A real-time music visualizer in the MilkDrop tradition, written in Rust. Seoul captures whatever your desktop is playing, turns it into bands, beats, tempo and spectrum, and feeds that to shader visuals. Each frame is drawn over a warped, fading copy of the frame before it, so the image builds trails and keeps moving.
Download for Windows or macOS (untested) · Project page
Windows 10/11: unzip the release, play some music, and run seoul.exe.
Keep the presets folder beside the exe. The exe isn't code-signed, so
Windows SmartScreen may warn the first time: choose More info → Run anyway.
macOS 14.6+ (untested): the Mac build is produced by CI, but nobody has
run it on a real Mac yet, so reports are very welcome. It's ad-hoc signed,
not notarized, so the first launch needs System Settings → Privacy &
Security → Open Anyway. Then allow system audio recording when asked. Seoul
keeps its presets, settings and screenshots in
~/Library/Application Support/Seoul. Fullscreen is ⌃⌘F, and Cmd-Q quits.
| Key | |
|---|---|
| Space / Backspace | next / previous preset |
| R · A · L | shuffle · auto-advance · lock |
| F · 1–9 · X | favorite · jump to favorite · hide |
| H · F1 · P | help · stats · screenshot |
| F11 · Esc | fullscreen · quit |
seoul.exe --synth plays a built-in 124 BPM test track. Use it when nothing
else is playing. --help lists all the options. Settings live in
seoul.toml and apply while Seoul is running.
cargo run --release # debug builds are too slow for 60 fps
cargo test
.\packaging\package.ps1 # → dist\seoul-<version>-windows-x64.zip
./packaging/package-macos.sh # on a Mac → dist/seoul-<version>-macos-universal.zipSystem audio comes in through cpal's loopback capture. That's WASAPI on
Windows and a CoreAudio process tap on macOS 14.6+. The HUD uses fonts the OS
already ships. Every push to main builds and tests both platforms in GitHub
Actions, and pushing a v* tag drafts a release with both zips.
A preset is a TOML file plus one or two WGSL functions. The TOML holds motion
expressions driven by the audio (zoom = "1.0 + bass * 0.012"), a palette
and a look. The WGSL draws each frame and can also warp the previous one.
Saving either file while Seoul is running reloads it right away. A broken
shader shows its error on screen with the file and line, and the music keeps
playing. See docs/PRESETS.md for the authoring guide.
Seoul runs three threads. The audio callback fills a ring buffer. An analysis thread runs an FFT every 256 samples and writes the results to a triple buffer. The render thread runs the pass chain:
warp → particles → composite → bloom → post → HUD
It renders over a pair of Rgba16Float feedback textures. Transitions between
presets are per-pixel masks stored in the feedback alpha channel, so a
dissolve, a radial wipe or a clock sweep never touches preset code.
docs/ARCHITECTURE.md has the full story.