Win bigger activations with the crew you have.
One graph drives the installation and the visitor's phone. Build the experience, deploy it on-site as a web link, and skip the fragile hand-off between tools and vendors.
The maths of a bid changes when one crew covers all of it.
One crew covers the whole activation
The TD who authors the wall authors the companion piece too, in the same graph. No second vendor, no second timeline, no fortnight spent reconciling two builds that drifted. Hours that used to leave the building stay billable inside it.
The licence stack stops scaling with the job
A loaded seat today stacks several expensive desktop licences — one for the real-time system, one for 3D, one for compositing, plus whatever the web side needs. This covers the ones a project actually touches, in a browser. Consolidating that is a real number on a job you are already bidding.
Previs in the venue, and the previs becomes the show
Build the space as geometry, put the content in it, and walk a client through it in a browser before anything is fabricated. Because it is the same graph either way, the mock is not thrown away when the pitch is won — it is the first version of the piece you install.
A change on site doesn't cost you a call-out
The author exposes the handful of controls a show actually needs to move and locks the rest, so the producer already standing in the room makes the change. You are not putting the one person who knows the patch back on a train the week you can least afford it.
The show machine is a browser and nothing else
The delivered piece is an ordinary web page. Point a full-screen browser at it on the show machine and that is the install — no bespoke player, no licence server phoning home mid-show, and no per-machine runtime cost sitting in the budget for every screen on the floor.
Touch screens, LED walls, totems, and rooms that react.
A stranger walks up and it answers. What you ship is the installed piece itself — not a file you hand to somebody else to build.
A touch wall people walk up to
Pointer, touch, keyboard, sensors and live data arrive as operators you wire like anything else. Interaction is not a scripting layer someone adds at the end — it sits next to the geometry, the materials and the state machine, wired into what it drives.
A tradeshow totem that reacts to who's in front of it
A camera or a proximity sensor is an input operator like any other, and a generation step can sit in the same graph reading from it. So the totem can answer the person standing at it — their choices, their scan, whatever they just held up — instead of cycling a playlist and hoping somebody stops.
A companion piece in the visitor's pocket
The installation and the companion piece are two outputs of the same source, sharing state in real time. Someone changes the colour on their phone and the wall answers. The two cannot drift, because there's nothing to reconcile — they were authored together.
A room that answers, rather than a loop with triggers on it
The visuals are computed as they run — real geometry, real materials, real physics responding to real input. Visitors work out very quickly whether a thing is listening to them or playing back a loop with a few triggers on top, and they stop trying once they decide it's the second one.
And all of it ships as a plain web link
Whichever of these you build, the finished piece is an ordinary web page. There's no runtime to license per machine, no engine build to keep in sync with a show file, and nothing exotic to install on the show computer beyond a browser. That's one less fragile thing to babysit at 3am on a Sunday load-in.
A producer can adjust it on-site without opening the graph
The author exposes the controls that will need to move during an install — brightness for the room's actual lighting, timing, copy, the attract-loop delay — and locks the rest. The producer on-site changes them from a panel. Nobody has to fly the technical director back out to nudge a number.
Where the file stops being the deliverable.
Every other node-based AI tool ends at a file. You run the graph, it finishes, you download what it made and put it somewhere. That's a real job and we do it too — but it isn't this job. Here the output is the experience: a person walks up, touches it, and it responds. That last step is usually where the work gets handed to somebody else.
A workflow makes the material. A scene is what someone opens. An installation is a scene in the most literal sense — a stranger, a surface, and no manual.
The install, and the visitor's phone.
Today
The web half goes out of house — a second vendor, a second timeline, a second margin. Your TD is booked on the wall while somebody else builds the phone piece, and reconciling the two is a fortnight nobody put in the quote.
With nooodles
One crew authors both from one graph. The hours stay in house, the licence count stops tracking the number of show machines, and the previs that won the pitch is what you install.
The rendered deliverables fall out of the same build.
