Tutorial — Single-Camera Body Capture

Augmented Traveler local tutorial. Back to Mocap — Tutorials · capture day: Tutorial — Running a Capture Day · cleanup: Tutorial — HumanIK Mocap Cleanup for MetaHuman

How to get body and finger animation from one ordinary video camera with Epic's MetaHuman Animator Markerless Motion Capture plugin (UE 5.8). We plan to use it for previs, the gesture library and idles, and to keep the suit or optical stage for hero monologues until the test in section 8 says otherwise.

Draft. Written from Epic's mono-video and video-capture guidance, the MetaHuman 5.8 release notes, CG Channel and Digital Production coverage, and user reports on Epic's forum. The feature is experimental: Epic says it is not production-ready and may change or be removed. The technical animator validates this page. Correct menu names here.

1. What it does (verified)

2. Requirements

Item Requirement Source
Engine Unreal Engine 5.8 or later Epic
OS Windows only for body or face-and-body Epic
Plugins MetaHuman Animator; Capture Manager Editor; MetaHuman Animator Markerless Motion Capture (Fab) Epic
App Live Link Hub (for Capture Manager ingest) Epic
Hardware Recommended: 16 physical high-performance CPU cores, 32 GB RAM, 8 GB VRAM GPU (RTX 3070 class), DirectX 12 default RHI Digital Production, quoting Epic
Tools ffmpeg path set in Live Link Hub for video conversion note.com walkthrough (unconfirmed by Epic; check)

Our RTX 5090/5080 workstations exceed this (Hardware — Development Specifications).

Speed: about one second of 1080p footage processes in under a minute (CG Channel). A 60 s idle take is therefore up to an hour. Queue takes overnight.

3. Camera, lighting and clothing

From Epic's video capture guidelines:

  1. Static camera on a tripod. Results fall off with a moving camera.
  2. Resolution HD or 4K; 60 fps; H.264 at CRF 18 or better (about 25 Mbps for HD).
  3. Framing: the whole body, head to feet, with clear margins on every side. Straight-on, neutral height.
    • Leave margin for arm extension on the telescope and pointing gestures.
  4. Lighting: even, diffused, covering head to feet. No harsh key light, no strong shadows.
  5. Background: high contrast with the performer.
  6. Clothing: form-fitting or neutral fit, a different colour from background and floor.
    • This conflicts with period costume. Full skirts hide the legs. Capture in fitted rehearsal clothes plus a short proxy skirt, and record a second take in full costume only as reference.
  7. Nothing occluding any body part. Keep props small and away from the torso where the blocking allows.

Our additions (not from Epic):

4. Record

  1. Set the camera and any phone to 60 fps. Forum users report problems with 29.97/30 fps; check the frame rate in the Capture Data settings after ingest.
  2. Slate and jam timecode as in Tutorial — Running a Capture Day.
  3. Start each take with the actor still, full body visible, arms slightly away from the body for 2 s (our convention).
  4. Record. Keep the actor inside the framed area.

5. Process (steps as documented)

  1. Enable plugins: MetaHuman Animator, Capture Manager, Markerless Motion Capture. Restart.
  2. Ingest: open Live Link Hub → Capture Manager → add a mono video source (a walkthrough calls it Mono Video Ingest; check the exact name) → set the take folder → queue the take → start. This creates a Capture Data asset.
  3. Create a MetaHuman Performance asset (Content Browser → right-click → MetaHuman → MetaHuman Performance).
  4. Set it up (Epic, body only):
    • assign the Capture Data;
    • Facial Tracking off, Body Tracking on;
    • Head Movement Mode = Disable.
      For face and body from one take, leave Facial Tracking on.
  5. Trim the timeline to the useful range.
  6. Process. Users report five passes and long run times.
  7. Export Animation → choose Existing Skeleton (our MetaHuman) or Performer Skeleton.
  8. Apply to the MetaHuman in a Level Sequence and review against the reference video.

The plugin exposes Python and Blueprint automation (forum product page); batch processing is a later task for the technical animator.

6. Known weaknesses

Weakness Evidence Status
Floating feet in some solves CG Channel Confirmed as known issue
Worse with a moving camera CG Channel Confirmed
One actor per solve Epic, CG Channel Confirmed
Hands occluded by body or props Digital Production mentions testing occluded hands Expected; unconfirmed how bad
Foot sliding, ground penetration, jitter Not stated by Epic Unconfirmed; check on first use
Install errors (Smart App Control, antivirus, error MD-0011-0) Forum reports Workarounds in the forum thread

7. Fixes for feet and contacts

No solve setting for ground contact is documented yet (check the Performance asset's details on first use). Fix in cleanup instead:

  1. Maya route: HumanIK foot pinning and hip height, as in Tutorial — HumanIK Mocap Cleanup for MetaHuman.
  2. Unreal route: a Control Rig with foot IK to the floor plane, baked in Sequencer. The Animation Mixer (5.8, experimental) can layer a foot-fix pass over the solve.
  3. Standing idles: often the simplest fix is to lock both feet for the whole clip and let the pelvis carry the weight shift.
  4. Seated takes: pin the pelvis to the chair seat as well as the feet.

8. Test against a suit on the same take

Run this once before we decide which route each asset type uses.

  1. Same performer, same take, both systems: the inertial suit records while the static camera films. Jam both to one timecode.
    • The suit's straps change the silhouette slightly; note it.
  2. Material: a 60 s standing idle, a 60 s seated idle, 10 gestures from Tutorial — Gesture Library and Tags, a 20 s walk and turn.
  3. Process the video with the markerless plugin; retarget the suit data to the same MetaHuman.
  4. Compare side by side in one Level Sequence with the reference video.
Measure How Pass mark (our proposal)
Foot contact Frames with visible float or slide per minute ≤ 5 after 30 min cleanup
Hand and finger fidelity Director rates gestures 1–5 against reference ≥ 4 average
Jitter Visible pops at kiosk scale, life-size None after light filter
Cleanup time Minutes of animator time per minute of footage Within 1.5× of the suit
Processing time Wall-clock per minute of footage Record only
  1. Decide per asset type: idles, gestures, monologues. Record the decision in Mocap — Process via the technical director.

9. Practice exercise

Film a 20 s clip of yourself on a phone at 60 fps against a plain wall, full body in frame. Process it, export to a test MetaHuman and list every frame where the feet float. Fix one foot contact with Control Rig.

10. Check before using it for a delivered asset

Sources