Stay stealthy.

The toolkit for bass producers who'd rather stay anonymous.

VOICE MASK · REMIX · VISUALS

Mask your voice into a deep, distorted drop that locks to your grid. Remix your own tracks into VIPs, mashups and genre flips, with tons of takes and no generative AI. Then turn them into stem-reactive visuals and promo clips, with your face masked on camera. Everything runs on your Mac.

The toolkit

🎙 VOICE MASK	🎛 REMIX	🎆 VISUALS
Type a line or record your own voice A 12-module mask rack, locked to your grid Club-loud drops for Rekordbox, CDJs and your DAW	VIPs, mashups and genre flips of your own tracks ROLL for takes, rate them, shape one on a lite-DAW timeline AIFF with rekordbox cues, MP3, an Ableton set	Stem-reactive visuals that follow builds and drops Your face masked on camera: fox, low-poly, your own Promo clips at 9:16, and a projector output for live sets

How it works

  TYPE / RECORD  ─▶  MASK  ─▶  LOCK TO GRID  ─▶  MASTER  ─▶  EXPORT / CLIP
  Kokoro TTS         12-module   AUTO bars,        −7 LUFS      AIFF + rekordbox.xml,
  or your own mic    FX rack     last word on      short-term,  or a face-masked
                                 the beat          −1 dBTP      video

🎙 Source Kokoro-82M TTS on the Apple GPU (MLX): 28 voices, faster than real time. Record or import your own voice. DeepFilterNet3 cleans it, and Whisper + a forced aligner give word-level timing. Persona designer (optional, Qwen3-TTS): describe a voice once and reuse it on every line. Markup for timing: | beat break · [0.5] / [2b] pause · *word* echo throw.

🎛 Mask WORLD vocoder resynthesis: pitch and formant moved independently, plus monotone and scale-lock. Layers: a sub octave, a ghost whisper, and a stacked-voice robotic collage aligned word by word. Machine: channel vocoder, LPC talkbox in key, ring mod, frequency shifter. Colour: Airwindows ToTape9 / Tube2 / DeRez4 / Galactic3, drive, crush, 3-band OTT. Four macros: DEPTH · GRIT · MACHINE · SPACE.

📐 Grid Exact length: 4 bars at 140 BPM is exactly 302,400 samples at 44.1 kHz. AUTO bars picks the length that fits the phrase and its FX tail. END → BEAT / BAR: the last word is warped onto the grid (Rubber Band R3, ≤ 8%). Beat-Lock puts every | chunk on a beat. Speech and tails are never cut.

💿 Master + export CLUB: −7 LUFS short-term max, true peak ≤ −1 dBTP. BAKE-IN: −6 dBFS peaks, no limiting. AIFF 24-bit / 44.1 kHz with ID3 BPM, key and cover art. CDJ-safe PCM WAV (format tag 1). Dry and wet variants, and alternate presets in one export. rekordbox.xml: beatgrid, hot cue A on the first word, a memory cue at voice-out. Drag the output cartridge straight into Rekordbox or your DAW.

🔊 Hear it

▶ Play the 35-second demo

One line, typed once: WHAT THE FUCK IS UP *GITHUB*, rendered by FoxBox at 140 BPM with the last word on the beat and an echo throw on GITHUB, through each voice in turn:

0:00 Dry TTS · 0:02 SIGNAL · 0:04 PACT · 0:10 LEGION · 0:15 ABYSS · 0:20 UNIT · 0:25 GHOST · 0:33 RAW

These are straight exports; only the silence between clips was trimmed.

Presets

	Preset	What it does
🜂	PACT	Deep entity. Pitch −9 st and formant −5 set separately, sub −12 st, a faint robotic stack, tube into hard clip, a dark plate.
📡	LEGION	An intercepted broadcast. A monotone voice chorus, 60 Hz ring mod, GSM codec, a radio band-pass, squelch.
🕳	ABYSS	Pit-demon. Growl subharmonics, detuned doubles, heavy tube drive, a 2.5 s dark hall.
🤖	UNIT	Robot in key. A talkbox on the key root, ring mod, hard clip, a 1/16 slap.
👻	GHOST	Whisper transmission. Breath resynthesis, a reverse swell into Galactic tails.
⚡	SIGNAL	Glitch. A 6-bit DeRez crush, a +200 Hz shift, stutter, tape-stop.
◯	RAW	An anonymizer base for your own voice: formant shift plus McAdams.

A mask-strength badge (SYNTHETIC · WEAK · MEDIUM · STRONG) tells you plainly when a chain is only pitch-shifted, and so reversible.

