MUSE∞ — The Impossible Museum

Why this exists

Humanity already built its answer machine. We just can't walk into it.

For as long as people have faced the questions that don't fit in a search bar — what makes a life meaningful? — they have answered in paint. Those answers exist. They hang in museums: scattered across continents, sealed behind glass, separated by centuries, and silent. Van Gogh cannot tell you what the yellow meant, and Socrates never stood in the same room as Monet.

MUSE∞ is a museum with those constraints removed:

It collapses space. Works that could never share a wall hang in one room, because your question — not geography — decides what belongs together. The curator is the thing you are trying to understand.
It collapses time. The minds behind the work stand beside you — not as wall labels, not as a chat panel, but as a roundtable across centuries. They read the same painting differently, answer you in their own voices, and disagree with each other. Disagreement is where thinking starts.
It listens back. Every answer you give changes the room — the light shifts, the route bends, the ending moves. A museum stops being an archive you browse and becomes a conversation that reshapes itself around what you say. No two people can walk the same MUSE∞.
It ends where no museum can: with you. Your walk — the paintings you stopped at, the answers you gave, the philosophy your choices revealed — is synthesised into a final world that exists for no one else. You don't leave with a postcard. You leave standing inside your own answer.

What it does

You type one real question. The museum curates a themed exhibition around it — title, chapters, route — and you invite up to three masters (Monet, Van Gogh, Socrates…) who walk beside you through a spine of World Labs Marble worlds rendered at native metric scale, with real Art Institute of Chicago open-access paintings on the walls. No painting appears twice.

Click a painting and a game-style dialogue rises: a master opens in character, you answer, a different master answers you — then three parallel live LLM readings of that exact painting land, each spoken in its own MiniMax cast voice, each re-lighting the room through a constrained effect vocabulary. Click the guided-tour HUD and you auto-walk to the next stop. At the end the masters hold a closing roundtable about the walk you actually took, name the world your choices built — and you step into it.

How we built it

World Labs Marble — nine walkable worlds from authored prompts + reference images, rendered natively as splat + collider pairs (zero bounding-box renormalisation), collider ground-snapping, per-world tuning
Tripo — the masters' 3D bodies: documented public-domain portraits → multiview turnarounds → reviewed GLB companions that walk with the visitor
MiniMax speech-2.8 — seven voices cast per master; sentence-segmented narration with prefetch, and a public-domain score (Mussorgsky, Debussy, Satie) that ducks under every spoken line
GPT-5.x — three-perspective readings + the roundtable via strict json_schema; one shared effect enum feeds both the server schema and the renderer's lighting so the two sides cannot drift
Three.js + vanilla JS, zero build step — and zero keys required to walk: fallbacks are local and clearly labeled

Challenges we ran into

Native-scale splat worlds fight back: collider junk below real floors buried feet mid-body; hilly open worlds put terrain metres above the flat profile ground; frames placed at naive coordinates float in the void. We built a three-tier ground resolver, wall-probing artwork placement that backs frames onto real geometry, and a session digest that is capped client-side and re-clamped server-side, so the finale can only speak about what actually happened.

The bet

Generated worlds are not for escaping reality — they are for entering ideas. A question is an invisible thing; MUSE∞ makes it a place you can walk through. That is the shift this project argues for: from reading knowledge to inhabiting it, from museum-as-collection to museum-as-instrument-for-thinking, from visitor to curator, interlocutor — and finally, the exhibit itself.

One question in. One world out. Everything in between is a conversation the real world could never host.