The new tool can train any face mesh to perform complex expressions in real-time.
Ziva Dynamics, a developer of virtual character production tools, has announced the release of Ziva Face Trainer, the company's new Machine-Learning-based facial rigging service. Powered by ZivaRT, this new technology is capable of training any face mesh to perform the most complex expressions in real-time.
According to the team, Ziva Face Trainer is built on a comprehensive 15-terabytes-large library of 4D data and proprietary machine learning algorithms and comes with high-performance animatable puppets that can express over 72,000 training shapes and novel face poses. What's more, the service was tested on various heads, both life-like and fantastical, allowing you to work with almost any character you want.
"Before you upload your mesh, you must follow the simple Face Registration instructions to prepare your mesh for training," comments Ziva." Map your mesh to our generic face mesh using a combination of Point Selections and Polygon selections in Wrap3, clean your mapped mesh results in either Wrap3 or Maya, and increase the resolution of your mesh using OpenSubDiv in Maya."