Light-based instrument and graphic score by Interspecifics A/V Installation -- 2026
In this new project, Interspecifics shift away from the microscopic and biophysical phenomena that have shaped much of their practice. Instead, they turn their attention to the human scale and to the ways in which collective bodies organise themselves. They observe movements that resemble the behavior of a swarm, propelled by a shared need to respond to global urgencies affecting communities around the world.
The resonant material in An Instrument Made of Light comes from audio recordings of marches and demonstrations collected in different countries. The collective voice -- the vibrating social body that emerges from political urgency -- becomes the source material for constructing a kinetic light-driven sound installation.
From these recordings, Interspecifics create a new sound: a contemporary mantra in which the energy of global protest is transformed into a luminous and sonic gesture that rearticulates the energy of protest. An automated system selects audio files from open archives such as Amnesty International, Cities and Memory, and other collections of protest recordings. When a fragment is chosen, an instrumental process of transformation begins. Through spectral analysis, the system translates the characteristics of the audio into a luminous score performed by an array of moving spotlights, motors, crystals, and refractors. This choreography of light recomposes the original material in real time, unfolding before the audience as a visual organism. The sound is then resynthesised and performed live through a quadraphonic setup. Light and audio thus operate together as a unified performative structure.
The work proposes an expanded form of listening that centers protest as an energetic field -- an instrument of light in which collective memory becomes vibrating energy, continually activated and reimagined.
The system has three main programs that communicate via OSC and DMX:
SCORE APP (IMOL_CV_GRAPHIC_SCORE_QT_V2.py)
|
|-- Audio pipeline (librosa + k-means on audio features)
| |-- /pattern N -------> Pattern Controller (port 9000)
| |-- /set N -----------> Pattern Controller (port 9000)
| '-- /vel/*, /feat/* --> Max/MSP (port 9001)
|
'-- CV pipeline (OpenCV + MiniBatchKMeans on camera frames)
|-- /system/state ----> Max/MSP (port 9001)
|-- /system/stateB ---> Max/MSP (port 9001)
|-- /state/N ---------> Max/MSP (port 9001)
'-- /track/A..E ------> Max/MSP (port 9001)
PATTERN CONTROLLER (IMOL_PATTERN_CONTROLLER_QT.py)
|-- Listens for OSC on port 9000
|-- Runs behaviour engine (sine, square, saw LFOs per channel)
|-- Manages pattern slots and pattern sets
'-- Sends DMX frames via OLA --> Showtec Net 2/3 --> Physical lights
MAX/MSP PATCHES (port 9001)
|-- SpectralSynthesis.maxpat
| Receives /system/state, triggers partial resynth note
| bursts and bank sweeps via osc_to_jitter_notes.js
| 83-voice additive synthesis from partials text files
|
|-- Sampler_md.maxpat
| Receives /track/A..E, /system/state, /system/stateB
| Granular polyphony (poly~ grains 16) + delay/feedback
|
'-- granulStrig.maxpat
Multi-channel groove + feedback/delay effects
Driven by shared Max send/receive buses (no direct OSC)
PHYSICAL FEEDBACK (the room closes the loop)
Camera sees the light field produced by the DMX fixtures
--> CV states change --> Max reshapes audio --> new analysis
--> new patterns --> lights change --> camera sees change
IMOL_CV_GRAPHIC_SCORE_QT_V2.py is the central hub. It runs two parallel pipelines:
Audio pipeline: Loads a protest recording via librosa (22050 Hz, mono). Extracts per-hop features (RMS, onset strength, spectral centroid, bandwidth, flatness, rolloff, 8 MFCCs) in ~1s windows. Runs k-means clustering on windowed feature vectors to produce up to 7 pattern labels. A meta layer plans set changes across the track duration (1-4 set transitions depending on track length). Sends /pattern N and /set N to port 9000 with minimum hold times (30s for patterns, 18s for sets).
CV pipeline: Reads camera frames, applies background learning, detects bright regions (threshold 180, morphology, area/intensity filters, max 50 contours). Feeds 8-feature vectors (blob count, area, spread, velocity, intensity, left/right balance, top/bottom balance, cluster density) into an online MiniBatchKMeans model (14 states). An adaptive OSC controller adjusts transmission rate based on activity level: static (<0.3), smooth (0.3-0.7), or dynamic (>0.7). Sends /system/state (smoothed float) and /system/stateB (slow exponential glide) continuously to port 9001.
The score app does not receive any OSC -- it only sends.
IMOL_PATTERN_CONTROLLER_QT.py receives /pattern, /set, /blackout, and /pattern_random on port 9000. It manages 9 DMX fixtures with per-channel behaviour modes (off, static, sine, square, saw) at configurable LFO rates. Patterns store full behaviour states for fixtures 1-8 (not fog). The fog machine (fixture 9) runs on an independent timer. Outputs a 512-byte DMX frame at ~20 Hz via OLA to the Showtec Net 2/3 Art-Net node.
