Twenty-nine installations at the edge where AI stops being a tool and starts being a mirror — observing, reflecting, sometimes revealing things you didn’t know you were showing. Everything runs on the machine in the room. Nothing leaves it unless you say so.
Two installations where the AI learns to reflect. The technology disappears; what remains is the encounter.
An oracle that speaks in your own voice
You speak. The installation listens — not just to your words, but to the patterns in how you say them. Across six phases it asks questions that reach toward vulnerability, progressively cloning your voice as the encounter deepens. Then it speaks back what it observed — in a voice that is unmistakably yours.
An entity that observes without judgment
A camera reads your face through a FER+ emotion classifier. A prosody analyzer reads the shape of your speech — pitch, energy, pace, the pauses between words. An AI entity forms and reforms, tracking your emotional trajectory. Nothing is simulated. The mirror never lies; it only shows what you are already doing.
Each piece is a different kind of observation — sonic, visual, generative, ambient. As of the July 2026 sweep, every installation in the collection runs against the headless tools layer; one awaits a vision-capable backend.
Speak and watch your words bloom into drifting, luminous particles that carry their own momentum across the screen.
Say anything and the room responds — your words become generated images almost before you finish speaking.
Speak a phrase and it blooms into a living nebula — your words choose the stars, your mood paints the void.
A living forest of mathematical forms that grows, contracts, and shifts colour with every sound in the room.
Your voice leaves fingerprints in sound — each utterance distilled into gestural brushstrokes on a slowly accumulating canvas only you could have made.
Live music remakes the visual world in real time — each beat triggering a new style-transferred reality over the camera feed.
You speak, the story continues — a narrative that bends around whatever you bring into the room, periodically summarizing itself to remember where it has been.
Your spoken words paint scenes that persist, building a visual record of everywhere the story has travelled.
Visitors in different locations paint together over the network, with an AI whispering suggestions into the silence between strokes.
One person draws, another receives — the AI mediates between them, translating gesture into generated image.
Solve puzzles that only exist because you spoke them into being — the AI parses your spoken answers and each solution unlocks a new scene.
Runway looks no human designed stream past in a timed cycle — each one a different AI vision of beauty.
Walk through a space and hear it narrate itself — the system reads where you are through the camera and begins to speak. Awaiting a vision-capable backend.
Describe a dream aloud and watch it materialize — surreal generated imagery and a spoken interpretation arrive together.
Speak, and the room reads the feeling in your words, rendering a landscape that matches the emotional weather inside you.
Touch a MIDI controller and galaxies shift — a living mural that blends and morphs at your fingertips.
The artwork changes by the hour — visit at dawn and see something that will never exist again at midnight.
Your face feeds through a corruption machine — pixel sorts and datamosh producing a portrait no camera could have made.
Slide a MIDI fader and watch geometry obey — light, shape, and colour bending to your hand in real time.
A city that breathes with live data — real-world weather reshaping a generated skyline in real time.
Speak and particles bloom outward as if disturbed by an invisible event — physics made from words, with spoken narration woven in.
Say a word and a tile appears — a growing wall of illustrated couplets accumulating into a collective poem.
Every message from every visitor becomes a visual patch — the conversation stitches itself into a growing collective image.
The room listens to what visitors say and colours itself accordingly — collective mood made visible and public.
Speak and your voice returns to you — delayed, layered, pitch-shifted, a little unfamiliar.
Turn a MIDI dial through history — each era renders itself with a scene, a narration, an atmosphere.
Stand inside the end of the world — desolate generated landscapes and ambient dread rendered on demand.
No installations in this collection.
The entire stack runs natively on Apple Silicon via MLX. A privacy gate — local-only by default — blocks any cloud backend at construction time, and rejects non-local URLs even for nominally local services. Voices, faces, and words stay in the room.
Microphone, webcam, and MIDI feed Whisper speech-to-text, FER+ emotion detection, and librosa prosody analysis — pitch, energy, pace, and the pauses between words.
A pluggable LLM backend — Ollama, MLX, Claude, or OpenRouter — reasons over what the room observes. Cloud providers require an explicit opt-in.
The Afterwords TTS server (Qwen3-TTS on MLX) synthesizes speech and clones voices at runtime — the Audio Mirror’s oracle voice is built from the viewer’s own.
A unified image generator defaults to mflux — MLX-native Flux — rendering through an OpenGL pipeline to whatever surface the installation lives on.
# Clone and install
git clone https://github.com/adrianwedd/lunar_tools_prototypes.git
cd lunar_tools_prototypes
python3 -m venv env && source env/bin/activate
pip install -r requirements.txt && pip install .
# Run any installation
python lunar_tools_demo.py --demo audio-mirror
python lunar_tools_demo.py --demo ai-mirror-of-truth
# Run the test suite (deterministic fakes, no hardware)
LUNAR_HEADLESS=1 pytest -q # 197 passedclick to copy