES decoding. Including the tables of constants in the code would blow up the bundle size unnecessarily, so instead, I generate the tables at startup using a bit of math! Entropy at play; exact same data, 10x smaller bundle.

Find the bug

HLS final boss

Today I was facing the problem that a Worker I keep around for performance reasons prevents a Node/Deno/Bun process from self-terminating. Instead of having to terminate the Worker manually, there exists Worker.unref(), which gives the runtime the permission to exit. Neat!

Huge news! Screen Studio, who need no introduction, are Mediabunny's next gold sponsor, donating $1k/month! 🚀 Hugely appreciated. You guys have upped the screen recording game, and I'm excited to work together!

I created a new extension package for Mediabunny, @mediabunny/ac3! 🚀 This one provides decode & encode support for AC-3 and E-AC-3 right in the browser, meaning you can work with a lot of broadcast media without conversion. ~260 kB of gzipped WASM. www.npmjs.com/package/@med...

Mediabunny has exceeded 1 million monthly npm downloads holy shit Never expected this to happen this fast, but here we are. It's just a number at first, but it's kind of surreal when you picture what it represents. Thank you! ❤️

You can tell Claude Code is written in React because it's slow and buggy

Mediabunny v1.31.0 ships with ID3 metadata read/write support for ADTS (.aac) files, and has a more resilient MPEG-TS demuxer. 🚀

I live in crazy times where I get emails like this

When the debug videos are so good you forget you're working on Mediabunny and just start watching them

Mediabunny's now at ¾ of a million monthly downloads!!! (almost) 🎉 Time to treat myself to some malatang

Finally added this to Mediabunny: a warning if the library is bundled twice. This is almost always unintentional and leads to runtime errors, so it's good if this mistake is surfaced early. Using a symbol on the global object for the detection.

Mediabunny has world-class seeking performance across all file formats. On the left, VLC *trying* to seek through a .ts file, it just kinda gives up. On the right is Mediabunny. It just works.

I find that LLMs often struggle with writing *confident* code that fully understands all invariants at play. This code will work, but the highlighted condition is unnecessary (always evaluates to false). I always try to catch these cases to make the code tighter.

A huge milestone: Mediabunny v1.29.0 now ships with read & write support for MPEG Transport Streams! 🎉 This format is commonly used in live broadcasting. Quite proud of the implementation. It supports AVC, HEVC, AAC, MP3, O(log n) seeking, and it's very fast. HLS is next!

Mediabunny reached 5k GitHub stars!

Does anybody know if V8 (or other JS JITs) use hidden classes for common tuple types? So let's say I use a lot of [number, string, boolean], will these get optimized to a tight 16-byte struct or will these remain the slow, unoptimized array type?

B-frames are truly a mess with how out-of-order they can be; here I need to read ahead by SIX packets just to correctly compute the duration of the packet with timestamp 10.1. Very annoying! I might use the bound given by max_num_reorder_frames in the future.