ZibraAI, the company, emerged in 2020 as a result of successfully applying machine learning to 3D processing. While conducting in-house ML research, the team developed a way to encode the Signed Distance Field (SDF) of a 3D object into compact vectors that can be then unpacked by a small neural network to restore the original SDF. For better quality, the object is divided into a sparse voxel grid, and the local geometry inside each voxel is compressed into a vector using another neural network. All of this is done in a matter of milliseconds since the proprietary method doesn’t require training new networks to represent an unseen object. This technology can be applied to game physics simulation, which natively uses SDF for particle simulation. And this is how the startup’s ML-based core technology (later also dubbed as ZibraAI) was born.
So, long story short, ZibraAI’s team came up with an idea of creating affordable tools based on their core ML technology. The tools are designed to simplify the game development process, resolve the issue of poor visual appearance, and reduce the amount of computational power used in the process. It will let game developers and VFX artists unleash their imagination and focus on creativity instead of game optimization. One of the many challenges that they face is particles and objects of complex shapes interaction simulation. Usually, they would have to divide an object of complex shape into several simple objects and then set up the simulation of interaction with each of them separately. This process takes a lot of time and effort. In general, all existing solutions fail to provide interactivity and require powerful PCs to be run on.
Inspired by the idea of empowering game creators and making the path easy for them, ZibraAI decided to bring semi-automation to the game development pipeline. Developers will see graphics improvement, in-game physics optimisation, and 3D content generation at a lower computational cost, across all devices.
Zibra Liquids — the first-ever ML-driven fluid simulation plugin available for commercial use.
So, as we mentioned before, ZibraAI is an ML-based technology that will be at the core of many tools that are currently in the works. However, there is already a plugin — Zibra Liquids — that has been available for beta-testers for weeks now. Here’s what you need to know about it.
Zibra Liquids is a real-time and cross-platform physically correct fluid simulation plugin for game developers. It is also the first-ever ML-driven plugin for fluid simulation pipeline available for commercial use. Quite impressive, isn’t it? There’s more.
Zibra Liquids: under the hood
Zibra Liquids can be used for real-time fluid simulation and the simulation of liquids interacting with animated objects of any shape, even complex ones. Needless to say, the fluid-object interaction is a killer feature as no one has done it at the same level before. It works in real-time and is embedded in the game engine, meaning developers no longer need to use expensive and time-consuming third-party software. Zibra Liquids takes a couple of minutes to install, enables new game mechanics, and allows for water to become a part of the new gameplay.
So, how exactly does it work? The plugin results from the synergy of proprietary physical solvers that allow for a physically accurate fluid simulation and ML-based neural object representations. The fluid simulation technology was also developed by ZibraAI in-house by applying the Moving Least Squares Material Point Method. The initial prototype was able to simulate 300,000 particles in 7 ms running on a GTX 1050 laptop without any optimization and with the intention to accelerate implementation up to 3x times at least. The second challenge was to implement a rendering technique for particle visualization. For the first prototype, the team developed a local voxel particle ray-trace method which is great for simulation, since it also operates with voxels on a simulation grid. This version could render fluid on the same hardware at 30 FPS which was enough for beta testing, so the company started talking to potential customers, created their first demos, and gathered initial feedback.
After combining the fluid solver and the core ML technology, the team created a Unity plugin with interface and server backend which was able to receive mesh from Unity, generate SDF, and process it using the ML technology, which packs the data into series of per voxel neural embeds which were stored on the clients’ site. This combination allowed users to easily set up the effects and reach a new level of visual quality.
The team wanted the plugin to be natively compiled on C++ to control resources, minimize CPU load, and use a full range of techniques for optimizing GPU shaders.
It included well known common approaches like:
- resource packing
- register usage optimisation
- memory access patterns optimisation
- using indirect draw/dispatch calls to reduce CPU involvement
But that was not enough and we moved further. All performance critical parts of the Zibra Liquids solver were optimized. First of all we updated our sorting algorithm with a custom radix sort sort implementation which is 6x faster than the old bitonic sort.
Secondly instead of drawing quad meshes with a custom shader for our spheres we instead use an atomic based rasterizer instead of quads.
The last optimization is the particle to grid pass of the MPM algorithm, where we also use an atomics + radix sort approach, all in all that allows us to get a ~6x performance boost on particle to grid.
In total the new solver is about 3.5 times faster than the previous implementation.
Zibra Liquids key features:
Ease of use
The focus is on giving the plugin users the smoothest transition to new technology by designing an intuitive and simple interface. It takes only a few clicks inside the game engine to set fluid up and see the desired results.
Zibra Liquids is a GPU-powered technology that simulates several millions of particles on high-end hardware, hundreds of thousands of particles on laptops, and tens of thousands of particles on mobile devices in real-time. It assures CG-level physical and visual quality in real-time and best-in-class game performance.
Multiple visual and physical presets
The technology is capable of real-time transition between physical and visual parameters, such as stiffness, boundary force, metalness, opacity, refraction, reflection, and many others. No complicated fine-tuning is required — users get the desired solution straight out of the box.
Everything is in users’ control now, even gravity. This feature brings new ideas for game mechanics that used to be impossible to create. Users can unleash their creativity and approach game design from a brand-new angle.
AI-accelerated interaction with complex objects
ZibraAI’s core ML technology allows for creating a neural SDF representation for any geometry and using it as a collider for the liquid component in a scene. This technology makes the old-fashioned boxes and assets-representing spheres a thing of the past.
This feature allows the use of fluids as a gameplay feature. It influences the game’s appearance, and it’s also essential for creating realistic simulations of liquids that can affect objects in a scene.
How game developers can benefit from Zibra Liquids?
- Working with water becomes simple as never before. Water can now be interactive, it accurately “tells” the difference between complex 3D objects, and new game mechanics become available:
- The Gravity control and Force Interaction combo allows you to move objects within the scene
- Water guns battles
- Puzzles involving water
- Water surfing
- Any magic tricks that involve water
- Watercraft races
2. The plugin is designed to be used on any platform, including PC, game consoles, and mobile devices.
3. A free version of the plugin is available for anyone who wants to pursue game development with a limited budget.
4. Creators can now predict the game performance, which is crucial as many VFX artists have to contend with their ideas being discarded as they don’t fit the computational budget. Water created with our plugin can be reconfigured for performance purposes without actually being cut out of the game.
5. The visual appearance of any game is improved due to physically accurate fluid and objects interaction simulation, which brings an immersive experience.
6. Simple UI and an easy setup process of all necessary parameters.
7. The GPU-powered technology simulates several millions of particles on high-end hardware, hundreds of thousands of particles on laptops, and tens of thousands of particles on mobile devices in real-time. Before ZibraAI, all of these computations were CPU-based.
8. The game development process is now more accessible and cheaper. The games can be developed and tested on any hardware, and users can play them even on smartphones that were released a few years ago.
What skills does one require to start using Zibra Liquids?
The good news is you don’t need any knowledge of machine learning to use the plugin, but you need to be familiar with the Unity game engine interface.
What platforms are coming next?
We’re counting down to the official release in the Unity Asset Store. After that, iOS, Unreal Engine, and Android support will be added and the plugin performance will be improved by 2x. If you want to take part in the Unreal Engine plugin beta testing and be the first to know about all ZibraAI releases, join their Discord community.
Eventually, Zibra Liquids will become a platform-agnostic solution available for any game engine, including custom ones.
ZibraAI is open to collaboration and deep integration requests from enterprises of any size.