Live 3D · prompting itself
Concept pitch · Built for live world models like Google DeepMind's Genie 4

Revelation Pavilion

Express your intent. Walk through a tunnel of screens that answers it, and keep prompting as you go. One non-stop shot, no cuts. At the end, a magical box gives you a revelation that's yours alone.

🎬 Watch the non-stop shot → 🚶 Walk the journey Read the concept
Why now

World models just made this possible

On October 27, 2026, Google DeepMind announced Genie 4, a real-time interactive world model. Sundar Pichai described it as letting you steer dynamic environments live (events, camera angles and actions), with native audio generated to match the world. That is exactly the engine this pavilion needs. The tunnel stops being pre-rendered video and becomes a live world each guest steers with their own words, in one continuous shot.

Prompt: sunset, over real desert-valley mountains
Prompt: “sunset”: golden hour over the valley
Prompt: party and aurora
Prompt: “party, aurora”: lasers, fireworks and an aurora over the grounds

Prompt any time

Guests aren't limited to one prompt at the door. They keep steering the world as they walk, and every prompt changes it live. The demo on this site even prompts itself while you watch.

One non-stop shot

No cuts and no loading screens. The world model keeps one continuous scene going while guests reshape it.

Sound that matches

Genie 4 is described as generating native audio, so the tunnel can sound like the world each guest asked for. The demo has a simple version: press Sound and the soundtrack follows each world.

This 3D concept simulates that interaction in the browser so you can feel it today: type into the prompt bar at interactiveworldsimulator.com. Genie 4 is currently a research preview for trusted testers, with access announced for Google AI Ultra subscribers (Project Genie) and early-access partners via the Gemini API.

The idea

One walk, personal to every visitor

Most festival activations show everyone the same thing. The Revelation Pavilion starts by asking each person what they're looking for, then builds the whole walk around their answer. Every guest leaves with something no one else got, and wants to share it.

Open this view in 3D ↗
01 · ENTRANCE DOME

Express your intent

Guests step up to an intent kiosk and a glowing prompt table and say what they're looking for: calm, adventure, joy, love, courage. That intent becomes the prompt for every screen that follows.

Open this view in 3D ↗
02 · TUNNEL OF SCREENS

A world drawn from your intent

A wraparound LED tunnel covers the walls and ceiling with a living world generated from the guest's intent: an ocean with rolling waves, deep space, a forest at golden hour. It flows past them as they walk, so no two walks look the same.

Open this view in 3D ↗
03 · REVELATION PAVILION

The magical box

In the last chamber, under a sky of LED dots, the box opens with warm light and gives each person a unique, inspiring revelation. The whole pavilion's skin pulses with it, and the guest leaves with a share card.

Open this view in 3D ↗
04 · ON THE GROUNDS

A landmark at night

The pavilion's glossy black skin carries "comets" of LED dots that glow after dark. It sits between the Spectra rainbow tower and the Ferris wheel as a sculpture you can see from across the field.

Why it works

Personal, shareable, and built for the crowd

Every walk is unique

The guest's own words drive the visuals and the ending, so the experience feels made for them, not just shown to them.

Made to be shared

A tunnel that changes colour for each person, plus a personal revelation at the end, gives guests a natural reason to post about it.

Works day and night

By day it's a sculptural walk-through. After dark the LED skin turns the pavilion into a glowing landmark on the festival grounds.

One guest, start to finish

One guest's walk

She came alone. She's 29, quit her job on Tuesday and bought the ticket on Wednesday. She joins the queue without telling anyone why.

The table asks what she's looking for. She almost types "fun". Then she types the truth: "I'm scared to start over."

The tunnel answers with a slow dawn. Soft clouds, a sun coming up over her. Nobody else sees this version. Her pace slows.

The lid lifts. She looks down into a tiny harbour town at dawn, the world she described, small enough to hold. Across it, her line: "You are allowed to begin before you feel ready." She photographs the card and sends it to her sister before she leaves the dome.

🚶 Follow one guest →
The surprise in the box

A god's-eye view of the world you described

While the guest walks, a fast image model paints a miniature of the world they asked for, seen from directly above through a tilt-shift lens, like a tiny model village. When the lid lifts, that is what they are looking down into: their world, made small enough to hold, with their revelation written across it.

Fast enough to feel instant

FLUX schnell on Replicate renders in about two seconds. The walk takes two minutes, so the miniature is ready long before the box.

Personal, not random

The image prompt is built from the guest's own words plus the world the tunnel chose, so the miniature matches what they just walked through.

It leaves with them

The miniature is the share card. In the real pavilion it can print from the box as a postcard, with the revelation on the back.

