Next version of Unity will be called Unity 6

Unity 6 will include a number of software and quality-of-life updates, such as the debut of Unity Cloud. In particular, two major generative AI tools from Unity will be featured in early access: Unity Muse and Unity Sentis. Muse is an AI-powered content creation tool that allows developers to create assets and animations "faster, without interrupting workflow." Unity Sentis is an inference engine that enables developers to integrate neural networks into their games "on any platform."

Developers will likely be keenly interested in the benefits of Unity 6, as it will be the first version of the software to charge high-earning developers a fee based on either "initial engagement" of players or a percentage of their revenue, whichever is lower. The structure of this program was introduced after the initial installation-based runtime fee was rejected by the developer community.

This is Unity's first step into a world where the company will collect money from developers post-launch, and the first opportunity to regain long-lost trust.

What's in a name? On the transition to Unity 6

The company aims to make Unity 6 "the best version of Unity that has ever existed" – a claim that will certainly be tested, as this version will have "a different business model associated with it." Comments that point to difficulties with the Unity LTS naming convention may cause some communication challenges.

Developers accustomed to working on Unity's experimental Tech Stream should expect this model to be available in some form in the coming years.

We were curious to ask about one topic we heard from developers during the Runtime Fee uproar: some developers began to feel a lack of stability in Unity's LTS versions, even though the company promised it would be the most stable version of the software. The concern was that Unity had become so stretched (and so over-invested in higher-revenue services) that it allowed LTS to stall.

The answer to this question is that the company will "pay huge attention" to the stability and performance of Unity 6, and that, as before, it intends to continue fixing bugs for older Unity versions with the LTS designation, even as it moves forward.

According to Unity's metrics, LTS stability has been "growing" year over year for the past several years. "But that doesn't mean there aren't bugs that shouldn't exist, and we need to fix them."

According to Unity developers, the most common requests are for improvements in iteration speed, performance, and stability. "I believe this should be our foundation – we constantly improve these metrics."

Are there non-invasive ways to apply generative AI?

Unlike Microsoft, which in its recent announcements eagerly discussed using generative AI for content creation, the speaker preferred to focus on more specific use cases for Unity Muse and Unity Sentis. According to him, earlier this year (when he stated that AI would be "disruptive" for the gaming industry) he was most excited that developers would use Unity Muse's natural language, which allows developers to tell the chatbot what features they are adding to their game.

Here, he enthusiastically described how these tools integrate into features like Unity Texture, explaining how developers can use Muse as a "multimodal" input to quickly find a texture mix, experiment with it, preview it on a model, and quickly modify it using Muse-based sliders.

Developers who are wary of data ownership issues in generative AI tools (which involve copyright infringement risks) should, in theory, feel confident using Muse and Sentis. "We place special emphasis on ensuring transparency about the data on which models are built," he says. "Texture uses our data – data we have rights to – and we created a synthetic dataset with variations so you have a starting point."

Image via Unity

These variations, we hope, ensure that, say, your AI-generated textures won't look exactly like another developer's AI-generated textures.

Whitten shared some interesting use cases for Sentis that he said he hadn't anticipated when early access to the technology opened earlier this year. "What surprised me was that people are using many small, custom-built neural networks to achieve various results," he said. He gave two examples: a VR meditation app and a test version of the game Pong.

In the VR meditation app, the developer was able to use Sentis and a custom neural network to measure the user's breathing and suggest adjustments to help them meditate, turning the Meta Quest headset into a "breath sensor" headset.

In the Pong example, he described how a developer used Blender to create "a million frames" of what the game could look like with full ray tracing, then created a version of the game that captured ball and paddle position data, and then used a neural network to generate frames that give it the look of a fully ray-traced game. "Anything you can think of in Unity to create such objects, they actually replaced at runtime with a system that takes inputs and runs a neural network to create that thing."