variable.gallery presents
UNDEFINED
An infinite AI art gallery, generated as you walk.
Admission free, and priced accordingly.
Wall text
UNDEFINED is a first-person walk through an endless chain of 3D maze galleries. It is a virtual museum in which a real image model (Stable Diffusion, entirely in your browser) paints everything on the walls, live. A small in-browser language model with strong opinions and no credentials invents every exhibition. A dry, faintly disappointed narrator reads the room aloud. There is no server and there are no API keys. After a one-time download, the whole institution runs client-side and works offline. That is more than most institutions can say.
Each doorway leads to a new exhibition: a name, painting subjects, styles, moods, palettes, and curator remarks. The exhibitions get stranger the deeper you go — classical, liminal, surreal, bureaucratic absurdism, fever dream. You may watch the diffusion noise crackle on the canvas while each painting forms. Every piece has a placard that tells the truth, including the verbatim prompt. The gallery believes in provenance.
Gravity is yours. Push squarely into any wall and it becomes your floor. The narrator will have thoughts about this.
Frequently asked questions
Is the art real?
An AI model on your own machine generates the paintings in real time, from prompts the curator wrote moments earlier. The gallery declines to define "real" and suggests you do the same.
Who painted this?
Stable Diffusion (SD-Turbo / SDXS, ONNX) paints the canvases. A small LLM (Qwen 2.5, 0.5B) curates. Kokoro TTS narrates. None of them have signed anything.
Who is responsible?
The artist Shahab Nedaei conceived and built UNDEFINED. It is a work of new media art about generative art, authorship, and museums that refuse to end.
Do I need a supercomputer?
No. Any modern browser will do. WebGPU paints quickly. Without it, a WebAssembly fallback paints slowly. The gallery finds slowness contemplative.