getModelObject resolves a node's model id to a built
IAutoMovieModelObject. If the same model id appears in multiple nodes
it should return a distinct object each call (a three.js object can live in
one place only). Each node is wrapped in a group placed at its world
transform, so node placement and a pose's own root transform compose
cleanly.
Cameras and the four light kinds (point, area, directional and spot) map onto their three.js equivalents.
The first scene.nodes.length top-level children ARE the designed nodes,
in design order. Both mask passes read that: the legacy ramp colours the
Nth child with the Nth colour, and the stable semantic palette resolves a
designed node to root.children[index] (see
applyAutoMovieSemanticMask). Anything this function adds of its own
goes after them, and a host that prepends a child of its own breaks the
second one exactly as it always broke the first.
A scene carrying a space also gets its structural ground (#1173): the
standable surfaces become real meshes under one SPACE_GROUP_NAME group (see
buildSpaceObject), so guide passes describe a world instead of actors
floating in a void. Their ordinary material writes no beauty colour or depth,
because a semantic support declaration is not evidence that physical floor
geometry exists. The group follows the nodes and lights so the whole
ground reads as one mask colour rather than one per surface.
An optional frame of soft props follows the designed scene objects.
Its returned handle updates plain surfaces at later times and releases
the meshes. Pass an empty frame to opt in before the first prop appears.
The cited refusal contribution covers the supplied shadow profile and unresolved node model. Complete hierarchy, material and numeric admission, lowering receipts and structured safe-retry policy remain with their corresponding owners.
Build a
three.jsscene from an IAutoMovieScene.getModelObjectresolves a node'smodelid to a built IAutoMovieModelObject. If the same model id appears in multiple nodes it should return a distinct object each call (athree.jsobject can live in one place only). Each node is wrapped in a group placed at its world transform, so node placement and a pose's own root transform compose cleanly.Cameras and the four light kinds (point, area, directional and spot) map onto their
three.jsequivalents.The first
scene.nodes.lengthtop-level children ARE the designed nodes, in design order. Both mask passes read that: the legacy ramp colours the Nth child with the Nth colour, and the stable semantic palette resolves a designed node toroot.children[index](see applyAutoMovieSemanticMask). Anything this function adds of its own goes after them, and a host that prepends a child of its own breaks the second one exactly as it always broke the first.A scene carrying a
spacealso gets its structural ground (#1173): the standable surfaces become real meshes under oneSPACE_GROUP_NAMEgroup (see buildSpaceObject), so guide passes describe a world instead of actors floating in a void. Their ordinary material writes no beauty colour or depth, because a semantic support declaration is not evidence that physical floor geometry exists. The group follows the nodes and lights so the whole ground reads as one mask colour rather than one per surface.An optional frame of soft props follows the designed scene objects. Its returned handle updates plain surfaces at later times and releases the meshes. Pass an empty frame to opt in before the first prop appears.
The cited refusal contribution covers the supplied shadow profile and unresolved node model. Complete hierarchy, material and numeric admission, lowering receipts and structured safe-retry policy remain with their corresponding owners.