🎛 REMIX ^{new in 1.5}

1.5.2: REMIX is back in the workshop (greyed out) while its sound is rebuilt. Your saved remixes are kept.

Remix your own tracks into bass-music edits, on your Mac. Pick a recipe, press BUILD, and ROLL for as many takes as you like.

• VIP / DROP SWAP: keep the track and rebuild its drops. HYBRID keeps the held 808 and answers it with designed growls, RESAMPLE re-sequences the track's own bass, and ONE PATCH puts the drops on a single sound.
• MASHUP: A's build or vocals into B's drop, key and tempo matched. MASH RADAR finds partners for A among your own tracks.
• GENRE FLIP: the same track as trap-hybrid, riddim, half-time, 140, four-on-the-floor or DnB, on a kit of your choice.
• TAKES: every BUILD and ROLL is a take. Rate it + / − (ROLL leans toward what you like), keep it, name it, A/B two of them.
• A lite DAW, not a toy: move, duplicate, split and cut sections, SWAP SOUND on one clip or a whole lane, mute, solo and gain per lane, snap, zoom, loop, undo. The drop's anatomy (GAP, FIRST HIT, PAUSE, SWITCH) is marked on the ruler.
• BASS DNA: see how the bass moves (notes, 808 glides, wobble, growl), audition sounds for 2 s, and compare the old bass with the new.
• EXPORT: AIFF with rekordbox cues at every drop, MP3 320, an Ableton LIVE SET (beta), and straight into VISUALS.

🎆 VISUALS ^{new in 1.5}

Make visuals and promo videos for your music, or run them live behind your set. Pick what the visuals listen to, what sits underneath, and stack effects on top. Every layer can react to its own stem.

🎚 The new VISUALS console ^{new in 1.5.2}

• The stage is the hero. A header strip holds the audio source, latency and a live level meter; the layers sit beside the stage as a visible stack.
• LAYERS: drag to reorder, an opacity knob, mute and solo, what each layer reacts to, and LOCK for AUTO-VJ. Face hiding is always the top layer, and turning it off takes two clicks.
• Effect browser: hover a tile to preview it on the stage, click to add.
• PERFORM (press P): the stage full screen with 8 scene pads, AUTO-VJ, BLACKOUT, FREEZE, DROP FX and a hold-to-talk VOICE pad.
• AUTO-VJ new looks: at each new part of the track the unlocked layers change, with a countdown to the next one.
• FACE CHECK: OUTPUT, REC LIVE and SAVE CLIP ask first whenever your real face could show.
• VOICE strip: the live voice mask folds into one row under the stage, with PUSH always in reach.

🧠 Smart visuals ^{new in 1.5}

• AUTO-VJ: the visuals read the song's structure. Builds tighten, the breath before the drop goes still, and the drop hits. LOCK the layers you want to keep as they are.
• Made for bass music: held subs, stabs, wobbles and 808 glides each move the picture their own way, and half-time sections slow it down.
• The preview is the output: pick 9:16, 16:9 or 1:1, and the stage, the projector window, SAVE CLIP and REC LIVE all match exactly.
• Camera: 12 face-hiding styles and AUTO-FRAME. Lean in or reach out and you break through the effects, still hidden.
• TEXT: your lyrics or drop line decrypt, slam and shatter on the drop. MILKDROP cuts on the drop too.
• Safe flashes: strobing is held to 3 flashes a second (WCAG 2.3.1).

Audio source	TRACK: your song. SPLIT STEMS separates drums, bass, vocals and other on your Mac (HT-Demucs on MLX, about 15× real time). LIVE INPUT: your DJ output from an audio interface (e.g. your mixer's USB record out) or Mac system audio. MIC: the live voice mask.
Base	Nothing, the track's waveform, the voice core, your camera (faces hidden), a photo, or a video
Effects	Stack styles and filters, each with its own opacity, blend mode and the stem it reacts to (kick → drums, sub → bass, …)
Sync	Ableton Link locks tempo and beat to rekordbox's decks (Performance mode)
Output	The stage, a full-screen output window for a projector or LED wall, SAVE CLIP and REC LIVE

Family	What it is
FOXBOX	FoxBox's own WebGL styles: the voice core, feedback tunnel, point cloud, spectral terrain, scope, datamosh, flow field. All react to builds, drops and the bass line
MILKDROP	100 classic Milkdrop presets via Butterchurn, with AUTO cycling on the bar and a cut to a high-energy preset on the drop
SHADERS	16 original ISF shaders, all driven by build, drop and bass, plus your own .fs files
FILTERS	15 effects that work on the picture beneath: RGB split, datamosh, kaleidoscope, pixel sort, halftone, VHS, trails, thermal, edge glow, zoom pulse, plus the smart ones: depth focus, motion trails, palette from image, beat strobe and bass wobble
TEXT	5 styles for the words before the drop: sharp GPU text that decrypts, slams and shatters

