Skip to content

ROADMAP.md

Latest commit

 

History

History
527 lines (431 loc) · 32.2 KB

ROADMAP.md

File metadata and controls

527 lines (431 loc) · 32.2 KB
Raw

CAYADEV Visualizer — Roadmap

This document records, honestly, what has actually shipped and what is planned. A row is only marked done when the feature works in the application and is covered by a test or by the GPU self-test.

Current release: v3.1.2 · Next release: v3.1.3

Release Theme Released State
v1.3.1 Modes, backgrounds, recording 2026-08-06 Shipped
v2.0.0 Shader editor, MIDI, OSC, OBS source, offline render 2026-08-30 Shipped
v2.1.0 Layers, post-FX, 3D geometry, Art-Net, Auto VJ 2026-09-01 Shipped
v3.0.0 Modulation, deep analysis, MilkDrop language, mapping, transitions 2026-09-02 Shipped
v3.1.0 Timeline, Clip Deck, accidental-close protection, Electron 43 2026-09-02 Shipped
v3.1.1 Cross-platform builds, OpenRGB, Spout and Syphon 2026-09-04 Shipped
v3.1.2 MilkDrop shader engine, Now Playing overlay, transparent visualizer 2026-09-05 Current
v3.1.3 Per-application audio capture — Next
v3.1.4 Comprehensive video export — Planned
v3.1.5 Broadcast layout editor — Planned
v3.1.6 NDI output — Deferred
v3.2.0 Redundancy, failover and frame sync — Planned

