Build one compact formation as chunked instance batches.
Heroes are deliberately absent: the builder promoted them to explicit scene
nodes. Each LOD tier's model is flattened into one mesh, keeping exactly one
64-byte instance matrix and one 4-byte phase scalar per anonymous slot and
tier.
Declared gaits are baked into part-matrix tables and each member reads the selected table at its seeded phase. Positive-stride takes advance by cue travel/turn and the stride heuristic; zero-stride takes instead advance by elapsed time and their declared period. Two scalar uniforms and one texture reference carry a tier's cadence, while culling, spacing/reform and sparse cues require additional update work.
No production scene/film caller is connected in the current tree; direct users and owning tests exercise this producer, so this API alone does not establish that compiled shot populations reach the capture application.
Build one compact formation as chunked instance batches.
Heroes are deliberately absent: the builder promoted them to explicit scene nodes. Each LOD tier's model is flattened into one mesh, keeping exactly one 64-byte instance matrix and one 4-byte phase scalar per anonymous slot and tier.
Declared gaits are baked into part-matrix tables and each member reads the selected table at its seeded phase. Positive-stride takes advance by cue travel/turn and the stride heuristic; zero-stride takes instead advance by elapsed time and their declared period. Two scalar uniforms and one texture reference carry a tier's cadence, while culling, spacing/reform and sparse cues require additional update work.
No production scene/film caller is connected in the current tree; direct users and owning tests exercise this producer, so this API alone does not establish that compiled shot populations reach the capture application.