You already live in scenes — what the visitor walks up to and uses. What's usually missing is everything around the show: the stills for the campaign, the cut for the case study, the frames the client's marketing team asks for on the Monday after.
Run the same graph as a workflow and it hands you those as files — the same visual system, the same type, the same materials, at whatever ratio and duration the ask arrives in. The case-study reel isn't a re-creation of the show. It comes out of the show.
The installation
The wall, the sensors, the companion piece on the visitor's phone, sharing state.
- Mode
- Scene
- Who opens it
- Your visitors
The material around it
Campaign stills, social cuts, case-study frames — files your team places.
- Mode
- Workflow
- Who opens it
- You and your team
When something works, save it and run it again.
These ship with the beta. Most of them began as something we had to build on site.
Touch Wall
SceneDrive an installation surface from gesture and input.
Companion Piece
SceneShare installation state with the visitor's phone.
Live Data Board
SceneA board that keeps running unattended for the length of the show.
Social Cutdowns
WorkflowCut the show footage into every platform format, for the campaign around it.
Frame Extractor
WorkflowPull graded stills and cutdowns out of the live piece, for the case study.
What comes up when you're bidding.
Who on our team actually authors this?
Whoever already builds your real-time work. It is a node graph, so a TD or a creative technologist is at home in it quickly; the parts that usually need a second discipline — the web build, the phone piece, the rendered deliverables — come out of the same graph rather than out of a second contract. What it does not do is turn a producer into a TD. Somebody still has to author it.
What happens if the network drops on site?
A scene is a bundle of files, so it can be served from a machine on the show floor and doesn't need the internet to run. Once it's loaded, nothing has to phone home. What does need the network is anything you deliberately wired to a live feed — data, or the companion phone piece — so decide in the graph what a quiet feed means: hold the last known value, fall back to a state, or show a designed message rather than a broken one.
Can it drive a projector or an LED wall?
The picture goes out the machine's display chain, like any full-screen browser window — projector, LED processor, or a capture card downstream of a display output. Canvas size, aspect and warping-friendly geometry are parameters in the graph, so odd resolutions and non-16:9 canvases are normal rather than a fight. There's no NDI or Spout output today; if your pipeline depends on that, tell us — it's on the list and we'd rather build it against a real show.
Can we previs in the venue before we fabricate?
Build the space as geometry, drop the content into it, and walk the client through it in a browser on whatever screen is in the room. Because the mock and the delivered piece are the same graph, what gets signed off in previs is what installs. What it will not tell you is that the projector is two degrees off, or that the plinth is in the wrong place — that is still a site visit.
Can we run the AI parts on-premises for a brand client?
Generation runs where you choose — in the cloud on a key you paste, so nobody needs a GPU to start, and on your own hardware when the work can't leave the building. The managed version of that is an offering we're defining with early customers rather than something you can buy off a page today. If you have an activation with that constraint, it's exactly the conversation we want. See motion design studios for how the IP-bound version is set up.
Does it need special hardware?
It needs a machine that runs a current browser with hardware acceleration — the same class of machine you'd already be putting behind a screen. Heavier scenes want a real GPU, the way any real-time work does. There's no dongle and no per-machine runtime licence.
What happens if the network drops?
A scene is a web page: once it's loaded, it runs in the browser on the show machine, so a dropped connection does not stop what is already playing. Anything bound to a live feed holds its last value until the feed returns. For a venue with unreliable connectivity, serve the page from a machine on-site and keep the network out of the critical path entirely.
How do visitors get it on their phone?
A link — usually behind a QR code on the plinth or the wall itself. They open it in the browser they already have. No app, no store, no install, nothing to download while standing in a queue.
One crew, one graph, the whole activation.
Free to start, nothing to install. Or tell us what you're building — the surfaces, the inputs, the site — and we'll walk through how it comes out of one graph.
A live bid is the best brief you could give us
Tell us what you're quoting — the surfaces, the inputs, the crew, the site — and we'll work through how it comes out of one graph. What your production actually needs is what we should be building.
No spam. A real reply from a human.