Neural colliders, Mesh renderer and more. Key updates for Zibra Liquids 1.4.4

Two months ago, we released a major update for Zibra Liquids, our tool for real-time liquid simulation. You can find the information about all of the changes for Zibra Liquids 1.4 – including Force fields, Liquid initial state baking and other features – via this link.  

Now, another update of the Unity version – Zibra Liquids 1.4.4 –  is out. Let’s talk about what’s changed.  

What’s new in Zibra Liquids 1.4.4 

The newest version of Zibra Liquids supports the same number of build platforms as the previous one, with one exception – it now has Windows DX 12 support. In general, the full list of technical requirements currently needed for Zibra Liquids to work smoothly looks this way:

  • Editor platforms: PC (DX11/DX12, x86 and x64), macOS(Metal, x64 and arm64) For M1 Mac users, the M1 native version of Unity Editor is strongly recommended.
  • Unity version: 2019.4 or later (use the latest feature update)
  • Build platforms: PC(DX11/DX12, x86 and x64), UWP(DX11/DX12, x86 and x64), macOS(Metal, x64 and arm64), iOS (iOS version 12 or later, iPhone 6s or newer)
  • Supported Render Pipelines: SRP, URP, HDRP
  • URP: 7.5 or later
  • HDRP: 7.x or later

Mesh render

Now, let’s move on to the major updates. The newest version of Zibra Liquids has a new liquid renderer.

Previously, we rendered liquid as many particles, which we subsequently blurred so that it looks like a liquid. This allowed us to render millions of particles on PC hardware in real-time. But with the move to mobile, it was no longer the best option. While PC GPUs have huge memory bandwidth allowing blur to be very fast, on mobile devices such an operation suddenly becomes very expensive in terms of performance.

Due to this fact we implemented a new renderer, one that utilizes a mesh extraction and a rendering algorithm. You can read about how it works here, but in short, it takes the data from the simulation grid and produces a mesh using the Dual contouring algorithm.

This mesh consists of triangles, the number of which can vary depending on the amount of liquid. These triangles are adjusted to better resemble the liquid’s surface. Then our custom shader renders the mesh. 

Compared to particle rendering methods, rendering the mesh allows for a much better scaling of the performance with the resolution. It also helps to improve the plugin performance. 

That’s why Zibra Liquids 1.4.4 shows up to a 3-4x better performance on mobile devices than the previous version. The new version also provides much better visuals, especially refraction. On top of that, 1.4.4 introduces underwater rendering, which will allow you to use liquids in even more situations.

Check out the difference. 

With Zibra Liquids 1.4.4, it is now possible to render your liquid in 100% resolution on a mobile, and to have much smoother framerates.

Neural colliders

Another important update are the Neural colliders. Technically, it’s not a new feature per se. As you may already know, our solution supports two types of liquid colliders – analytic and AI-powered ones. 

Analytic colliders allow you to add simple geometry, such as spheres, boxes, capsules, toruses, and cylinders as liquid colliders. There, collisions are calculated with analytic functions. AI-powered colliders, on the other hand, are based on our custom technology of a neural SDF representation and enable the adding of arbitrary geometry as a collider for liquid. 

It is no longer necessary to divide complex shapes into simpler ones. Liquid interaction is realized with any complex object as a whole and can be adjusted in just a few clicks. 

For a long time, our cutting-edge AI-powered colliders were called Voxel colliders. But recently, we changed their name to Zibra Neural Colliders. Here is why: 

Zibra Neural Colliders VS. Voxel colliders

Compared to the old voxel colliders, Zibra Neural Colliders use a new, more advanced and performant neural network architecture.  It is almost as good as sampling SDFs from a 3D texture and takes only about 1.8 nanoseconds per sample on an RTX 2060 mobile. 

They also allow for a greater level of compression while taking up to 300x less memory  compared to standard 3D texture with the same quality. You can see the difference in this picture. 

The functionality of our neural colliders remains the same for the most part. But with our new neural network architecture, you can add even more AI-powered liquid colliders to your scene. The new neural network also handles high-detail objects much better, including those with really thin walls. 

Overall, it grants a much higher quality SDF, quite similar to one computed directly from mesh, while also being much faster and requiring less memory.

Apart from mesh renderer and neural colliders, Zibra Liquids 1.4.4 has many other updates – for example, we added orthographic camera support and improved emitter UX. You can check the full list of improvements here. And don’t forget – if you have any ideas on how to improve our plugin, feel free to share them via email, support@zibra.ai, or in our Discord community. In terms of development, we are following our roadmap, but your feedback helps us to correctly choose the priority of our tasks.