Listens on UDP port 9001. SpectralSynthesis.maxpat uses /system/state to trigger note bursts and bank sweeps across 83-voice additive synthesis driven by partials text files. Sampler_md.maxpat uses /track/A..E and both state streams to modulate granular polyphony with delay/feedback processing. granulStrig.maxpat uses shared Max buses for multi-channel groove and effects. Audio output goes to speakers.
| # | Type | Channels | DMX Start |
|---|---|---|---|
| f1-f4 | Moving head spots (14ch) | color, strobe, dimmer, pattern, prism, focus, pan/tilt, speed | 1, 15, 29, 43 |
| f5-f6 | Varytec Hero mirrors (8ch) | pan/tilt, speed, rotation, auto show | 65, 73 |
| f7-f8 | MBM40D mirror ball motors (1ch) | rotation speed/direction | 81, 82 |
| f9 | AF-150 fog machine (1ch) | fog output (independent timer) | 90 |
IMOL/
|-- python-light-engine/ Python applications
| |-- IMOL_CV_GRAPHIC_SCORE_QT_V2.py Score app (V2, ML edition)
| |-- IMOL_CV_GRAPHIC_SCORE_QT.py Score app (V1, audio-only patterns)
| |-- IMOL_PATTERN_CONTROLLER_QT.py Light pattern controller
| |-- main.py DMX sender core (OLA helpers, DmxByteArray)
| |-- cv_state_analyzer.py MiniBatchKMeans state analyzer
| |-- adaptive_osc_controller.py Activity-driven OSC rate controller
| |-- camera_roles.py USB camera role identification
| |-- light_geometry_analyzer.py Background subtraction / geometry
| |-- fader_gui.py Fader GUI component
| |-- IMOL_FIXTURE_TESTER.py Single-fixture DMX channel tester
| |-- IMOL_HERO_PATTERN_CONTROLLER.py Hero mirror controller
| |-- test_displays.py Display detection utility
| |-- fixtures.yml DMX fixture definitions
| |-- requirements.txt Python dependencies
| |-- fog_timer_config.json Fog timer settings
| |-- camera_roles.json Persisted camera assignments
| |-- cv_state_model.json Persisted ML model
| |-- motor_homing.json Motor homing positions
| '-- set_state_memory.json Cross-track set memory
|
|-- max_resynth/ Max/MSP patches and helpers
| |-- SpectralSynthesis.maxpat 83-voice additive partial resynth
| |-- Sampler_md.maxpat Granular sampler + delay/feedback
| |-- granulStrig.maxpat Multi-channel groove + effects
| |-- partials_bank_to_coll.js Partials file loader -> coll
| |-- partials_loader_status_to_dict.js Loader status -> dict bridge
| |-- osc_to_jitter_notes.js /system/state -> note bursts + bank sweeps
| |-- osc_tracker_keyboard.js Tracker -> 83-bin amplitude map (utility)
| |-- partials/ Partial analysis text files
| '-- tan_*.txt Tonal analysis files
|
|-- audio/ Audio archives (gitignored)
| |-- archives/ Source MP3s + RTF metadata (19 tracks)
| '-- processed/ Processed WAV files
|
|-- config/
| |-- osc_addresses.md OSC endpoint reference
| |-- dmx_universes.md DMX universe mapping reference
| '-- audio_archives_metadata.json
|
|-- docs/
| |-- ola_setup_mac_mini.md OLA + Art-Net setup (step-by-step)
| '-- lights_reference/ Fixture manual photos
|
|-- logs/ Runtime logs (gitignored)
|-- img/ Documentation images
|-- start_imol_gallery.sh Gallery auto-start launcher
|-- start_imol_gallery.command macOS double-click wrapper
|-- stop_imol_gallery.sh Gallery stop script
'-- .envrc direnv auto-activation for imol-venv
- macOS (tested on Mac mini and MacBook Pro)
- Python 3.14 with pip
- Max/MSP with FluCoMa externals
- OLA (Open Lighting Architecture) --
brew install ola - Showtec Net 2/3 Art-Net node on a 2.x.x.x network
- 2 USB cameras (one for CV score, one for motor homing)
- DMX fixtures as listed above
git clone <repo-url> IMOL
cd IMOLpython3 -m venv imol-venv
source imol-venv/bin/activate
pip install -r python-light-engine/requirements.txtbrew install direnvAdd to ~/.zshrc:
eval "$(direnv hook zsh)"Then allow the repo:
cd /path/to/IMOL
direnv allowbrew install ola
brew services start olaOpen the OLA Web UI at http://localhost:9090:
- Enable the ArtNet plugin, bind it to the correct network interface (2.x.x.x subnet).
- Patch Universe 0 to an Art-Net output port.