🦊 Face masks ^{new in 1.5}

With the camera as the base, your face is always covered, on the Mac, before anything is drawn or recorded. Pick how:

• FOX MASK: a lit 3D fox face that moves like a face rig. It follows your head turns, blinks, brows and open jaw, its ears sit on your head, and the far ear folds away when you turn. If tracking is lost for 300 ms, your face falls back to a pixel blur, so it's never shown.
• LOW-POLY: your face as a coarse, lit polygon mesh in privacy colours. The shape moves with you but your features don't show.
• Your own mask: press TEMPLATE for a 1024 × 1024 drawing guide laid out on the face (eyes, brows, nose, mouth, jaw), draw over it, then + MASK to import it as SVG, PNG or WebP. It wraps onto your face like the fox and stays with you.
• Ten more: MOSAIC, BLUR, SOLID, GLITCH, DEPTH GLITCH (beta), ASCII, REDACTED, STATIC, HALFTONE and THERMAL VOID, each with a STRENGTH and an optional pulse from the mix or a stem.

Lean in or reach out and you break through the effects on top, with the mask still on. It all runs on MediaPipe's face and hand tracking on your Mac; nothing is uploaded.

🎭 MASKS ^{new in 1.5.1}

A character creator for your mask. Start from one of eight masks in the LINEUP, then swap any part: the shell, eyes, mouth and jaw, ears and horns, material, colours, pattern, glow and effects. RANDOMIZE rolls every part you haven't locked. Try it on the head model or on your camera, save it to MY MASKS, and WEAR it in VISUALS.

📹 SAVE CLIP

Render the drop, the whole song or a bar range as a 9:16, 16:9 or 1:1 MP4 (H.264 + AAC) for Reels, TikTok and Shorts. It's rendered frame by frame from the stems (faster than real time, with picture and sound in sync), carries the fox watermark, and lands in your exports folder, ready to drag. Hit CLIP on a finished drop in the Studio to start from it. REC LIVE records the stage as you play. Set it to THE DROP with the CAMERA base and it films you through the drop. The camera hides your face with MediaPipe on the Mac, and nothing is uploaded.

🎙 Voice panel

The live mask is still on the page: every Studio preset and the four macros on your mic, push-to-talk (hold, latch or mute), FX pads (THROW · STUTTER · SWELL · TAPE STOP · DROP OUT, landing on the song's grid), MIDI learn, REC SET for the whole performance, and TAKE → STUDIO for a dry take to finish as a drop.

🛠 Also fixed in 1.5

• Drag in your tracks: drop WAV, AIFF, FLAC, MP3 or M4A on the Studio or on VISUALS' TRACK strip. Anything else gets a clear "can't read this file".
• Nothing personal on screen: exports, errors and settings show ~ instead of your home folder, so a stream never shows it.
• The projector window matches the stage exactly, in every format.
• Safe mid-set: the projector keeps playing when you switch pages, the display stays awake while it's on, and FoxBox asks before it quits while the output is live.
• Updates install before you start, and never lock you out: offline or a slow server means you just CONTINUE.
• Easier to read: no text under 11 px, bigger click targets, calmer headers.

🎸 STRINGS ^{new in 1.5.5}

▶ Watch the one-minute STRINGS explainer

A bass-remix instrument you play with your hands in front of the camera. Load any track (or press ▶ TEST BEAT): FoxBox splits it into stems and starts two bars before the drop. Glowing strings stretch between your fingers, and every move remixes the song.

• Hand shapes fire beat FX while you hold them, locked to the grid. Let go and the song snaps back in time. ✊ TEAROUT · ✌️ RIDDIM CHOPS · 🤏 WOBBLE · 🖐 GROWL · 🤘 HALFTIME · 👇 SUB DROP · ☝️ BUILD ROLL (let go: the DROP). Raise your hand for more; one effect per hand, two at once.
• Play the strings: stretch them taut to open the song up (slack is a dark build), tilt them for an 808 slide on the bass, and shake them for vibrato.
• GLASS: frame a window with your thumbs and index fingers and look into another world: THERMAL, X-RAY, HALFTONE, PRISM, KALEIDO or DATAMOSH, each with its own sound. Let go and it shatters.
• The strings perform each effect: they fray, gate, wobble, sag and snap, and the cheat sheet on the right lights up what you're doing.
• Scrub the track: tap a section (INTRO · BUILD · DROP · BREAK …), drag the waveform, or press ←/→ to jump by section; every jump lands on the bar.
• Cleaner stems: the splitter keeps 808 subs on the bass and stops synth leftovers from leaking into the vocals.
• Your face, your call: the raw camera shows by default with ▲ FACE VISIBLE. One click on FACE HIDING masks it, and RECORD and SEND TO OUTPUT ask first.
• RECORD films it; SEND TO OUTPUT puts it on the projector. Tracking runs on your Mac with MediaPipe; nothing is uploaded.
• Also in 1.5.5: tracking no longer freezes when the room goes dark, so face hiding keeps following you when the club lights drop.

