Copy of fractal stream video creations: Live Generative Audiovisual System

Fractal Stream — Live Generative Audiovisual System Real-time generative visuals and a self-recombining audio mix, built in the browser and streamed to flat or 360° displays. Free and open-source. *GitHub:* [github.com/m8w](https://github.com/m8w) · *YouTube:* [@h92o](   / @h92o  ) --- A live, never-repeating audiovisual stream. The image is several generative visual engines blended into one frame; the sound is four of my own YouTube videos mixed together at random. Nothing is pre-rendered — every frame and every audio combination is generated on the fly. The visual layer — three engines composited in real time The picture you see is a single frame built from multiple generative sources running at once: *Fractal Stream* — a generative visual program built from the ground up with Claude AI, running on **Three.js / WebGL**. GPU shaders render evolving fractal and metaball geometry: particle trails, iridescent surfaces, continuously morphing form. *Butterchurn ×2* — Butterchurn is the WebGL reimplementation of Milkdrop. I had Claude extend it to load and cycle *Milkdrop `.milk` presets* directly, and I run **two independent instances side by side**, each driving its own preset field. Those two Butterchurn instances are *composited additively* into the same frame — their light values sum, so the two preset fields interpenetrate and react together rather than simply stacking. Fractal Stream renders into that same composite, giving **three live visual layers blended into one image**. All of it is audio-reactive: Web Audio FFT feeds the preset and shader parameters, so the visuals move with whatever's playing. The audio layer — a 4-lane random mixer The soundtrack pulls the *audio from my own YouTube catalog* and runs *four videos at once* through a randomized mixer. Sources are selected and blended on the fly, so four streams are always overlapping and the combination is *never the same twice* — a continuous, self-recombining mix rather than a fixed playlist. 360° output pipeline The live composite is authored as a *3-panel rectangular mix**, then remapped into a **seamless equirectangular projection* (the standard 2:1, 360°×180° format). The panel edges are wrapped and blended so there's *no visible seam at the meridian* — the three rectangles dissolve into one continuous sphere. That single source then drives two viewing modes: *360° screen / dome* — the equirectangular frame is re-projected to fit the display geometry. *Navigable 360° video* — with spherical spatial metadata embedded, players like YouTube 360 render it as an explorable sphere. You physically *move your phone (gyroscope / device-orientation)* to look around inside the stream space. Stack | Layer | Tech | |---|---| | Generative visuals | Three.js, WebGL, GPU shaders | | Milkdrop presets | Butterchurn (×2 instances), `.milk` preset runtime | | Compositing | HTML5 canvas, additive blend | | Audio reactivity | Web Audio API (FFT) | | Audio mixer | 4-lane randomized YouTube-audio mix | | 360 output | Equirectangular projection + spherical spatial metadata | Projects & links Everything is free and open-source. This and the rest of my work — including [`newbold-radio`](https://github.com/m8w/newbold-radio) (the broader four-source catalog mixer) and [`qoix`](https://github.com/m8w/qoix) (QOIX software synthesizer) — live on my GitHub: *→ [github.com/m8w](https://github.com/m8w)* License Released under the [MIT License](LICENSE) — free to use, modify, and build on, with attribution. See the `LICENSE` file for full terms. Copyright © 2026 William Victor Newbold