6.2 is now available

[Treyrannosaurus-Rex]

Our commitment for Unity 6 is to provide a faster, more reliable, and more stable Engine. We’re excited to announce that Unity 6.2 is now production-supported.

Get access to this and more by downloading Unity 6.2 today, or see more details and bug fixes in release notes.

Unsure how to go about upgrading? Check out the upgrade guides to help you go from previous Unity releases to Unity 6.2. For complex productions with a high number of dependencies, find out how our Success Plans can ensure the upgrade process goes smoothly.

Highlights from the Unity 6.2 release

You can find a full list of what’s new in Unity 6.2 here, but take a look at some highlights below and join us live on stream on August 19th to break it down.

• Unity AI Beta: Make creating with Unity faster and easier. Natively integrated in the Unity Editor, the Unity AI Beta is included as part of the production-supported Unity 6.2 release. Unity AI provides contextual assistance, automates tedious tasks, generates assets, and lowers the barrier to entry.

  • All users who install Unity AI get free, unlimited Unity Points during the beta. Points will expire when Unity AI transitions to a paid model in the future. In the meantime, we’d love your feedback. Check out the roadmap and share your suggestions.

• Android XR updates in 6.2: Unity 6.2 introduces key updates to strengthen Android XR development. Unity’s Android XR package transitions into a verified state with 6.2, providing developers with a stable and production-quality foundation for building Android XR applications. We strongly recommend upgrading to Unity 6.2 as the verified package aligns with the latest engine improvements and offers the most robust support for Android XR. Notable updates include:

  • Hand Mesh Functionality: Users can now visualize hand meshes and utilize them for occlusion.
  • Dynamic Refresh Rate: Support for dynamically adjusting the display’s refresh rate during runtime ensures smoother performance.
  • Visibility Mesh Occlusion: This feature reduces GPU overhead for post-processing effects in URP, making it more practical to apply image effects like color grading and vignetting on untethered XR devices.

• Graph Toolkit is an API framework that empowers you to build custom editor-based graph tools.

• Mesh LOD is now available to automatically generate levels of detail at import to reduce iteration times.

• World Space UI for UI Toolkit is now available to render UI directly in world space for immersive XR and gaming environments.

• Developer Data Framework: Our recently announced Developer Data framework, reflects our commitment to providing developers with transparency and control over their data usage within the Unity ecosystem. This transparency and control will remain pivotal as we release additional live service capabilities from within the editor.

• New Diagnostics: Unity 6.2 brings new diagnostics features that help you monitor performance and stability across devices in real time. With these new features, you will gain enhanced crash and ANR reporting to diagnose issues faster, see how your game performs across diverse device types and specs, and more.

See more details in the release notes.

Remember that Unity 6.2 is a Supported release, which means it receives the same level of support as an LTS (including bug fixes and critical platform updates) until the next release is published. Being able to confidently upgrade to the latest Update means you can always keep your productions on the most current release without sacrificing stability or performance.

Still have questions? Learn more about release support.

Note: If you’re capturing screenshots or videos of the Unity Editor, follow these guidelines from our online evangelism team.

[LeonhardP]

Hi @TheCelt, please ask specific questions like this in a dedicated topic. Mesh LOD topics are covered by the Mesh tag.

[stonstad]

@Treyrannosaurus-Rex

Our commitment for Unity 6 is to provide a faster, more reliable, and more stable Engine.

Hi Trey!

I’d really appreciate an editor that doesn’t crash several times a day. This is a known issue associated with using multi-scene physics. Not a single LTS version has a fix.

Open For 8 Months:

Unity IssueTracker - Crash on CleanupAfterLoad() when exiting Play Mode...

[altepTest]

I’ve enjoyed the AI feature presentation. I know is a few months old but is a preview of great things to come in the development world. Yes, is still in its infancy in the way is controlling the editor can’t do too much. But give it a few years and that will change for the better.

Now I know the point of this is to make money so I understand the pay for use option. But I believe is a losing method on the long run.

These engine AI integration will be replicated by other platforms especially open source and those would allow local AI models. Those that I have on my machine are far superior in generating uncensored data which I suspect is not allowed by unity.

You should plan for the long run, provide tools to integrate unity with the AIs models people already have (locally or in other places) so they can stick with your software in the long run. You will never be able to beat uncensored models.