- Set the Mac's Ethernet to a static IP on the Art-Net subnet (e.g.
2.0.0.10, mask255.0.0.0).
For the full step-by-step, see docs/ola_setup_mac_mini.md.
source imol-venv/bin/activate
python python-light-engine/main.py --universe 0 --raw-channel 1 --raw-value 255A fixture on channel 1 should respond.
./start_imol_gallery.shThis launches in order:
- OLA daemon (if not already running)
- Camera initialization (3s wait)
- Score app on display 1 (projector, fullscreen, hidden cursor, auto-loads audio from
audio/archives/) - Pattern controller on display 0 (control screen, fog timer auto-enabled)
- Three Max patches (SpectralSynthesis, Sampler_md, granulStrig)
Logs go to logs/ with timestamps. To auto-start on boot, add start_imol_gallery.sh to System Settings > General > Login Items.
Edit start_imol_gallery.sh to adjust the IMOL_DIR path for a different machine.
source imol-venv/bin/activate
cd python-light-engineScore app:
python IMOL_CV_GRAPHIC_SCORE_QT_V2.py --display 1 --hide-cursor --audio-folder ../audio/archivesPattern controller:
python IMOL_PATTERN_CONTROLLER_QT.py --display 0 --fog-autostartMax patches (from terminal or open in Max):
open -a /Applications/Max.app/Contents/MacOS/Max ../max_resynth/SpectralSynthesis.maxpat| Flag | Default | Description |
|---|---|---|
--display N |
-1 (largest) | Target display index |
--hide-cursor |
off | Hide cursor for gallery mode |
--audio-folder PATH |
none | Auto-load a random audio file on startup |
--camera-index N |
auto | Camera device index |
--cv-state-count N |
14 | Number of ML states to learn |
--cv-state-model PATH |
cv_state_model.json | Model persistence file |
--cv-state-use-rules |
off | Disable ML, use rule-based states |
--osc-out-host |
127.0.0.1 | Pattern controller OSC host |
--osc-out-port |
9000 | Pattern controller OSC port |
--osc-track-host |
127.0.0.1 | Max tracker OSC host |
--osc-track-port |
9001 | Max tracker OSC port |
--osc-out-disable |
off | Disable all OSC output |
| Flag | Default | Description |
|---|---|---|
--display N |
-1 (smallest) | Target display index |
--fog-autostart |
off | Enable fog timer on startup |
--osc-autostart |
on | Auto-start OSC server |
--no-osc-autostart |
-- | Disable OSC auto-start |
| Address | Payload | Description |
|---|---|---|
/pattern N |
int (1-based, 0=stop) | Activate pattern N |
/set N |
int (1-based) | Load pattern set N |
/pattern_random |
none | Activate a random pattern |
/blackout |
none | All fixtures off |
/link/enable 0|1 |
int | Enable/disable Link sync |
/link/tempo BPM |
float | Set Link tempo |
/link/beat POS |
float | Set beat position |
| Address | Payload | Source | Description |
|---|---|---|---|
/system/state |
float | CV | Smoothed ML state (1-14), rate-adaptive |
/system/stateB |
float | CV | Slow exponential glide of ML state |
/state/N |
int 0|1 | CV | One-hot toggle on state change |
/track/A..E |
i f f f f f f f | CV | Blob index, x, y, w, h, area, vx, vy (normalized) |
/vel/0..4 |
int 0|1 | Audio | Discrete energy level toggles from RMS |
/vel/value |
float 0-1 | Audio | Continuous normalized RMS |
/feat/* |
float 0-1 | Audio | Optional features (rms, onset, centroid, flatness, rolloff) |
Apps on wrong screens: Use --display 0 or --display 1 explicitly. Run python test_displays.py to check detection.
DMX not working: Verify pgrep olad returns a PID. Check http://localhost:9090 -- the universe must be patched to an Art-Net output. Confirm Mac and node are on the same subnet. See docs/ola_setup_mac_mini.md.
Cameras not detected: Increase the sleep value in start_imol_gallery.sh. Check camera_roles.json for stale device indices.
ML states not evolving: Allow a few minutes for initial learning. Delete cv_state_model.json to restart. Tune --cv-state-learning-rate or --cv-state-smoothing-tau.
Patterns not changing: Verify the Score app's OSC is reaching the controller (check the OSC IN monitor in the controller UI). Ensure "Enable OSC control" is checked in the controller.
Max not receiving OSC: Confirm Max patches are listening on port 9001. Check the Score app is sending (--osc-out-disable must not be set).
View logs: tail -f logs/*.log
./stop_imol_gallery.shOr manually:
pkill -f "IMOL_CV_GRAPHIC_SCORE"
pkill -f "IMOL_PATTERN_CONTROLLER"
osascript -e 'quit app "Max"'(c) Interspecifics, 2026.