Feedback Request: Changes to Unity's Dynamic GI Roadmap

[StevenK_Unity3D]

We aim to provide you with the best tools and features to enhance your everyday light creation experience with Unity. To achieve this, we’d love to hear your feedback about changes to Unity’s Dynamic Global Illumination roadmap, as well as hear about your Unity 6 lighting experience. Please take 10 minutes to complete the survey linked below and help us refine and improve your Unity experience.

(this survey is now closed, thank you all for taking the time to contribute!).

Update 26th May, 2026: You can now try Unity’s first milestone to dynamic GI through Surface Cache: Surface Cache GI Preview.

---

What’s changing on our Dynamic Global Illumination roadmap?

We’re making changes to our Dynamic APV (Adaptive Probe Volumes) roadmap.

Dynamic APV is functionality built on top of APV, to enable more dynamic lighting situations. APV itself is core new functionality released in Unity 6.0 and will continue to be supported.

For some background: Adaptive Probe Volumes (APV), is a new system for indirect lighting and automates the placement of Light Probes based on scene geometry delivered in Unity 6. This vastly improves artist iteration time by eliminating the need for manual probe placement. Additionally, features such as APV Sky Occlusion and APV Lighting Scenarios allow dynamic transitions using prebaked lighting data. Effects such as time-of-day changes or toggling lights on and off are possible using Lighting Scenarios, with support for multiple pre-baked lighting setups. Sky Occlusion provides plausible sky lighting transitions using baked, multi-bounce occlusion data for static contributors.

With Dynamic APV, Unity embarked on a journey to enable even more semi-dynamic GI use cases through APV, with the “light propagation” method as its foundation.

However, despite significant effort from our engineering teams, this method ultimately failed to meet our standards for image quality and robustness.

After multiple attempts and considerable time investment, we recently made the difficult decision to change our approach. While we considered exploring an alternative light transport solution for Dynamic APV, the compounded risk of technical unknowns and rapidly evolving industry expectations made continuing with this solution unlikely to yield the best outcomes for our creator community.

With this in mind, we are pivoting from Dynamic APV to explore a new plan for delivering fully dynamic GI, with broad platform support. This shift involves developing a scalable and high-performing dynamic GI system built on modular “GI building blocks” to meet the wide-ranging needs of our audience over time. Choosing to instead focus on fully dynamic GI offers additional value through support for dynamic environments, as well as unlocking a simpler authoring and significantly faster iteration experience.

Unity’s new approach to Dynamic GI

Even though we aren’t able to commit to a specific date, our first milestone in this multi-year endeavour is a fully dynamic, diffuse-only GI system, currently based on a ray traced surface cache, that we intend to support in both the Universal Render Pipeline (URP) and the High-Definition Render Pipeline (HDRP). Initially, this system will focus on simplicity, reliability, and scalability with an emphasis on hardware compatibility. This will enable creators to engage with dynamic lighting features in medium-sized scenes on devices ranging from high-end mobile devices to platforms like PS4 Pro and above.

Building our solution based on a ray traced surface cache has the following benefits:

1. Easier workflows, little to no additional authoring steps required
2. Faster iteration, no precomputation step required
3. More dynamic, works with procedurally generated content, user generated content, and construction and destruction

While this initial feature set will include some limitations, such as reduced performance in certain complex scenarios, it sets a strong foundation for future expansions that will enhance support for volumetrics, specular and glossy light scattering, with improved visual fidelity.

What’s next?

We understand that this pivot may come as a disappointment to those anticipating Dynamic APV, but we are confident this new direction aligns better with Unity’s core values as a versatile and forward-thinking cross-platform engine.

To ensure transparency and collaboration, we are announcing these changes here early, and are inviting you to share your feedback through the survey linked above so that we can refine our roadmap together.

We are committed to making dynamic GI a robust, scalable solution and appreciate your patience and trust as we work to deliver industry-leading tools for creators worldwide. We’ll share more details as soon as we are able to.

What other Global Illumination solutions were delivered in Unity 6?

