Unity 6.2 Beta is now available

Hi everyone,

We’re excited to announce the first Unity 6.2 beta release, with a continued focus on Engine stability and performance, and invite you to try the new features we’ve rolled out. Your feedback is important to getting Unity 6.2 production-ready. We value your input and will have a team of engineers and customer QA folks available here on Unity Discussions to engage with you.

Introducing Unity AI

With the Unity 6.2 beta, we are introducing Unity AI – our integrated suite of AI tools designed to assist with your development workflow directly within the Unity Editor. When Unity 6.2 enters GA (general availability), Unity AI will replace both Muse and Sentis, providing improved features, better Editor workflow integration, expanded model choices, and more flexible pricing.

Note that all Unity AI features are completely optional to install and use.

Unity AI consists of three main components:

  • Assistant – A contextual helper that answers questions about your project, generates pre-compiled code, and can execute agentic actions like batch renaming files or placing multiple objects in a scene. Drag project assets into prompts or query selected console errors. Assistant replaces Muse Chat.
  • Generators – Improved tools for creating sprites, textures, animations, and new sound generation capabilities. Generators publish directly into supported Unity assets for a seamless workflow. Generators replace Muse Sprite, Texture, and Animate
  • Inference Engine – Inference Engine continues to focus on local AI model inference performance gains for unique runtime experiences. ML-Agents will be updated soon to be compatible with Inference Engine. Inference Engine replaces Sentis.

Interested in trying Unity AI? Here’s what you need to know:

  • In-Editor installation – Install Unity AI through the “AI” button at the top of the Editor after downloading Unity 6.2 Beta, and reference our documentation to get started.
  • Free during Unity 6.2 beta – All users who install Unity AI will receive free Unity Points to try the features. Free points will expire at the end of the beta, when Unity AI will transition to a paid Unity Points model.
  • Data usage and privacy – When you use Unity AI during this initial Unity 6.2 beta release, data related to your use of Unity AI including prompts, responses, and interactions will only be used to provide the AI service for your exclusive benefit, and will not be used for model training purposes. For example, if you upload a reference image to generate a truck sprite, we do not use your source image to train or improve the generative image model. In a future Unity 6.2 beta release, you will be able to manage your data settings, including whether to share your Unity AI data for AI model training purposes. We will share an update here at that time.
  • Regional Availability - Unity AI is available in most global regions, but currently unavailable in certain regions, including Russia and China. For Unity customers located in China, please refer to this announcement.

For complete details on Unity AI features, the transition from Muse and Sentis, and information for existing Muse subscribers, please visit the Unity AI FAQ.

Other highlights

MeshLOD

  • Mesh LOD – Automatically generate levels of detail (LOD) at import, optimizing assets and reducing iteration times without the need for external tools. By storing all LODs within a single Mesh, it minimizes memory usage while simplifying workflows for both static and skinned meshes. Ideal for high-poly models, it ensures seamless scalability and improved performance. See our dedicated Discussions topic for more information.

  • World Space UI for UI Toolkit – Render UI directly in world space for immersive XR and gaming environments to power richer UI experiences.

Try Unity 6.2 Beta

Unity 6.2 beta provides an opportunity to evaluate these new tools while providing valuable feedback on how they can better serve your development needs. Check out the release notes for a comprehensive list of features.

Download Unity 6.2 Beta

Our alpha and beta releases are open to anyone, so no signup is required. Get started by downloading them from the Unity Hub. Because there may be feature stability issues with early test versions, we do not recommend them for projects in production, and we highly recommend that you back up any project before opening it with an alpha or beta release.

As a reminder, Unity 6.1 remains the current Supported Update release during this beta period. When Unity 6.2 enters GA (general availability), it will become the only Supported Update release.

Through the 6-2-beta tag, you can hear about further announcements, and discuss the 6.2 Beta with us and the wider community on Unity Discussions. Our goal in sharing beta versions is to identify and resolve any remaining issues to improve overall quality before the official release.

We will announce all 6.2 beta releases in our dedicated release announcements topic here on Unity Discussions. You can choose to receive notifications for new 6.2 beta releases via the bell icon on the right.

If you have any questions or other feedback regarding the beta and its features, be it positive or negative, please let us know about it.

Thanks for participating in the beta!

