Hard cut
Place B immediately after A, without position alignment. Watch the return toward B’s original origin.
NO ALIGNMENTTRANSITION LAB / EXPERIMENT 01
Two clips.
One continuous move?
Watch what changes when a travelling animation meets a new one. Four methods, the same source motions, evaluated in Blender—not an artist’s before-and-after edit.
Inspect the transition ↓SYNTHETIC BENCHMARKGenerated glide + arm gesture. Not Moonwalk, Thriller, motion capture, or a realism guarantee.
01 / SEE THE DIFFERENCE
Same inputs. Different joins.
PLAYBACK
A 4-second viewing excerpt from the generated benchmark. One combined recording keeps every panel on the same scene clock. The full incoming clip is retained in the tested Director sequence.
Unaligned NLA
Native Blender NLA crossfade. Smooth influence does not remove the underlying position mismatch.
OVERLAPAligned NLA
The stronger control: B is aligned to A at the overlap’s start, then Blender crossfades the poses.
ALIGN + OVERLAPDirector bridge
Align the incoming clip and add a pose/velocity-aware bridge. Both complete source clips keep their native speed.
ALIGN + EXTRA BRIDGE02 / FOLLOW THE EVIDENCE
Measured, not assumed.
Unaligned endpoint gap—Measured before Director placement
Additional bridge—Not a change to source speed
Full incoming clip—Retained, beyond this video excerpt
Source result bytes—Rechecked after capture and rendering
Where the hips actually go
HORIZONTAL X / METRESLocal stage coordinates with one constant reference for every method. The green floor trace in the video is the projected hip trajectory, not a foot-contact classification.
The timing tradeoff
Crossfades overlap A’s ending and B’s opening. The Director inserts time between complete clips. After the join, identical scene time does not mean identical source phase.
03 / UNDER THE HOOD
Continuity is the start.
Naturalness is the next test.
The Director reads the outgoing endpoint, measures position and velocity, aligns the incoming motion in the target’s space, and creates a separate reversible transition action.
Its quintic position and quaternion-log bridge is kinematic interpolation. It does not understand weight shifts, enforce balance, solve full-body IK, or guarantee believable choreography.
Read the engine design ↗What this demonstration establishes
Actual engine output. The Director panel uses the sequence produced by the reviewed executor. The controls use actual Blender Actions and NLA, not a JavaScript recreation of blending.
Identical presentation. Every view uses the same locked camera, lighting, joint geometry and elapsed clock. Only the animation composition changes.
An honest stronger baseline. Aligned NLA may also look good. This benchmark does not claim the Director wins for every pose pair.
Limited motion. The generated rig glides and moves its arms; it is not a skilled dancer. Smooth sampled boundaries are not proof of natural human performance.
No private assets. No Adventurer, private Mixamo file, user configuration or production scene is published.
Loading measured build information…