mantra.
Cycles renders stills. Mantra renders frames — and frames carry sub-frame information. Motion blur is not a post-effect; it is what the camera always saw.
The signature is sub-frame motion. Real cameras don't capture instants; a still photograph is the integral of light arriving across the time the shutter is open. A mechanical shutter at 1/48s on a 24fps camera captures roughly half the frame's duration — a 180° shutter. Anything moving in the scene smears across that interval, and that smear is half of what the eye reads as motion on screen.
Production renderers built before the path-tracing era treated this as a primitive, not a post-effect. Mantra is one of those renderers. Sub-frame motion vectors are computed at every shaded point; the renderer integrates over time as well as over the hemisphere; the resulting image is what the camera would have seen with its shutter open across this frame, not what the scene happened to be at the instant the frame stamped.
The deeper integration is with Houdini itself. Mantra reads the same SOP network the artist is editing, with VEX shaders that compile alongside the geometry's VEX wrangles. Volumes, particles, deep camera maps with per-sample depth ordering for compositing, point-cloud occlusion baking — all of it lives in the same node graph as the scene's geometry. The renderer is not a separate stage. It is the last node in the network.
Houdini 18 began phasing in Karma as Mantra's USD-native successor; by Houdini 20, Karma is the recommended path for new projects. Mantra still ships and still renders productions — the studios that built decades of pipeline on it haven't migrated overnight. The shutter is still open.