Latest Release
6000.2.0b12
6000.2.0b11
6000.2.0b10
6000.2.0b9
6000.2.0b8
6000.2.0b7
6000.2.0b6
6000.2.0b5
6000.2.0b4
6000.2.0b3
6000.2.0b2
6000.2.0b1

39 Likes

Sound generation may be just what I currently need..
Interesting!

8 Likes

Post-processing filters like blur effects or color shifts for sub-sections of your UI are fully integrated with the render chain, ensuring visuals remain sharp at any size or distance.

Can you share more about this? I don’t see anything about filters in release notes and I’m pretty sure they were planned for 6.3 release at best.

You are absolutely correct. This is aiming to land in 6.3. I’ve removed the sentence from the post. Thanks for flagging.

6 Likes

Are there any limits to AI image creation now?
As far as I remember it supports pixel art and upscaling right?

2 Likes

During the beta, asset generations and assistant queries are free. And yes, we now support some cool post-processes like pixelization, upscaling, remove background, and recolor. Let us know if there are more that might be useful!

6 Likes

Any plans for custom materials for UI Toolkit?

1 Like

Is there any chance of a UI converter in the future for transitioning to ui toolkit? Also please consider adding sonnet as a provider

1 Like

there arent release notes here (well not yet) the download is, sure, but the notes arent.

1 Like

Hi @Murgilod, we’re actively working to deliver support for custom shaders and post-processing filters later this year. Read more about our current roadmap here.

4 Likes

They’re collapsed by default, but they’re there, just under the Component Installers section.

5 Likes

I’m working with ML_Agents. Is your new pipeline ready to work with AI to integrate with them?

That sounds fair…

now this:
It is the developer’s responsibility to determine the appropriate AI-based declarations to make with app stores, distribution platforms, and end-users when you ship a commercial project that has used generative AI.
i think is better to show it here, because most indies are probably not going to read the FAQ about AI (which is not your fault)

now can i ask, if our data is not going to be used to train unity’s ai models, then how those ai models are going to be improved? :thinking:

5 Likes

When can we expect some lighting improvements? For example, real-time global illumination?

8 Likes

The ease of use of the AI tools is really good. While the generation features in my first tests are not always at the output quality level that I am looking for they are nicely done and having this split into multiple specialized models is a nice idea also for brainstorming.

The AI chat is surprisingly good for all kinds of questions I threw at it. And I like that it integrates Asset Store knowledge which is a great touch.

Can we expect some sort of 3d asset generation coming as well, with the quality of what Meta just announced for their v2 of 3d asset generation? This would be amazing for prototyping.

4 Likes

Hmm there werent categories there earlier… under release, it was blank

re: AI-based declarations → This is pretty commonly understood and we try to keep these posts brief. It’s there in the FAQ for anyone who needs it :slight_smile:

re: improving / fine-tuning our AI models → During the beta, you are right that we turned OFF using your prompts and responses for improving our models by default. Note this is the opposite of every other company as far as we can see. We did this to be conservative with our community and seek an understanding of expectations for what the default should be and how data controls should work. We will be launching some controls in a new AI cloud dashboard during the beta that will provide options for you to control data sharing for model training, as well as options for enabling/disabling the AI features. We know some users will be OK to provide their data to help us improve the Unity AI models for improved accuracy, and some will not be OK with that. Not forcing (defaulting) all users to do something that might be controversial is more important to us than improving the AI models. If you have a further opinion on this please DM me.

8 Likes

Thanks for the honest feedback. Anything you found that the Assistant was not so good at?

re: 3D → We are indeed working on integrated 3D mesh and texture gen, as well as skybox gen, and a few more asset types too. 3D is still imperfect for production use, but it’s very good for prototyping, and getting close to production quality in the near future. Look for an update later this year hopefully.

6 Likes

re: ML Agents → It’s not updated yet to work with Inference Engine, but we are getting to it soon. For now, you can continue to use Sentis to run your MLA models without much difference. We’ll publish an update when MLA is integrated with IE by default :slight_smile:

3 Likes

Here are two tests. One a 16-bit pixel “Wall” and it gave out not images for a wall but a wall full of 16 bit images. The text in the images looking really bad though.

Second one an animation where I tried to get somebody “programming” and it did some really weird hand motion stuff. I was a bit reminded of the meme video where the security guard at a stadium scans people but doing some really generic movements that don’t catch anything :smiley:


3 Likes