ShaderMotion Exporter

Source video

Avatar

0:00.00f / 0:00 | decoded held frame rejected missing

Curve Layer Wheel over the graph to zoom time.

Clips

I/O set in/out · N new clip · Space play · ←/→ frame step

Layers

Load a video to detect recorder layers.

Cleanup

Re-interpolate held and stale frames

Remove decode spikes

Reject frames with garbled tiles above

Frames where the color blocks are covered (menus, notifications) are dropped and filled from their neighbors.

Smoothing

Preset

Zero-lag low-pass cutoff in Hz (0 = off). Lower is smoother. Applied after held frames are filled.

Output

Frame rate

Keyframes

Start clip at origin, facing forward

Face blendshapes

Face mesh path

Visemes (lipsync vowels)

Root from center of mass

Bake root motion into pose

Loop

Face tracking is exported as blendshape curves on the mesh at the given path (relative to the avatar root), named for ARKit or Unified Expressions avatars. "Match preview avatar" uses the loaded VRM's own blendshape names and mesh paths instead (other spellings like browOutterUp, _L/_R sides). Recordings made with the ShaderMotion VRCFury prefab carry it; other recordings have none.

Visemes come from ShaderMotion's lipsync slots, which store VRChat's 15 visemes as one 2D point, so only the five vowels come back (vrc.v_aa, v_ih, v_ou, v_e, v_oh; consonants arrive as vowel mixes). On a face-tracked avatar they add to the tracked mouth; untick to leave the mouth to face tracking.

Unity places a humanoid clip by its center of mass. Deriving it from the decoded pose keeps hips and feet where they were recorded. Untick to use the hips position like the original web exporter.

Data area

Left % Top % Width % Height % Sample area

Only change this if the recording is cropped or letterboxed: the color grid must fill the outlined area. Changing it requires decoding again.

Open a ShaderMotion recording