Unity 6 introduces several advancements in global illumination (GI) and lighting workflows, aimed at improving both the visual fidelity and performance of lighting in your projects.

You can read more about the Unity 6 lighting features on the blog post: Unity 6’s New Global Illumination Lighting Features.

[Michail_P]

Please add realtime indirect lighting and APV

[stonstad]

+1 for anything that brings Unity closer to Unreal Lumen-quality lighting with comparable performance. A modern approach to lighting procedural content is beneficial.

[Qleenie]

I think it makes sense to drop the idea of APV and go for a more dynamic approach, and the idea of surfels or surface cache is a tech which is well known and often used.
If it uses ray tracing to generate these cache, will it be DX12 only and will it require to build BVH?

[thunderwire]

Thank you for finally listening to your community. I’ve been begging for fully dynamic GI for years. I was committed to switching to Unreal after finishing my game, just because of missing features like real-time GI, but I might stick with Unity a little longer.

[Redrag0n]

God, I hope we can do better than a noisy and ghosty solution that requires an extra 5+ ms and a mandatory upscaling to work.

Shame about the dynamic APV, it could be a nice cheap solution where a fully dynamic GI is too expensive and iteration speed is much needed, even if it doesn’t look as good.

[Wolfos]

Good to hear this is being looked into again. With much of the industry moving to real-time GI, it feels increasingly awkward that Unity doesn’t offer a great solution for it.

We’re planning to ship with HDRP’s ray traced global illumination feature. It’s unfortunately missing specular occlusion. But it runs well. Even on the Steam Deck we’re running at native resolution. I think this bodes well for the viability of real-time GI probes.

This will enable creators to engage with dynamic lighting features in medium-sized scenes on devices ranging from high-end mobile devices to platforms like PS4 Pro and above.

@StevenK_Unity3D
Why not large scenes? The current BVH system has no problem with large scenes because you can just exclude anything after a certain distance. Surely you don’t intend to keep the entire scene in the BVH?

[Onigiri]

I love this infinite loop of switching between trying to make dynamic GI > semidynamic GI > dynamic GI again without delivering anything for ~10 years. At this point just buy H-Trace WSGI and make it official solution. This asset is better than any realtime GI you could ever create (it’s even better than another surface cache slop - Lumen).

[DrViJ]

I think it could be better to have APV first on all platforms (including WebGPU) and then to make something new, one of unity problems is jumping from tech to tech leaving them all unfinished and unpolished (Pre-tech-beta-alpha-preview version, and then oops, we deprecate it)

[Wolfos]

I don’t think they ever planned on dynamic GI? The plan was always to replace Enlighten without paying Enlighten because Unity is somehow allergic to tried and true technology these days.
At least radiance caching works.

H-Trace is not a universal solution and far from production ready. Voxels have big limitations thanks to their uniform density.

[Onigiri]

@Wolfos

I don’t think they ever planned on dynamic GI?

No, there was a DDGI attempt

@Wolfos

Voxels have big limitations thanks to their uniform density.

Yes, but H-trace devs completely solved them

[Wolfos]

H-Trace uses 800MB of VRAM to render up to a maximum of just 40!!! meters away from the camera. This is not solvable. There’s a reason other engines moved away from voxel GI. It doesn’t scale.

RTGI is the scalable solution, for big games and small, at higher resolutions and higher distances. With less VRAM consumption.

We need a solution for all games. If we were only rendering Sponza we could just bake lighting.

[Onigiri]

@Wolfos

H-Trace uses 800MB of VRAM to render up to a maximum of just 40!!! meters away from the camera

Everyone have 16+ gb ram nowadays, I don’t think this is a problem. And it works perfectly fine in open world scenarios, it fallbacks to screen space tracing after the threshold.

[Onigiri]

Don’t get me wrong, I will be only happier if they’ll create something better, but so far they failed all their attempts at this and it will be 5 more years of dev time for trying it again. So buying asset that already works and praised by community looks like decent solution to me.