Status table

Feature v1.3.1 v2.0.0 v2.1.0 v3.0.0 v3.1.0 v3.1.1 v3.1.2 Note
Multi-monitor ✅ ✅✅ ✅✅ ✅✅ ✅✅ ✅✅ ✅✅ A separate window on every selected display
System audio ✅ ✅ ✅ ✅ ✅ ✅✅ ✅✅ WASAPI loopback on Windows, CoreAudio on macOS, PulseAudio/PipeWire monitor on Linux
Multi-source mixing ✅ ✅ ✅ ✅ ✅ ✅ ✅ Mixed before the FFT
Layer compositing ❌ ❌ ✅✅ ✅✅ ✅✅ ✅✅ ✅✅ Unlimited layers, 17 blend modes, groups, solo/mute/lock
Layer masks ❌ ❌ ❌ ✅ ✅ ✅ ✅ Alpha from another layer, plus shape and gradient masks
Post-FX ❌ ❌ ✅ ✅✅ ✅✅ ✅✅ ✅✅ 40 GPU effects, orderable, audio-bindable, per-layer chains
Visualizer modes 14 31 32 48 48 48 50 Includes 14 generative modes, nowplaying and geometry
Spectrum metering ❌ ❌ ◐ ✅ ✅ ✅ ✅ Four frequency scales, dB amplitude, attack/release ballistics, spread and smoothing
Broadcast layouts ❌ ❌ ❌ ✅ ✅ ✅ ✅ Bar placement, logo beside the bars, track and artist text
Backgrounds 10 19 19 31 31 31 31 All share the palette and template system
Colour presets 10 10 58 58 58 58 58 Seven groups; apply to Studio and the 3D engine too
Formulas ❌ ❌ 35 98 98 98 98 30 plane curves, 12 space curves, 29 surfaces, 27 attractors
3D solids ❌ ❌ ❌ 13 13 13 13 Platonic solids, geodesic spheres, L-systems, IFS clouds
True 3D ❌ ◐ ✅ ✅ ✅ ✅ ✅ Own matrix maths; no third-party 3D library
Modulation engine ❌ ❌ ❌ ✅✅ ✅✅ ✅✅ ✅✅ LFOs, envelopes, S&H, random → any config path
Deep audio analysis ❌ ❌ ◐ ✅✅ ✅✅ ✅✅ ✅✅ Constant-Q chroma, key, chords, HPSS, YIN pitch, loudness
Scene transitions ❌ ❌ ❌ ✅ ✅ ✅ ✅ 18 transitions, switchable off
Projection mapping ❌ ❌ ❌ ✅ ✅ ✅ ✅ Corner pin, mesh warp, soft edge, per-output masks
MilkDrop ❌ ❌ ❌ ❌ ❌ ❌ ◐ Functional baseline in v3.1.2: closure compiler solves CSP (#559), HLSL warp/comp translated (91.7% stage compile), 8 wave modes, repeat wrap, feedback clamp. Phase 2 quality pass scheduled
Live shader editor ❌ ✅ ✅ ✅ ✅ ✅ ✅ GLSL, live preview, error line, custom sliders
Built-in shaders ❌ 5 5 42 42 42 42 All compile on the GPU in the self-test
Shadertoy / ISF import ❌ ✅ ✅ ✅ ✅ ✅ ✅ Local converters; no service is contacted
Scene templates ❌ ❌ ❌ 72 72 72 72 Nine groups, each verified not to damage a working setup
Text and lyrics ❌ ❌ ❌ ✅ ✅ ✅ ✅ Audio-reactive typography, LRC/SRT import, timing editor
MIDI ❌ ✅ ✅ ✅ ✅ ✅ ✅ Learn; CC/note → any setting or action
OSC ❌ ✅ ✅ ✅ ✅ ✅ ✅ UDP listener, hand-written OSC 1.0 parser
Art-Net / DMX ❌ ❌ ✅ ✅ ✅ ✅ ✅ ArtDMX output; packet layout tested byte by byte
BPM / tempo ❌ ❌ ✅ ✅ ✅ ✅ ✅ Period histogram; tested to ±0.5 BPM
Auto VJ ❌ ❌ ✅ ✅ ✅ ✅ ✅ Bar-aligned scene, mode and palette changes
Recording ◐ ✅ ✅ ✅✅ ✅✅ ✅✅ ✅✅ One-key capture, GIF export, 4× PNG snapshot
Video / webcam input ❌ ✅ ✅ ✅ ✅ ✅ ✅ Also readable as sv_media inside shaders
OBS integration ❌ ✅✅ ✅✅ ✅✅ ✅✅ ✅✅ ✅✅ Browser source — no plugin, real transparency
Offline render ◐ ✅ ✅✅ ✅✅ ✅✅ ✅✅ ✅✅ Frame-exact and deterministic — the regression net
Windows Dynamic Lighting ✅✅ ✅✅ ✅✅ ✅✅ ✅✅ ✅✅ ✅✅ Unusual in this class. Windows only — elsewhere the card explains why and OpenRGB takes over
Mobile remote ❌ ✅ ✅ ✅ ✅ ✅ ✅ Scenes, templates, Studio presets
Automated tests ❌ ◐ ✅ ✅✅ ✅✅ ✅✅ ✅✅ 971 unit tests + a GPU self-test over every engine (808 at v3.1.1, 703 at v3.1.0)
Timeline ❌ ❌ ❌ ❌ ◐ ◐ ◐ Shipped in v3.1.0. Tracks, clips, automation lanes, markers, one shared transport. Partial: no multi-select on the canvas, no tempo map editing
Clip deck ❌ ❌ ❌ ❌ ◐ ◐ ◐ Shipped in v3.1.0. Sparse grid, beat-quantised launch, follow actions, performance view. Partial: one deck, and only scene/template slots apply
Accidental-close protection ❌ ❌ ❌ ❌ ✅ ✅ ✅ Shipped in v3.1.0. Recovery and an Esc lock, both off by default
Windows build ✅ ✅ ✅ ✅ ✅ ✅ ✅ NSIS installer and a portable build
macOS build ❌ ❌ ❌ ❌ ❌ ◐ ◐ Shipped in v3.1.1. .dmg and .zip, Apple Silicon, built on a macOS runner. Unsigned, and system audio needs BlackHole. Never launched on a real Mac
Linux build ❌ ❌ ❌ ❌ ❌ ◐ ◐ Shipped in v3.1.1. AppImage and .deb, x64, built on a Linux runner. The audio engine loads; never launched on a real desktop
Runs without a Node install ✅ ✅ ✅ ✅ ✅ ✅✅ ✅✅ Windows never needed one: every build since v1.3.1 carried its own 93 MB node.exe. v3.1.1 drops that payload and covers all three platforms — the helper runs on the app’s own binary (ELECTRON_RUN_AS_NODE)
OpenRGB ❌ ❌ ❌ ❌ ❌ ◐ ◐ Shipped in v3.1.1. All three platforms, per-LED, sharing one renderer with Dynamic Lighting. Tested against a protocol-level server, not real devices
Spout / Syphon ❌ ❌ ❌ ❌ ❌ ◐ ◐ Shipped in v3.1.1. GPU handoff, measured end to end on Windows at 30 fps with none dropped. Syphon shares the code path but has never run on a Mac. Absent on Linux
Now Playing / SMTC ❌ ❌ ❌ ❌ ❌ ❌ ✅ Shipped in v3.1.2. Windows SMTC session reader via persistent PowerShell loop, anchor interpolation, 7 animations, OG & Modern styles
Transparent window ❌ ❌ ❌ ❌ ❌ ❌ ✅ Shipped in v3.1.2. Full window and canvas transparency for desktop overlay use
Electron 33 33 33 33 43 43 43 33.4.11 reached end of life in April 2025
Legend: ✅ present · ✅✅ best-in-class · ◐ partial · ❌ absent_NODE`)
OpenRGB ❌ ❌ ❌ ❌ ❌ ◐ Shipped in v3.1.1. All three platforms, per-LED, sharing one renderer with Dynamic Lighting. Tested against a protocol-level server, not real devices
Spout / Syphon ❌ ❌ ❌ ❌ ❌ ◐ Shipped in v3.1.1. GPU handoff, measured end to end on Windows at 30 fps with none dropped. Syphon shares the code path but has never run on a Mac. Absent on Linux
Electron 33 33 33 33 43 43 33.4.11 reached end of life in April 2025
Legend: ✅ present · ✅✅ best-in-class · ◐ partial · ❌ absent