The biggest aspect of AIs that I can control locally, is that I can add my own information to their knowledge database. For people that are new to this. For example asking unity AI (or other proprietary models) to generate specific things that the models don’t know about is not supported. Let say I’ve create a particular character with certain clothes and hair and all. Because I can “train” my local models to know about my custom character it will replicate it with good detail and precision when asked about.

[JiRo_Dev]

Apparently it’s fixed, and they’re considering to backport it to 6.1 and 6.2

[Jonathan-Westfall-8Bits]

Think they are preparing a backport of the fix that was done in the Unity 6.3 alpha for the next alpha release. It could be in the stable versions of Unity within a couple weeks if all goes well.

[hamzakhan201021]

the best feature in this update is Mesh LOD

[bugfinders]

awesome work!! did my eyes deceive me or did it not ask to install vs community?

[zulo3d]

Because of all the hostility there is towards AI I suspect you’ll make much more money if you add an option for users to pay you not to add AI to Unity.

[stonstad]

@JiRo_Dev @Jonathan-Westfall-8Bits

Absolutely, I hear you. But hey, it’s not “fixed” until it shows up in a stable release…

I just want the editor to stop crashing. There is literally no supported release available with a fix.

[altepTest]

honestly I understand the hate against AI, because it destroys everything creative, and we are not even at the worse point ,as I will explain this lengthy post. but for now is still does have big advantages

Recently I was trying to figure out an operation to do in Blender for a 3D model. But could not figure out how to do it. Searching the web didn’t worked, google search could not figure out what I was asking. Reading a bunch of forums proved unsuccessfully.

For some reason, I’ve had a mental block, I’ve kept trying these old ways of figuring stuff, searching web and browsing forums. Then it hit me.

I’ve took a screenshot of blender window with the object and I’ve put that into a chat prompt on Gemini, asking how do I do this? Lo and behold, two seconds later I’ve got back a clear explanation using my actual layers names and indications where to find all the tools and how to do what I wanted to do.

No sane developer should look down on that “fix my code” functionality that unity had shown. You don’t know, I mean you do know, how many hours one can lose for some small bug that should had worked, asking on forums, delaying the work, pulling your hair out. If you got lucky then you figure and get even more annoyed because it was some random stupid thing. These AI models can identify all these major time destroyer problems immediately.

Unfortunately unity is at the moment on the wrong path with the closed source approach. That is because all these powerhouse AI companies like google will roll out some sort of interaction over your applications in real time.

So for example google screenshare where at the moment it only watches your screen in real time and answers on questions based on that. I have no doubts they will have some interface to control the mouse and keyboard and apply work on the application. It already knows where all the tools are, it can identify in real time what state they are in, he knows all the key shortcuts.

Instead of asking “how do I do this” you will just say “do this then” and the model will take control of the computer, open Photoshop, create the UI textures, export in the right place, generate the code. Then that will be like the end for computer jobs.

I guess unity could try to protect their application and their AI by somehow hiding the interface and the functionality behind a prompt, I don’t know, I don’t see unity keeping up with google and openai. Why should unity worry then? well if you have 10 seats you just fire all of them, keep one seat, then use the AI to work at same speed those 10 human would had worked. Why pay for 10 unity seats lol.

[zulo3d]

@altepTest

I don’t know, I don’t see unity keeping up with google and openai

Yes, AI is going to take Unity’s AI’s job. In future even AI is going to hate AI.

[andreiagmu]

@altepTest

You should plan for the long run, provide tools to integrate unity with the AIs models people already have (locally or in other places) so they can stick with your software in the long run. You will never be able to beat uncensored models.

I haven’t tested this, but I think you can use the Unity AI Inference Engine (formerly known as Sentis) to run local models, both in the Editor and in runtime.

(See Inference Engine section at the bottom of the Guiding Principles page)

Unity AI Guiding Principles

[Thomas-Mountainborn]

AI natively integrated with the editor? No, thank you very much. I refuse to use any generative AI as it’s the world’s grandest theft. It’s deeply unethical. How long until Unity starts training on our assets and scripts, without our express consent? I want gen AI stuff out of my editor by default.

[RadialMoss]

I tested this new 6.2 version to create an Android build, but it failed. Is this a new bug, or am I doing something wrong?

FAILURE: Build failed with an exception.

* What went wrong:
A problem occurred evaluating project ':launcher'.
> Failed to apply plugin 'com.android.internal.application'.
   > The option 'android.enableDexingArtifactTransform' is deprecated.
     The current default is 'true'.
     It was removed in version 8.3 of the Android Gradle plugin.
     If you run into issues with dexing transforms, try setting `android.useFullClasspathForDexingTransform = true` instead.