Looking down into the open chest at a miniature island with a lighthouse and sailing boats
“an island with a lighthouse and sailing boats”
Looking down into the open chest at a miniature neon city at night with a ferris wheel
“a neon city at night with a ferris wheel”
Looking down into the open chest at a miniature volcano island with a castle
“a volcano island with a castle”

The lid lifts and you look straight down into the chest. Move your head (or your phone) and you sway over it, with clouds drifting between you and the tiny world. On this site the miniature renders live once an image model is connected; until then the demo paints one itself from the words the guest used, so a lighthouse, a castle or three fishing boats still show up.

How it would run

From a guest's words to the walls, live

Four stages and one continuous stream. The world model does the rendering; everything around it keeps the stream safe, fast and on brand.

1

Intent capture

The kiosk and prompt table at the entrance take one line, by voice or touch. Along the walk, guests keep adding prompts from wall tablets or their phones.

2

Prompt composer

Combines the guest's words with the festival's visual style and the time of day into one world-model prompt, with content filtering before anything renders.

3

World model stream

Genie 4 renders the world as one continuous shot and reacts to each new prompt without a cut. Until API access lands, the same pipeline runs on pre-generated worlds plus the real-time graphics you see on this site. A fast image model paints the god's-eye miniature for the box in parallel.

4

LED mapping

A media server wraps the stream onto the curved tunnel and the stage wall, with the camera path matched to the tunnel's shape. Matching audio plays from the tunnel's speakers.

The main stage LED wall showing the same ocean world as the tunnel
The same stream can feed the main stage's LED wall, so a guest's prompt in the tunnel can ripple out across the grounds.

Latency budget

Under a second from a prompt to the first changed frame, so each guest's words land while they're still looking at the wall.

Never a black wall

If the stream stalls, the tunnel crossfades to the last good world and the shot never breaks. The whole walk still runs without the world model if it has to.

Revelation engine

A curated library of revelations matched to each guest's intent, delivered from the box as a card they can share on the spot. Try the Share button in the demo.

Same tunnel, different intent

Each guest sees a different world

Tunnel screens showing an ocean with rolling waves
Intent: “freedom, like the ocean”
Tunnel screens showing deep space with stars and nebula
Intent: “deep space, full of stars”
Tunnel screens showing a forest with light shafts and fireflies
Intent: “a forest at golden hour”
Tunnel screens showing flames and embers
Intent: “courage and fire”

Nine worlds respond to intent today (ocean, space, forest, fire, sun, clouds, confetti, bokeh and candy), each with a crossfade when the prompt changes. Try it: open the 3D walkthrough, type an intent at the entrance, and watch the tunnel change.

The build
Guests in futuristic festival beachwear queueing at the entrance dome
Guests in futuristic festival beachwear queue at the entrance dome. Palm-lined polo lawns and real desert-valley elevation data, with mountain heights exaggerated 2× for drama.

What's in the concept

ElementWhat it isConcept size (approx.)
Entrance domeIntent kiosk at the door and a touchscreen prompt table inside16 m long · 13 m wide · 6.5 m high
Tunnel of screensHourglass-shaped tunnel lined with curved LED on the walls and ceiling18 m long · 6–9 m wide
Passage domeA short reset space between the tunnel and the finale12 m long · 10 m wide
Revelation pavilionDouble-dome chamber holding the magical box on a plinth30 m long · 15 m wide · 6.3 m high
SkinGlossy black shell with LED dot "comets" that glow at nightAll four structures

Sizes are taken from the 3D concept model and are for scale only. They aren't engineering figures.

By the numbers

What it takes to run

Working assumptions for a first conversation, not a bid. Everything here scales with the number of lanes.

MeasureAssumptionResult
Walk timeEntrance 45 s · tunnel 60 s · passage 20 s · box 40 sAbout 3 minutes per guest
FlowA new guest every 15 s, two box stations alternatingAround 240 guests per hour per lane; 480 with two lanes
A day10 hours open, two lanesRoughly 4,800 guests a day; 14,000+ across a weekend
LEDTunnel walls and ceiling, curved tilesAbout 200 m² of screen, plus the LED-dot skin on all four shells
PowerLED at peak, render nodes, lighting, HVACOrder of 150 kW peak; generator or grid tie
ComputeOne render node per lane plus a spare; one image-model call per guestImage generation under $50 a day at current pricing
CrewPer shift6–8: entrance hosts, tunnel marshals, box hosts, one tech, one lead
FootprintFour structures in a line, plus queue and exitAbout 80 m × 20 m

Throughput is the lever: the box moment is the bottleneck, so two or three box stations per lane keeps the queue moving while every guest still gets their own reveal.

Next steps

From concept to the grounds

  1. Agree on the site footprint and how guests flow from the queue to the exit.
  2. Choose the structure and LED partners, and confirm the tunnel's screen layout.
  3. Build the intent-to-visuals pipeline, the god's-eye miniature and the library of revelations.
  4. Prototype the prompt table and the magical box moment with real guests.
🎬 Watch the non-stop shot → 🚶 Walk the journey