01 · Scale

Built for the many, not the one.

Scene tools are made for one artist at a desk, or a small team passing files. Formicum is made for dozens of AI agents and the people directing them, all working in one live scene at once. One agent dresses the set, another does the look, another lights it, a director steers the camera, and every one of them sees the same scene, changed in real time.

• Set dressing at scale Agents place, scatter and arrange thousands of assets in parallel.
• Lookdev in parallel Materials are shared nodes; change one and every object using it follows.
• Lighting passes A lighting agent works in the same scene the layout agents are still building.
• Direction People steer with a sentence; agents carry it out as small, visible changes.

02 · The message

Every change is a cue.

Agents and people change the scene the same way: by sending a cue, one short line of text. A path, a selector, a value. village/*/roof.mat = mat/gold is 29 bytes and changes every roof in the village. Cues are cheap to write, cheap to send and easy to read, for a model and for a person.

03 · The state

One state. One truth.

The whole scene, and the editor's view of it, lives in a single state. Cues go in one end; a new state comes out the other. Nothing changes any other way. Every component carries its own version, so an agent can tell exactly what moved under it while many others work alongside.

Updates are fast because they are small. A preview render runs light while things move, then refines when they stop: in our tests, six camera moves cost the renderer 3.4 seconds of CPU instead of 20.8.

04 · The trace

The trace replays exactly.

Every cue is kept, in order, with who sent it. Replay the trace and you get the same scene, to the last value. Step back to any moment, see which agent made which change, and why the scene looks the way it does. Like a cathedral raised over generations, no single contributor sees the whole build. The trace remembers it.

05 · The role

Formicum assembles. Your DCC creates.

Formicum is not a modelling package and does not want to be one. Models, rigs and simulations come from the tools you already use and arrive as assets. Formicum is where they are assembled: placed, dressed, given their look and lit, by everyone at once.

06 · The render

Octane first.

We think Octane is the finest renderer there is, so we started there. Formicum drives Octane live, writes Octane's own material nodes rather than an approximation of them, and reads and writes .ocsscenes that open in Octane as they are.

Next: the Render Network. We are building toward sending a Formicum scene to the network as naturally as rendering it on your own machine.

07 · Direction

Rendered in Octane today. Built to render anywhere.

A scene in Formicum is not tied to one renderer. MaterialX is becoming the common language for materials across renderers, and Formicum already carries its catalogue of 802 node definitions. Translating materials through it, so the same scene can render anywhere, is what we are building next.

Status