[rasmusn]

@Qleenie

he idea of surfels or surface cache is a tech which is well known and often used.
If it uses ray tracing to generate these cache, will it be DX12 only and will it require to build

Hi Qleenie. I cannot make any guarantees about what we will actually ship but I can say that our current implementation is ray tracing-based. It can exploit DXR/DX12 where available but doesn’t it require it as it has an automatic software BVH fallback. This enables it to run pretty much anywhere you have compute shaders (but of course it may be slower on such platforms).

[rasmusn]

@Redrag0n

God, I hope we can do better than a noisy and ghosty solution that requires an extra 5+ ms and a mandatory upscaling to work.

Shame about the dynamic APV, it could be a nice cheap solution where a fully dynamic GI is too expensive and iteration speed is much needed, even if it doesn’t look as good.

Indeed, many modern realtime GI solutions have the major drawback that they can be noisy, they have a high performance requirements, and often only work on high-end hardware.

In our solution we want to try avoid or at least minimize these negatives by optimizing for stability, performance, and platform reach instead of visual quality.

This means we probably won’t be able to match the visual quality of other high-end solution but on the other hand we think our solution will be faster, more visually stable, and run on a broader range of devices.

For example, our current implementation does not rely on temporal reprojection in screen space. This means that the dreaded screen space ghosting is simply not possible.

So while we are not exactly delivering Dynamic APV it sounds a bit like we are aiming for something you’d like. We’ll see :).

[rasmusn]

@Wolfos

Why not large scenes? The current BVH system has no problem with large scenes because you can just exclude anything after a certain distance. Surely you don’t intend to keep the entire scene in the BVH?

Unfortunately, BVH culling is not as simple as you seem to suggest in our context. For example, we cannot “just” use “the current BVH system” (I’ll assume you by this mean Unity’s DXR integration) because this would not work across platforms. Of course this can be solved but given our performance constraints and given that Unity is a large and old code base, this will take time to do properly.

So to avoid postponing the public release with 3-9 months, we aim to release an initial version which does not have great support for large worlds. Then, in following release we will improve on this. Rome wasn’t built in a day :).

[rasmusn]

@Onigiri

I love this infinite loop of switching between trying to make dynamic GI > semidynamic GI > dynamic GI again without delivering anything for ~10 years. At this point just buy H-Trace WSGI and make it official solution. This asset is better than any realtime GI you could ever create (it’s even better than another surface cache slop - Lumen).

I agree with you that our history has not been spectacular when it comes to realtime GI. There are many reasons for that which I won’t go into here. That said, let me clarify a few misconceptions in your message.

• “without delivering anything for ~10 years”: It is not true that Unity did not deliver anything in this area. HDRP has SSGI and SSR both of which has a ray tracing mode. It also has a full path tracing mode.
• “another surface cache slop - Lumen”: We do not aim to replicate Lumen. We want something that is more lightweight, less prone to artifacts, and runs on a broader range of hardware, even if this means reduced visual fidelity (compared to for example Lumen). This prioritization leads to a completely different algorithmic design.
• “trying to make dynamic GI > semidynamic GI > dynamic GI”: To my knowledge we never intended to ship a fully Dynamic GI solution before. It is true that we did an experiment with DDGI but to my knowledge this was never intended to support dynamic geometry.

H-Trace is great indeed but its approach (ReSTIR GI backed by voxel-based irradiance cache) is inherently too slow for our needs and it won’t run a broad range of platforms (this is not the author’s fault, it just an inherent consequence of its algorithmic design). It is possible that we want to build something like this in the future (and coincidentally I personally have built something like this as a spare time project ).

[Wolfos]

Thanks for the reply. If it’s usable by 2028 that’s soon enough for me. Just happy to hear it’s on the roadmap (provided there’s actual follow through this time around).

Can you clarify how it wouldn’t match the visual fidelity of, say, Lumen? It sounds like even on a high-end desktop GPU we can’t scale to that quality, so I’d assume it’s not an issue of probe density or bounce count.

Focusing on temporal stability is good though.