Not shipped yet

Kept separate from the table above on purpose: that one records what has actually been released, this one records intent. Nothing here is in a user’s hands, and a row only moves up once it ships.

Feature Target State today
Per-app audio capture v3.1.2 Not started. Needs a native module on both Windows and macOS; the same API family gives macOS its missing system-audio loopback
Comprehensive video export v3.1.3 Not started
Broadcast layout editor v3.1.4 Not started
NDI v3.1.5 Deferred for licence reasons, not difficulty — see "Not done, and why"
Redundancy / genlock v3.2.0 Not started

Shipped, but never run on real hardware

Four things went out in v3.1.1 that this machine cannot verify. They pass their tests and they are in users’ hands, so they belong in the table above rather than in "not shipped" — but calling them proven would be a lie.

What Verified Not verified
macOS build Builds on a macOS runner; arm64 Mach-O binaries confirmed inside the .dmg, and NSMicrophoneUsageDescription is present Never launched on a Mac. Gatekeeper behaviour, BlackHole capture and CoreAudio enumeration are all unobserved
Linux build Builds on a Linux runner; audify loads there in CI, and the .deb declares libpulse0 Never launched on a desktop. Monitor-source enumeration is untested against a real PulseAudio or PipeWire server
OpenRGB Protocol encoded and decoded against a server implementing the wire format, 38 tests Never driven a physical LED. Vendor quirks and per-device direct-mode support are unknown
Syphon Shares the whole code path with Spout, which was measured end to end on Windows. The macOS handle conversion is unit-tested Never run on a Mac. The IOSurface path has never touched a real GPU

This table shrinks as the hardware becomes available; nothing is moved out of it on the strength of an argument.

Verifiability

Counting features is easy; showing they are correct is not. Every claim above is backed by something runnable:

npm test
npm start -- --smoke
  • 971 unit tests, all passing on main. 703 of those shipped in v3.1.0; 105 came with v3.1.1; 163 came with v3.1.2. Formulas are checked against values derived by hand from their definitions — Viviani's curve staying on its sphere, the torus tube radius, Chladni's m↔n antisymmetry, every attractor staying bounded and landing inside the view volume.
  • Tempo is measured against synthetic signals of known BPM (90/120/128/140/174 → 89.8/120.4/127.9/140.0/173.7).
  • Analysis is tested against synthetic signals with known answers: a known chord must come back as that chord, a 220 Hz tone as 220 Hz.
  • Art-Net is verified byte by byte against the ArtDMX header.
  • The GPU self-test draws every registered mode, background, post-FX effect, shader, formula and transition on a real GPU and measures that the result is not blank; it then switches the interface to English and scans for untranslated strings. It also asserts that automation never opens the camera, after the screenshot generator was caught doing exactly that.

v3.1.0 — Timeline and Clip Deck · shipped

Two performance surfaces that share one clock and one quantiser. A deck with its own clock would drift from the timeline, and things meant to fire on the same beat would visibly separate.

