Was able to improve multithreaded perf by another 25%! 4K decode went from 182->227 FPS, 1080p decode from 638->803

I used to genuine believe that the 7 in CR7 stood for "istiano". Like i18n or a11y

I definitely cooked with this one. Library will be dropping soon!

Me: please draw a solid green rectangle Safari:

My ProRes decoder now also works in Firefox and Safari! Since it's written in WASM, it technically runs everything out of box. But the problem was that Firefox and Safari only support a limited range of pixel formats, meaning I had to add conversion logic to the decoder.

All format conversions in my ProRes decoder are working like a charm ☑️

Who is putting these cringe comments into my code?? (I am the sole developer of this project)

You also really can't trust LLVM to emit the proper instructions even if they exist, so I'm forced to write shit like this (inline assembly) Truly no way around handwritten assembly in the decoder space, maybe FFmpeg was right all along

Beta-test the Pikimov's particles system before it's released, and share your feedback on the current implementation www.patreon.com/pikilipita/p...

Bluesky message limit is too short, so here's a screenshot of my tweet because I thought the content was worth sharing here!

What kinda unit is 15^(25/lb)

The next version of Mediabunny fixes an oversight in the HLS request resolution logic: redirects! When a playlist request gets redirected, any subsequent subrequests must be relative to the *new* URL, not the original one.

Instantly seeking through a 4K ProRes HQ video in the browser, zero lag

This one's giving cosmic horror almost

My custom WASM ProRes decoder is able to decode 8 (eight!!) simulatenous 30FPS 1080p ProRes HQ streams in the browser. Overkill for most people, but a great stresstest! There's still some choppiness, but that's caused by the canvases and not the decoder. I'll investigate.

I use this script for all of my TypeScript libraries. It goes through the entire project and throws if it can find any symbol that does not have a docblock attached to it. This FORCES me to document every last thing, and if I don't, the deploy fails.

Learned about "Cross-Origin-Embedder-Policy: credentialless" today and it is amazing!! It enables powerful features like shared-memory multithreading WHILE still supporting cross-origin resources like images or videos! Only condition is that those don't require cookies to load.

The final JS bundle looks absolutely fucking mental (it is valid JavaScript!!)

I am "On Twitter" (I prefer calling it X) in the sense that I use it to share my work. As a self-employed open-source dev living mainly off donations, this is basically what I *have* to do because of the network effect there. I never browse the feed though, it usually just makes me feel worse.

Another instance of Google sucking. There's a much better match for this in the current Vite docs, but it decides to show me the v3 docs instead??

For my ProRes decoder library, I'll be trying out "errors as values" for the first time in TypeScript!

I finally made a proper website for Fragment 🎉 Right now, it is mostly a wrapper to the documentation available on GitHub but this is only the beginning! fragment.tools Built on top of @vite.dev, @svelte.dev and Mediabunny by @vanilagy.bsky.social

Next one

This one is wild

I really like this one

After adding shared-memory multithreading to my ProRes decoder, it now decodes a 130-frame 1080p video in ~200ms, right in the browser. This is 3x faster than multithreaded NATIVE FFmpeg, and 60% faster than Apple's hardware ProRes decoder. Relentless optimization pays off :)

This is one of those areas of software where bugs lead to kinda cool-looking results

A simple parallax demo, available now in Pikimov as a project template 🌸 pikimov.com 🌸 #parallax #motiondesign #anime

Got it down by another millisecond!

I started this project primarily to learn how to write optimized decoders, so almost all of it is written by hand without AI assistance. It's been insanely fun. There are some really cool optimization tricks involved; they would make for some banger blog posts for sure 👀

I've been working on an experimental, fully-custom ProRes decoder optimized for in-browser usage, and right now I'm about 3x faster than FFmpeg's decoder compiled to WASM! I'm certain I'll be able to bring this down to <2 ms per frame with threading.

It's been almost a year since V8's last blog post. Not sure what happened, but I hope they'll pick it up again.

1M soon!

Mediabunny fell off 😭

Why is it that Bun is talked about so much, yet I hear barely anything about Deno these days? Deno's even already written in Rust! (lol) Is it better marketing? The "backwards compatible" approach of Bun? What is it?

Mediabunny v1.45.0 now works basically everywhere 🚀 With the new @mediabunny/server extension, you get access to super-fast, hardware-accelerated media decoding and encoding on the server. Inspect, process and transform media on the server without ever leaving TypeScript.

@mediabunny/server is done and will be dropping later today. Here's the README for anyone who's interested in reading more: github.com/Vanilagy/med...

I love this so much This thing literally has O(1) memory usage. Media data is streamed in, gets processed, and is written to disk. None of it is accumulated anywhere. Even the uploader halfway across the globe is automatically slowed down due to stream backpressure!

This is all the code that's needed to compress a video on the server while it's being uploaded Right now it's being written to a file, but you could also stream it to S3 or anywhere else

Doing a performance pass right now. It's able to transcode 1080p at ~10 real time speed, while the main thread is basically almost idle during that time!! That's neat for Node applications where you don't wanna be blocking the main thread much.

Gonna support transparent VP9 as well!

It's a shame many people will talk to LLMs and not question the validity of the output. Hallucination is far from solved y'all! (I wish Claude was right tho)

Audio's done too! Super excited for this. Now, the last step is provide alternatives for all other corners of the codebase where the DOM is expected.

Video half of the upcoming @mediabunny/server is done! With this, Mediabunny will have full access to hardware-accelerated decode/encode in all server-side environments (Node, Bun, Deno). Gonna be huge 🚀

I've reported 16 Chromium bugs in total by now, kinda wild. Most of those related to WebCodecs!

Masterpiece

I swear it feels like Mediabunny is cursed on Google Search My HLS blog post is both linked to in the sitemap and is reachable through the /blog index, which Google has successfully indexed! So, what they're displaying makes no sense. Anybody have experience with debugging this sorta stuff?

YOOOO! Mediabunny's API is in GPT 5.5's weights!! That's awesome

Mediabunny v1.43.0 now supports encoding "studio quality" 24-bit FLAC directly in the browser

HLS added around ~32 kB to Mediabunny's bundle size 😫 The cool thing: if you don't use it, it'll just get treeshaken away!!