Facial Motion Capture From Video: Free Phone-Based MoCap Guide

Watch the original facial-capture tutorials

Both players use static YouTube iframes and require no runtime JavaScript. If a browser, region, network or local file viewer blocks embedded playback, use the red direct-link buttons. All article images are embedded in this HTML file.

What facial motion capture produces

Facial motion capture converts visible expressions into time-varying controls. In a blendshape workflow, each control represents a movement such as jaw opening, eyelid closure, eyebrow raising, lip compression or cheek inflation. The animation is the sequence of coefficient values over time—not the final character mesh.

Apple documents ARKit face expressions as a dictionary of named coefficients associated with specific facial-feature movements. Those coefficients can drive 2D or 3D characters. QuickMagic's OnlyFace documentation describes an ARKit-compatible 52-blendshape workflow.

Understand the source and target facial channels

Minimum useful validation set

• jawOpen
• eyeBlinkLeft and eyeBlinkRight
• mouthSmileLeft and mouthSmileRight
• browInnerUp
• mouthPucker
• At least one cheek/nose expression
• Eye-look channels when the target supports them

Three mapping cases

Blender script: create drivers for matching shape-key names

The script below creates clamped drivers for every target shape key whose name exactly matches a source shape key. Duplicate the Blender file before running it and change the two object names.

# Blender: create drivers for matching facial shape-key names
# Edit these two object names before running in Blender's Scripting workspace.
import bpy

SOURCE_OBJECT = "QuickMagic_SourceFace"
TARGET_OBJECT = "Target_Character"
GAIN = 1.0

source = bpy.data.objects.get(SOURCE_OBJECT)
target = bpy.data.objects.get(TARGET_OBJECT)

if source is None or target is None:
    raise RuntimeError("Source or target object was not found.")

if not source.data.shape_keys or not target.data.shape_keys:
    raise RuntimeError("Both objects must contain shape keys.")

source_keys = source.data.shape_keys.key_blocks
target_keys = target.data.shape_keys.key_blocks

created = []
skipped = []

for target_key in target_keys:
    name = target_key.name
    if name == "Basis":
        continue
    if source_keys.get(name) is None:
        skipped.append(name)
        continue

    # Remove an existing driver on the target value, if present.
    try:
        target_key.driver_remove("value")
    except TypeError:
        pass

    fcurve = target_key.driver_add("value")
    driver = fcurve.driver
    driver.type = "SCRIPTED"

    variable = driver.variables.new()
    variable.name = "src"
    variable.type = "SINGLE_PROP"
    variable.targets[0].id_type = "KEY"
    variable.targets[0].id = source.data.shape_keys
    escaped = name.replace("\\", "\\\\").replace('"', '\\"')
    variable.targets[0].data_path = f'key_blocks["{escaped}"].value'

    # Clamp to the common 0–1 target range.
    driver.expression = f"min(max(src * {GAIN}, 0.0), 1.0)"
    created.append(name)

print(f"Created {len(created)} drivers.")
print("Skipped unmatched target keys:", skipped)

After running the script

1. Scrub the timeline and confirm the target animates.
2. Review the printed list of unmatched target keys.
3. Manually map differently named channels.
4. Change GAIN or individual driver expressions where needed.
5. Test left/right channels carefully.

Optional Human Generator workflow

Human Generator is a paid Blender character-generation add-on, not a QuickMagic requirement. The current marketplace listing supports modern Blender versions, but purchase tier, Blender compatibility and transfer tools should be verified on the current product page.

1. Install Human Generator through its documented Blender add-on workflow.
2. Create or load a character with a facial shape-key system.
3. Import the QuickMagic source FBX.
4. Compare source and target shape-key names.
5. Use the add-on's documented transfer feature when it explicitly supports the source data, or use Blender drivers.
6. Adjust target-specific expression strength and corrective shapes.

Clean and calibrate the facial animation

Expression too weak

Increase the channel gain or use a steeper response curve on that specific expression.

Expression too strong or clips the face

Clamp the target value and reduce the gain. Check whether multiple shapes are over-combining.

Blinks are incomplete

Verify eyelids were visible in the video, then tune blink gain and add a corrective shape if the target eyelid deformation is incomplete.

Lip sync feels soft

Preserve jaw opening and lip closures. Avoid broad smoothing that removes consonant timing.

Left and right expressions are reversed

Test a one-sided smile and eyebrow raise, then correct mirrored source/target mapping.

Jitter

Smooth only the visibly noisy channel intervals. Preserve natural eye darts, blinks, speech articulation and asymmetrical acting.

Use the same facial data in Maya, Unreal Engine and Unity

Maya

Connect source coefficients to target Blend Shape node weights with the Connection Editor, driven keys or utility/remap nodes. Use the Graph Editor for calibration and bake the destination controls only when required.

Unreal Engine

Enable morph-target import when importing the relevant FBX. Unreal can encapsulate morph-target animation in an Animation Sequence and expose the individual morph curves. MetaHuman use requires a separate, compatible curve-mapping workflow.

Unity

Unity exposes mesh blend-shape weights through the SkinnedMeshRenderer. Animate them in an Animation Clip or set them at runtime through scripting. Confirm whether the imported FBX includes the required curves for the target mesh.

Common facial-capture problems

Final quality checklist

• The face is evenly lit and consistently visible.
• The performance begins from a neutral reference.
• Blinks, jaw, brows, smiles and asymmetric expressions are present.
• The active QuickMagic plan provides the required facial output.
• The source FBX retains its facial animation after import.
• The target has compatible shape keys, joints or controls.
• Left/right mappings are correct.
• Expression ranges are calibrated per channel.
• Smoothing does not remove speech or blink timing.
• The final exported animation is validated in the destination application.

Frequently asked questions

Related QuickMagic guides