Timeline

  • Tracks holding clips, and automation lanes writing into the same targets the modulation engine exposes — anything modulatable is automatable.
  • Transport with play, pause, stop, a loop region and scrubbing that updates the scene immediately, including while paused.
  • A zoomable canvas ruler reading in seconds and in bars, with grid density chosen from the zoom so it never turns into noise.
  • Clip placement by dragging, edge-trimming and snapping to bar, beat, half, quarter or frame, with Alt suspending the snap mid-drag.
  • A keyframe editor reusing the modulation engine’s curve set, so a curve named the same behaves the same in both places.
  • Named markers, jumpable, importable from an LRC or SRT file through the existing lyrics parser.

Seconds are the single source of truth; bars and beats are derived from a tempo map. Storing both would let them disagree with no way to tell which was right. Conversion accumulates across tempo changes, because multiplying by one constant BPM shifts everything after the first change and the error only shows during a show. retimeToTempo() preserves musical positions across a tempo change as an explicit action rather than a silent side effect.

Clip Deck

  • A sparse grid: empty slots are not stored, so an 8x8 deck holding one clip writes one record rather than sixty-four.
  • Beat-quantised launching from one frame up to four bars, with a countdown on the armed slot. Launching exactly on a grid line waits for the next one — otherwise hitting the beat squarely makes quantisation look random.
  • A cut is a true cut, with the fade forced to zero rather than left to the drawing side; a fade reuses the existing 18 transitions.
  • Column launch fires a whole row on one frame, aligned to the longest quantisation in that row so nothing staggers.
  • Follow actions: stop, loop, next, random in column, or go to a named slot. They chain, with a per-frame fire cap so a chain of near-zero-duration clips reports an overrun instead of pinning the machine.
  • A performance view with nothing but the deck, transport and blackout: large targets, high contrast, every slot reachable from the keyboard.
  • Deck activity records back onto the timeline as tracks, turning an improvised set into an editable one.

Determinism

Offline export is the project’s visual regression net, and it only works because every time-dependent source derives from the draw clock. The timeline obeys the same rule: automation evaluated at a time is the same value however the playhead reached it, and the exporter derives time from the frame index alone. The live path uses a separate wall-clock anchor so that several visualizer windows compute the same playhead in closed form, without a message per frame and without drift between them.

Also in v3.1.0

Accidental-close protection. Two switches in Settings, both off by default. With protection on, a visualizer window that closes unexpectedly reopens at once; recovery keys off intent, so closing from the panel, from the remote or with Esc is never undone. A second switch locks Esc, leaving the panel’s Close button and Ctrl+Alt+Shift+Esc as the ways out — a visualizer that cannot be closed would be worse than one that closes by accident.

Electron 43.5.1. 33.4.11 reached end of life in April 2025 and shipped Chromium 130. The upgrade is also the floor for the Spout and Syphon senders in v3.1.1, which need Electron 40 or newer.

Not in v3.1.0

Written down rather than quietly dropped:

  • Multiple decks (A, B, C…). The data model carries a deck list and every engine call is addressed by deck id, so the work is a deck selector rather than an engine change. One deck is what ships.
  • Timeline arming deck slots. The other half of the bridge — recording deck activity onto the timeline — is done; driving the deck from the timeline is not.
  • Multi-select and numeric nudge on the timeline canvas. Clips move and trim one at a time; exact positions are typed in the inspector.
  • Tempo changes mid-show through the interface. The engine and the tests support a full tempo map; the panel edits a single tempo.
  • Rendered clip thumbnails. Slots show the referenced scene’s colours. Capturing a real frame would mean switching the visualizer to that scene — changing the show to draw a preview of it.

v3.1.1 — Cross-platform builds, OpenRGB, Spout and Syphon · shipped

Live video output to other applications on the same machine, straight over the GPU. The visualizer feeds Resolume, OBS or any Spout/Syphon receiver without window capture and without a plugin install.

The release grew beyond that, and not by accident: Syphon needs a macOS build, and there has never been one. Nine releases have claimed Windows and macOS support while shipping only .exe files. So the platform work is not scope creep around Spout and Syphon — it is their precondition, and Linux comes along with the same CI job.

Running on three platforms

The application assumed Windows in four places that would each have broken a macOS or Linux build on its own.