TouchDesigner is paused for now. STRINGS is all FoxBox, and the TouchDesigner looks are greyed out while we focus here.

Tour

RACK: 12 modules, each bypassable, with its real parameters.	VOICE CORE: reacts to each render's pitch, words and tails, differently per preset.
TAKES: record from the VOICE panel with a count-in; clean-up and a word-level transcript you can edit.	VAULT: every take, searchable and draggable, exportable as a Rekordbox playlist.
VOICES: auditions, the persona designer, the lexicon, and optional model downloads.	WHAT'S NEW: after each update, what changed and why it matters.

Script markup

Markup	Effect
|	Beat break: the next part starts on the next beat
[0.5] · [2b]	Pause in seconds, or in beats at the session tempo
*word*	Echo throw: a delay/reverb send on exactly that word

GUY FVWKS IS IN THE BUILDING | MAKE SOME *NOISE*

A pronunciation lexicon (e.g. FVWKS → Fawkes) and ALL-CAPS handling keep names and acronyms right. 🎲 gives you a new hype line.

Install

Requires: Apple silicon (M1 or newer) · macOS 14+ · about 2 GB free · internet for the first launch only.

1. Download the .dmg from Releases, open it, and drag FoxBox into Applications.
2. First open only: FoxBox isn't notarized by Apple yet, so macOS asks once.
   • macOS 14: right-click FoxBox → Open → Open.
   • macOS 15+: open it once and click Done, then go to System Settings → Privacy & Security → Open Anyway.
3. Setup downloads the voices and the denoiser (about 365 MB) with live progress. The persona designer (9.1 GB), transcripts (2.9 GB) and the stem splitter (84 MB) are optional.
4. Updates install at startup: FoxBox checks GitHub Releases on the boot screen, verifies each download's SHA-256, and restarts into the new version. After an update, WHAT'S NEW shows what changed.

Privacy: audio, video and text never leave the Mac. The only network use is the model download (Hugging Face), the update check (GitHub) and, when you turn it on, Ableton Link on your local network.

Under the hood

Electron app (React · TypeScript)       IPC proxy · vbx:// audio · drag-out · camera · updater
   ├── VISUALS          compositor · AUTO-VJ director · three.js · Butterchurn (Milkdrop) · ISF shaders + filters · troika text · output window · WebCodecs clips
   ├── camera           MediaPipe (face, gestures, segmentation) in a worker · face styles · pass-through · AUTO-FRAME
   ├── STRINGS          hand shapes → beat FX · string tension, tilt + shake · GLASS worlds (WebGL2) · slip-FX worklet · live stems
   ├── live audio       AudioWorklet mask · Signalsmith Stretch · song deck · live input · stem estimation · MIDI
   └── link-helper      Ableton Link (tempo + beat sync)
        │  token-authenticated, loopback only
Python engine (FastAPI · uv)            library (SQLite) · jobs · exports · rekordbox.xml · songs
   ├── fvwks_voice      Kokoro-MLX · Qwen3-TTS · DeepFilterNet3 · Whisper + forced aligner · HT-Demucs stems
   ├── fvwks_fx         WORLD mask · layers · vocoder/talkbox · Airwindows · arrange · master · song, structure + bass-line analysis
   └── fvwks_contracts  the shared models and seams every package agrees on

Build from source

cd engine && uv sync --all-packages      # Python 3.12; compiles the small Airwindows module once
scripts/check.sh all                     # engine tests (contracts, voice, fx, server) + app tests
scripts/dev.sh                           # run the app against the local engine

Credits

Designed by SmittyTech

Built on open-source work by many people; see THIRD_PARTY_NOTICES.md. The drops you make are yours.

_{GPL-3.0. FoxBox links GPL components (pedalboard, espeak-ng, mutagen), so the app is GPL-3.0 as a whole. Model weights download at runtime under their own licenses; the stem splitter's weights (Demucs, by Meta) are provided for scientific use only, as Setup says before the download. TouchDesigner is Derivative's and isn't included.}