GDC2011: Fast and Efficient Facial Rigging

Talking about the Face Rig GOW3 in Gears of War 3 there are

a lot of characters and a lot of cutscenes, which brings some challenges:

• Emoji animations must be easy to edit
• A good workflow can be established on the basis of outsourced production
• Animation data can be shared with different models

Easy editing is reflected in the fact that all nodes are a curve, which also ensures the simplicity of the data and facilitates outsourced data transmission

Article inspired by MGS4 XSI

Rig is divided into 4 layers:

1. Cage: A low-poly geometry that defines most of the deformation area of ​​the face. Because of the low number of faces, deformation and iteration can be performed quickly.
2. Locator: some fixed points on the surface of the cage, similar to the key points used in facial motion capture
3. Offset Rig: moves with the Locator, so adjusting a Pose is adjusting these offsets
4. Joint: That is, bones, constrained by Offset Rig

A FACS (Facial Action Coding System) is mentioned here, which is very informative: that is, the face can use 32 action units to combine all facial movements. An action unit can be defined as the contraction or relaxation of one or more facial muscles.

Through the deformation of Cage (Morph, a common animation production method in DCC)
, the movement of the Locator Locator drives the Offset Rig control point, and the offset can also be calculated. The
Joint is a child node of the Offset Rig. Adjusting the offset will drive the The movement of the Joint, thus affecting the vertices of the skin.

Because the Face Rig structure of each character is the same, it is easy to build a Pose library, and these data can be reused between different characters

The principle of moving to another face: The new face is regarded as the Morph Target of the old face, and the obtained deformation data can be superimposed on all expressions in the Pose library, thus completing all Pose conversions.

Some blemishes can be corrected by overlaying a distortion

Use wrinkle map to add some wrinkle effect

Of course, there is still some room for improvement in this scheme:

• Weight skinning is a bottleneck
• Making Correction Distortions is a Bottleneck

The solution is to use a common head model:

if you need multiple different faces, make multiple different correction morphs. This way you don't have to skin each face, and you don't have to do the corrections individually.

Sigh: Epic's technical art is too NB~