The audio helper had no runtime. It ran on a system Node, and the Windows build carried a 93 MB node.exe so that users would not have to install one. The other two platforms got neither. Electron is itself a Node runtime: ELECTRON_RUN_AS_NODE=1 makes the application's own binary behave as one, and audify loads under it because it is N-API. That removes the bundled node.exe entirely, and the download shrinks with it.

Packaging demanded Windows files unconditionally. extraResources listed node.exe and the Dynamic Lighting native module with no platform condition, so a macOS or Linux build failed before it started.

Capturing system audio works differently on each platform, and only the Windows case existed:

How the system's own audio is captured
Windows An output device is opened for capture — WASAPI loopback
Linux The PulseAudio/PipeWire monitor, which is an input device
macOS Not possible. CoreAudio has no loopback at all

The old code looked for an output device on every platform. On Linux that picks the speakers and tries to capture them, which PulseAudio will not do. On macOS nothing would have worked at all. That rule now lives in a pure module, so the two platforms that cannot be run here are still covered by tests.

macOS therefore needs a virtual audio device — BlackHole is free — and when there is none the application says so and names it, rather than falling back to the microphone and letting someone visualise the wrong source without noticing. The real fix is CoreAudio's process taps (AudioHardwareCreateProcessTap, macOS 14.2+), which is the same API family as Windows' process loopback, so it belongs with v3.1.2 rather than here.

Windows Dynamic Lighting is a Windows API. Its card is now hidden off Windows instead of being shown saying only that it is unsupported. A settings file carried over from Windows explains why the setting is inactive.

What macOS users have to accept in this release

Written plainly rather than discovered after downloading:

  • The build is unsigned. No Apple Developer Program membership was bought, so macOS reports the app as damaged and it has to be cleared by hand. This is a cost decision and can be reversed at any time.
  • System audio needs BlackHole or another virtual device, per the table above.
  • Windows Dynamic Lighting is absent. OpenRGB and Art-Net/DMX are the RGB paths on macOS.
  • Apple Silicon only. Apple finished the Intel transition: the last Intel Macs left sale in mid-2023, and macOS Tahoe 26 was announced as the final release supporting them. An arm64 .dmg will not launch on an Intel Mac at all, so this is a real if shrinking exclusion rather than a degraded experience. A universal binary would cover both, at the cost of carrying two sets of native binaries in one package, and it is not worth that here. The GitHub Intel runners also sat queued for over fifteen minutes without ever starting, on two separate attempts.

OpenRGB — RGB on all three platforms

Dynamic Lighting leaving with Windows would have left macOS and Linux with no consumer RGB support at all. OpenRGB is cross-platform, supports far more devices than Windows' LampArray, and speaks a documented TCP protocol — the same shape as the Art-Net/DMX output that already exists, so it slots in beside it rather than replacing anything.

On Windows it is an addition, not a replacement: Dynamic Lighting stays, and the two can be used separately or together.

NDI was originally part of this release. It moved to v3.1.5 — not because it is harder to build, but because it is the only one of the three with a licence burden. Spout is BSD 2-Clause and Syphon is Simplified BSD: both need nothing more than a copyright notice. The reasoning is under "Not done, and why".

How it works

Chromium keeps the WebGL2 output inside its GPU process, and the D3D11 texture handle is not normally reachable from the application. Reading pixels back to the CPU would cost roughly 475 MB/s at 1080p60 and would defeat the point of a low-latency sender.

Electron exposes the GPU path directly. With webPreferences.offscreen.useSharedTexture, the paint event carries the texture itself — a shared ID3D11Texture2D handle on Windows, an IOSurface on macOS — with no CPU copy at any point.

The architecture this slots into already exists. A hidden offscreen window loads the same page src/main/stream-server.js already serves to OBS as a browser source, so audio frames arrive over the WebSocket that is already there and the rendering engine needs no changes at all. Spout and Syphon become output targets beside "OBS browser source" rather than a parallel pipeline.

Linux gets no texture sharing, and there is no way around it

Spout is a Windows technology and Syphon is a macOS one. texture-bridge ships prebuilt binaries for win32-x64-msvc, darwin-x64 and darwin-arm64 and none for Linux, and Linux has no established equivalent to port to. The option is not offered there at all rather than shown as a switch that does nothing; Linux keeps the OBS browser source, which already works.

