Unity ships Ziva Real-Time 2.0
Machine-learning-based toolset Ziva Real-Time is intended to create real-time 3D characters with the fidelity of those used in visual effects, complete with accurate muscle and soft tissue deformations.
Unity has released Ziva Real-Time 2.0, its AI-based system for generating high-quality real-time 3D characters, complete with accurate soft tissue deformation.
Previously only available in beta, the software uses source files showing a character in a range of poses to train a machine learning model to learn how to deform and skin the mesh.
As well improving existing real-time character rigs, the system can be used to generate real-time rigs that behave in the same way as offline rigs created for visual effects work.
The resulting data can be used inside Maya, Unity or Unreal Engine, enabling “film-quality characters to be deployed in real-time applications”.
A real-time character deformation system already used in production in videogames
Originally known as ZivaRT, before original developer Ziva Dynamics was acquired by Unity, Ziva Real-Time has already been used in production on projects like Insomniac Games’ Spider-Man: Miles Morales.
Unity’s blog post on the software pitches it at VFX and episodic animation as well as game development.
Now available outside the old ZivaRT beta program
As far as we’re aware, the release of Ziva Real-Time 2.0 marks the first time that the technology has been widely available: ZivaRT having been only available in beta.
A machine-learning-powered alternative to traditional skinning workflows
Ziva Real-Time takes a 3D character supplied by the user, and uses machine learning to predict how the mesh would deform in motion, including the deformation of soft tissues like muscles and skin.
Unlike Ziva Face Trainer, its counterpart for facial animation, the training process is done offline, in the Ziva Real-Time Trainer software.
Users supply the data on which to train the AI model: an Alembic file showing the 3D character mesh in a range of poses, plus a FBX file showing the joint hierarchy of the character rig in the same poses.
Pose shapes can be created by hand in DCC software, or come from scan data of live actors.
As well as human and humnoid characters, the software can be used for creatures: the cheetah shown at the top of this story comes from the free sample assets available in the Ziva Dynamics online store.
Can be used inside Maya, Unity and Unreal Engine via the Ziva Real-Time Player
Once training is complete, Ziva Real-Time Trainer exports a ZRT file: a lightweight data file that be processed by the second component of Ziva Real-Time, the Ziva Real-Time Player.
Available as a plugin for Maya, Unity and Unreal Engine, the Player makes it possible to drive the deformation of the character inside the host application.
According to Unity, “even the most complex [ZRT] character and creature assets are under 30MB”, so they can be processed at “sub-millisecond speeds”.
Price and system requirements
Ziva Real-Time Trainer is compatible with Windows 10+ and CentOS 7.6-7.9 Linux. The Ziva Real-Time Player plugins are compatible with Maya 2019+, Unity 2021.3+ and Unreal Engine 4.26+, all on Windows only.
The software is rental-only, with node-locked licences costing $1,800/year and floating licences costing $1,900/year, both including licences of the Maya Player plugin.
Additional licences of the Maya Player plugin cost $100/year for a node-locked licence; $200/year for a floating licence. The Unity and Unreal Engine Player plugins are free.