Greater geometric complexity can be simulated using detail maps on the diffuse channel of the material. Monochrome maps are assigned to each expression extreme and are combined at run-time.
To simulate the real world we must reproduce the quality of light visible to the observer. The visible spectrum has a much higher dynamic range than the display devices we use to view our simulation. That means we have to start thinking about exposure and how to change it dynamically.
Rag Doll rigs in games are usually less than ideal. The limbs tend to bend and behave in unrealistic ways, and yet multiple constraints present a processor hit that most developers are unwilling to accept. The Rag Doll in these demos was set up in Maya using only a hinge constraint that operates on 1 degree of freedom per joint. The rotation limit on the joint is set by the shape of the collision geometry.
This idea was first prototyped in the Moria environment for LOTR. Based on Debevec’;s Radiance idea for image based lighting, A line of pre-rendered spherical views was strung out along the path of the Cave Troll and the ambient / bounce light was provided by interpolating between the spheres based on the Troll’s position.
Noise functions can help to reduce the mechanical, linear feeling in animation, especially with the CG camera which often feels too perfect.
For some ‘on rails’ games (racing, fps) creating the surrounding environment with geometry and maps can be overkill- especially if the gameplay prevents you from getting too close. The midground and distant game world could be pre-rendered and streamed in based on the players location on a path.
For the Barbie cinematics, I created a more intuitive control system for the facial animation. Blend shapes still formed the basis of the technique, but they were controlled by a spherical giu that helped solve incorrect combinations of blend shapes creating faulty expressions.