Dependencies, and the question that is now answered

@napolab/texture-bridge covers both platforms in a single MIT dependency: Spout over DXGI shared handles on Windows, Syphon over IOSurface and Metal on macOS, with prebuilt N-API binaries for win32-x64-msvc, darwin-x64 and darwin-arm64.

It also ships a receiver, which the roadmap had not previously accounted for: Resolume or any other Spout/Syphon sender can be read into the visualizer as an input layer. That arrives free with the same dependency and is worth a layer source of its own.

One thing was assumed rather than proven, and it was checked before anything was built on top of it: audify runs in a separate Node subprocess precisely because its prebuilds target the Node ABI rather than Electron. texture-bridge has to load inside the main process, since that is where paint events are raised. Node-API is ABI-stable across both runtimes, so it should load without electron-rebuild — but that is exactly the kind of assumption audify punished.

It held. A bare require in the main process succeeded, and so did the second assumption underneath it: that Electron’s offscreen renderer really hands over a GPU texture rather than falling back to a CPU copy. Measured at 60 fps with a texture on every frame. Both were proven before the sender was written, which is why the sender itself worked the first time it ran.

Electron upgrade — a prerequisite, not a side quest

texture-bridge requires Electron 40 or newer. This shipped in v3.1.0: the application now runs Electron 43.5.1, so the floor is already met.

The target is 43.5.1. Electron 33.4.11 reached end of life on 29 April 2025 and carries Chromium 130 against 152 in the current line, which is the real security argument on its own. Two smaller reasons pick 43 over 44: 43 has had five patch cycles to settle where 44 is days old, and 43 carries electron/electron#51287, which removed the white borders frameless fullscreen windows had on Windows — this application creates exactly that kind of window. Version 44 would cost no extra migration work and buys a longer support runway, so it is a reasonable alternative rather than a wrong answer. electron-builder moves from 26.8.1 to 26.15.x in the same change.

What the upgrade does not fix: electron/electron#45774, the white flash on the first show() of a hidden window, is closed as not planned and untouched since August 2025. The workaround — creating the window already visible and already fullscreen, never calling show() — has to survive the upgrade intact, and verifying that it still does is part of the work.

v3.1.2 — MilkDrop shader engine, Now Playing overlay, and transparent visualizer · shipped

  • MilkDrop shader engine and bugfixes (#559):
    • Rewrote the expression compiler from new Function JavaScript generation to an AST closure tree, solving the Content Security Policy eval block without unsafe-eval.
    • Translated MilkDrop HLSL warp and composite pixel shaders into WebGL2 GLSL with multi-pass type inference. 91.7% of all shader stages compile across 10,347 official and community presets; 85.0% of presets have completely clean shader pipelines.
    • Fixed warp texture wrapping: MilkDrop defaults to REPEAT, whereas the engine previously clamped edges, causing effects to collapse and empty (e.g. Agitator brightness dropping from 16.7 to 0.7).
    • Restored 8-bit saturation clamping on feedback buffers, preventing runaway white/yellow blowouts on float buffers.
    • Implemented all 8 MilkDrop waveform modes, custom shapes, and custom wave geometries matching BeatDrop / MilkDrop reference implementations.
    • Aligned audio scale and decay mappings (bass/mid/treb normalized to long-term average, fDecay header mapped to per-frame decay).
    • Deferred to Phase 2 (quality pass): User sprite textures (img.ini), motion vectors, preset crossfade blending, and mouse interaction.
  • Now Playing overlay:
    • Live system media integration via Windows SMTC (GlobalSystemMediaTransportControlsSessionManager) through a persistent PowerShell 5.1 helper (0–1 ms steady-state overhead).
    • Anchor-based time extrapolation (positionAt / anchorMs) to provide smooth elapsed/remaining progress between sparse OS timeline updates.
    • 7 entrance animations (fade, slideUp, slideLeft, scale, typewriter, wipe), marquee scrolling for long titles, and audio-reactive bass pulse.
    • Modern and Classic OG Winamp styles with continuous or 40-segment progress bars.
    • Full admin configuration panel with live playback status indicator (green/dim/red).
  • Transparent visualizer window (Addendum 4.0):
    • Support for background.transparent toggle, creating an Electron window with transparent: true and #00000000 background, clearing canvases with clearRect, and removing body background to run overlays directly above the desktop or game capture.
  • Verification: 971 unit tests pass on main. Automated GPU smoke test verifies all 50 modes, 31 backgrounds, 40 post-fx, and zero untranslated UI strings.

v3.1.3 — Per-application audio capture

Pick which application's audio is analysed. Separate a game or a voice chat from the music, so the visualizer follows only Spotify or the DAW instead of whatever the system is mixing.

v3.1.4 — Comprehensive video export

Faster export, more formats, and enough presets that a content creator never has to think about encoder settings. Cutting export time is a first-class goal of this release, not a side effect.