BUILD FAILED in 3s

Edit: I could fix with this code:

#if UNITY_ANDROID && UNITY_EDITOR
using System.IO;
using System.Text;
using UnityEditor.Android;
using UnityEngine;

namespace PreBuildSettings.Editor
{
    public class FixUnityGradlePreBuild : IPostGenerateGradleAndroidProject
    {
        public int callbackOrder => 9999; // ensure it runs after everything else

        public void OnPostGenerateGradleAndroidProject(string path)
        {
            GradleProperty( path );
        }
        
        private void GradleProperty( string path )
        {
#if UNITY_6000_0_OR_NEWER
            /* When using Unity 6+: The build process fail if the line
             'enableDexingArtifactTransform' is present in the file 'gradle.properties'*/
            
            Debug.Log($"[{nameof(FixUnityGradlePreBuild)}] Post-processing Android project: removing " +
                "'enableDexingArtifactTransform' from gradle.properties");

            var stringBuilder = new StringBuilder();

            // Path to gradle.properties
            var filePath = Path.Combine( path, "..", "gradle.properties" );

            if (File.Exists( filePath ))
            {
                // Load in all the lines in the file
                var allLines = File.ReadAllLines(filePath);

                foreach (var line in allLines)
                {
                    if( line.Length > 0 )
                    {
                        // Add everything except enableDexingArtifactTransform
                        if (!line.Contains("android.enableDexingArtifactTransform"))
                        {
                            stringBuilder.AppendLine(line);
                        }
                    }
                }
            }
            
            // Add in line to set useFullClasspathForDexingTransform to true (replacement for 'enableDexingArtifactTransform')
            stringBuilder.AppendLine( "android.useFullClasspathForDexingTransform=true" );
			
            // Write out the amended file
            File.WriteAllText( filePath, stringBuilder.ToString() );
#endif
        }
    }
}
#endif

[Qleenie]

How do we activate DLSS4 Super Resolution mentioned here: Unity - Manual: New in Unity 6.2? I tested it, and it seems it still has DLSS 3.7.20 by default, is there a way to switch to v310.3.0 mentioned in the full list?

[soleron]

Congrats for managing to complicate something as straight forward as AI prompts

Why do I have to use switches like /ask when I am asking a question? Why do I need to use /run when I do not really want to run anything in order to make placements?? etc.

All I should have to do, should be attach an object and type “put this all over the place”

Or the simple existence of a question mark in the end of a phrase should be a question like normally is when we write.

What purpose do they serve? They defeat the purpose of modern AI.

[soleron]

@altepTest

with google

with google? nobody needs to keep up with google’s AI. Gemini is the worst.

[altepTest]

@andreiagmu

I haven’t tested this, but I think you can use the Unity AI Inference Engine (formerly known as Sentis) to run local models, both in the Editor and in runtime

ok so at first it looked really promising although one needed to create (or wait for someone to figure out) libraries to support unity API.

but looking closely I’m afraid at the moment it doesn’t do any of what I was hoping it will do.

on this page they mention some limitations

Supported models | Inference Engine | 2.3.0

I would loved to know what those things are by myself but honestly no one knows so I’ve asked gemini to tell me if SDXL or FLUX models can work with Unity AI Inference Engine. The answer is no.

Here’s why you cannot use any of these models directly in the Unity Inference Engine:

• String Tensors: All models (SDXL and Flux) rely on text prompts as their primary input. The Unity Inference Engine explicitly states that it does not support string tensors. This is a fundamental incompatibility. The text prompt (“a futuristic cityscape,” “a cute pony in a field,” etc.) must be converted into a numerical tensor format for the model to process it. Since the Unity Inference Engine cannot handle the string input directly, it cannot perform the first, most critical step of the generation process.
• Sparse Tensors: While it’s possible that a text-to-image model’s internal workings might not require sparse tensors, the frameworks they are built on (like TensorFlow and PyTorch) have extensive support for them, and they are often used for efficiency. The Unity Inference Engine’s lack of support for sparse tensors is another potential point of failure, especially if the model’s ONNX representation (the format Unity typically uses) includes sparse operations.

The other limitations (more than eight dimensions, complex number tensors) are less likely to be an issue for these specific models, but the lack of string tensor support alone is a deal-breaker.

In short, because SDXL and Flux are designed to be prompted with text, and the Unity Inference Engine cannot handle text input directly, you cannot use these models “as is” with that engine.