Demuxed landed in London for the very first time — and we were thrilled to join the celebration as Gold Sponsors!

The whole conference was a true celebration of video tech. The lineup was full of great talks – we particularly loved hearing the talk by Luke Curley titled "MoQ: Not Another Tech Demo" and Rafał Leszko's presentations Realtime Video AI with Diffusion Models. An honorable mention goes to the lightning talk slot, where we had a chance to see multiple 3-min presentations on all things business and video tech. One of the lightning talk slots was occupied by our Software Mansion colleague Łukasz Kita, who discussed combining AI and Prolog to demistify H264 spec. All Demuxed talks will be available on Youtube later on – we'll make sure to link our favorites here :)

For the special Demuxed occasion, our dev team created a super fun demo – a game called Pufferfish, where you join in, transform into one of the round creatures on screen, pucker your lips, and compete against fellow Pufferfish players.

Pufferfish is a silly little game, but it’s made using some serious technology. The whole game was running on Fishjam and Smelter, and all the special effects were made using TypeGPU. All of the tools were made in-house by the Software Mansion team. You can try all of them for free – click on the links and learn more!

During Demuxed, we’ve had a total of over a 100 players (that’s ¼ of the whole conference!), and at times the rivalry was getting as fierce as pufferfish’s spikes. Thanks to all of the video dev gamers who stopped by our booth and spend some great time with us!

Missed seeing Pufferfish in action? Don’t worry – we’ll make sure to show it on other occasions, too! We already have a pipeline full of ideas on how to make it even better – stay tuned for that!

Thank you Demuxed, it was lovely to join you in London. See you in 2026 in San Francisco!

KYUTAI BLOG | Link

Neural audio codecs

Currently, most audio agents use speech-to-text and then text-to-speech models underneath. However, there’s a better approach, which is pure audio-to-audio models. This extensive article explains step by step, how to create audio-to-audio models, giving you a great understanding of the topic.

RTCBITS | Link

us-east-1 - proof WebRTC Architecture

As you probably already know, recently aws’ us-east-1 region had an outage, which caused many services to become unavailable. This may be a good moment to revisit the resilience. Following this short article, you can make sure that you are prepared for such cases.

MOQ DEV | Link

MoQ-powered hangout

Media-over-Quic is still a very hot topic. Last month Luke Curley released a demo app - a hangout room, implemented using MoQ. Even though the technology is not mature, you can make some pretty cool stuff with it.

FFMPEG | Link

August FFmpeg release

Back in August, a new FFmpeg 8.0 version was released, featuring new decoders, Vulkan-based codec support, VAAPI VVC acceleration, and animated JPEG-XL encoding. This release also introduces AI-driven Whisper transcription support, better HDR handling, AVX-512 optimizations, and other enhancements for streaming and modern codecs.

Some time ago, our Software Mansion colleagues working with React Native Open Source came with an idea to bring the same tools available in browsers via the Web Audio API to React Native, offering precise, high-performant audio control and playback. They followed through on their idea and created react-native-audio-api.

By aligning with the Web Audio specification, this project delivers a single API that works seamlessly across native iOS, Android, browsers, and standalone desktop applications. It fills a key gap in the React Native ecosystem: a high-performance API for creating audio, applying effects, and managing parameters like volume on a per-audio basis.

Learn more about how the Audio API works and what you can build with it:

Hi everyone, my name is Feliks. My professional interests are firmly rooted in multimedia processing, the functional paradigm, and the mathematical foundations of computer science, such as combinatorics and algorithmics. Outside the realm of computing, I am a keen enthusiast of Brazilian Jiu-Jitsu, mountaineering, and long-distance trekking.

In my session, I would like to share with you the capabilities delivered by Boombox - an Elixir tool for multimedia streaming and processing that can be seamlessly integrated with AI models.

Together, we will build an Elixir application running real-time object detection models on frames from an external video stream. I will show you step-by-step how to integrate the newest computer vision AI models within the multimedia processing pipeline in order to analyze and modify the stream in real-time.

Prepare for a quick dive into the multimedia world with a pinch of YOLO ;)

Join me online or in person on November 6 (this Wednesday!). Use code Feliks10 and get 10% off on all tickets 💸

We hope you enjoy the RTC.ON newsletter as much as we do. It's great to see all of you joining us each month for a little multimedia walk through – thanks!

Here are some more ways to connect with us:

Want to share RTC.ON newsletter with a friend? Here is a link to our sign up page, including the archive of all past issues.