The 6-axis arm, the LED volume, Unreal driving the wall in real time — the Form Studios / PXO AKIRA / MRMC class of stage has spent five years productionising the hardware. What hasn't been productionised is the layer above it: turning a reference frame or one line of prose into the camera fcurves, the lighting keys, and the volumetric fall-off that match it. A DP and a tech-vis lead do it by hand. That's the layer this tier replaces.
Five tools, one deterministic compiler. A reference still becomes camera and light fcurves on
Blender, Maya, and Unreal by construction
(VCam Core-compatible),
and the same keyframes drive the techno-dolly path. No per-host drift, no re-keying between previz and final.
Eval(plan, manifest) → USD, pure.
They sit on top of your robotic stage the way Sequencer sits on top of Unreal — as the layer that turns intent into the exact motion the hardware will execute. See the full mapping with citations + roadmap on the Scene Stager page.
The AI · Agentic tier is not five disconnected products. It's one continuous loop: a reference image goes in, a staged scene comes out the other side, the skills converge into one USD scene, and that scene branches three ways — digital UE5, physical 3D print, or latent HDR backdrop. The recorded plate drops back into the reference library. The path broadens while the look refines.
$49 · BETA. Prose in: "Coalition dry-dock, abyssal dunes, 75 mm Kodachrome." A staged, topology-validated USD scene comes out — the skills below converge into one result. orchestration method withheld pending filing
$49 · $99 studio. Drop a reference still. It is inverted into a scene specification that fully captures camera and lighting, then baked into camera and light fcurves on Blender, Maya, or Unreal. Same math, same output, every host. method withheld pending filing
$29 · $49 pro. Browse your kit by shape, not by filename. Every asset gets a spectral signature on import — heat kernel + 32 lowest Laplacian eigenvalues — and cosine similarity in that embedded space returns the top-K matches in milliseconds. Eight shape classes, symmetry tags preserved through swaps. Zero machine learning. The math is enough.
$29. AI mesh repair in one button. Four hazards: self-intersection, pinch-off, curvature spike, Euler violation. Each triggers a localized fix from a small audited DAG. No global remesh, no UV rebuild — your bakes survive. This is Step 8 of the 11-step hybrid pipeline: the interchange gate. USD, FBX, Alembic all clean.
$79 · $149 studio. Boolean in, clean quads out. Every simply-connected polygon maps to the unit disk via Schwarz–Christoffel, and pulling the disk's orthogonal coordinates back through that map produces orthogonal isolines on the mesh — which ARE quad edges, by definition. Curvature-aligned, principal-direction-tracked, UDIM-safe. No artist knobs. No tensor fields. Conformal geometry does the work.
▸ Why it accelerates your workflow:
every tool in this tier compiles to the
same deterministic math substrate
that runs identically in Blender, Maya, and Unreal. There is no
per-host rewriting, no manual sync, no drift. A scene staged in Maya
bakes to the same keyframes in Unreal by construction. The compiler is
pure: Eval(plan, manifest) → USD. ML informs the plan,
proofs gate the plan, and deterministic execution produces the result.
1360 of 1360 tests green across the line.
▸ What's coming: a fifth modality (volumetric latent NeRF backdrops), a dedicated film-stock authoring app, native Houdini host, and the full demo workflow video showing the 5-act loop running on a mock studio (computer → 3D print → LED dome → recorded composite → back into the training corpus). The path broadens while the look refines.
The AI tier doesn't generate generic 3D. It calibrates against specific cinematic references — flash-on-black studio lighting, 75 mm kodachrome wide low shots, god-ray cathedral chambers, biomechanical chrome shells. Six stills below; the tools match them.