Feature What it gives you
Multiple formats MP4 (H.264/H.265), ProRes/DNx for editing, WebM, GIF/APNG for short loops
Preset profiles "YouTube 1080p60", "Reels 1080×1920", "Twitch overlay", "Master 4K" — one click
Duration sources Fixed seconds · timeline length · audio file length · N bars from the BPM
Audio muxing Muxed from the chosen wav/mp3/flac rather than from the system mix, so the master is in sync and clean
Alpha / transparency ProRes 4444 or WebM VP9 with alpha, for overlay stock
Batch Queue one scene across ten songs, or three resolutions
Chapters Split a long show at markers
Metadata Title, BPM and scene name written into the file name and ffmetadata
Frame sequence PNG/EXR sequences for post — slow but lossless
Two render modes (1) realtime encode that follows the playhead, (2) offline maximum-speed render. They stay separate modes

v3.1.5 — Broadcast layout editor

v3.0.0 already ships the broadcast template group: a restrained bar visualizer, corner or centre placement, a logo beside the bars rather than behind them, track and artist text, and a still or calm video background. v3.1.5 turns that from a set of templates into an editor.

  • Free placement of the logo, text and bar block, with snapping and safe areas.
  • Logo swapping driven by time, beat, bass or a random interval — the same video can show the label mark in one section and the release artwork in another.
  • Text styling presets, automatic track metadata, and per-element reveal animations.
  • Everything usable live and as an export preset.

v3.1.6 — NDI output

The same picture across a network: one machine runs the visualizer, another running Resolume or a mixing console receives it over IP. This is the standard in professional live and stage setups.

NDI inverts the difficulty of Spout and Syphon. It is technically easier — the send side takes a CPU buffer, so the shared-texture path is not needed and the image bitmap from the same paint event is enough. It is legally harder: the application EULA has to cover the NDI SDK terms, a link to ndi.video is required in the application, on the website and in the documentation, and the NDI tools may not be redistributed. Placing an EULA-bound binary inside an MIT-licensed project is the actual obstacle, which is why it ships as a separate optional package. grandiose provides the bindings.

v3.2.0 — Redundancy, failover and frame sync

The release aimed at staged, professional shows. This category is large and will be built carefully rather than quickly.

  • A backup machine takes over automatically, or very quickly, if the main one fails.
  • The same show file runs synchronised on two machines.
  • Genlock / frame-sync so multiple outputs advance on the same frame boundary.
  • Health monitoring and a pre-show check that reports what is not ready.

This is close to mandatory at large festivals and corporate events, and is standard in media servers of that class.

Not done, and why

These are deliberate omissions, not oversights.

NDI

Free to use in commercial products, and no fee is involved — but the licence is the whole difficulty. The application EULA must cover the NDI SDK terms, a ndi.video link is required in the application, on the website and in the documentation, and the NDI tools may not be redistributed. An EULA-bound binary inside an MIT-licensed project is a packaging problem rather than an engineering one, so NDI ships as a separate optional package in v3.1.5.

Spout (BSD 2-Clause) and Syphon (Simplified BSD) carry none of this, which is why they go first in v3.1.1.

What exists instead: the browser source for OBS does the same job with no plugin install and real transparency, and src/main/stream-server.js already abstracts the output.

WebGPU

The Studio engine is built on WebGL2. WebGPU (WGSL) is more modern, but today the entire Shadertoy and ISF ecosystem is GLSL. Compatibility was chosen deliberately.

Ableton Link

Sharing tempo over a network needs a separate protocol stack and a discovery service. What exists instead: own BPM estimation (tested), tap tempo and a BPM lock.

A central preset store

Deliberate: the application contacts no server, asks for no account and sends no telemetry. Sharing happens through files (.svpreset, .svpack).

"AI" scene generation

The Scene Generator is not a neural network and is not presented as one. It reduces text to four axes with a weighted keyword dictionary and builds the scene with a deterministic generator seeded from those axes. It runs entirely offline.