SonicEther's GI is looking good. Will this finally be 'the GI'? (Give this man 3D mipmaps!)

Through the ages, a thousand warriors have claimed to have figured out 'the GI'.
All of those warriors faded, but there always came one to take their place.
And so, today, will this finally be the legendary GI?


^That's Unity. Realtime. No baking.

And it could look and perform better. Just give this man 3D mip maps:
http://forum.unity3d.com/threads/fu...lobal-illumination.314748/page-9#post-2523668

This is the most beautiful I have seen the Sponza scene as.

EDIT: Thread http://forum.unity3d.com/threads/fu...lobal-illumination.314748/page-9#post-2523668

 

In a word: No.

SE's lighting is incredible, sure, but it's also been created with a specific purpose in mind. Namely, it only works with directional lights. Unity's GI is a more general purpose solution that makes sense to be "the" GI, while specialty GIs can, should, and will continue to be available to users in the Asset Store.

 

I think since SE is using voxel cone tracing in the current approach to his GI solution, the directional light limitation is no longer an issue. His initial approach mentioned in the first post of his thread was a different method that only worked for directional lights.

 

I don't like it.
I don't like how the light is positioned or that I don't see area with brightly lit columns on the 2nd floor.

You'll need a demo where you can fly around AND you'll light that can move/change color. Few point lights with shadows would be nice too. Or some emissive lights.

Current screen looks like it is carefully trying to hide some limitation of the technique.

 

@neginfinity, you can download the demo from a link in that thread and see for yourself. I don't think anyone would argue the technique is perfect; in fact, SE is pretty open about all the limitations and artifacts he knows about and how he can or cannot deal with them at this time.

 

Fair enough.

Checked out the demo. Impressions/thoughts/etc.:

* I didn't go "wow" from it.
* I see light bleeding through objects in some places.
* As far as I can tell, he's caching results from previous frame in order to simulate bounces. If you disable infinite bounces, light will lag behind sunlight movement by half a second. If you enable infinite bounces, light will remain in place where glowing object used to be for about half a second. That would make it hard to create some sort of flashing effects like a gunshot flash.
* Fps doesn't drop with increased number of glowing objects.
* I have 20 fps on max settings.
* I would like to see this being used on significantly larger scene.
* I would like to see it being used on large INTERIOR scene, like a courtyard connected to many corridors where only courtyard receives sun. Or just in dungeon with a single candle.
* I would like to see how that works on highly modular scene.
* I would like to see how that interacts with animated emissive materials (enlighten can handle that).
* I would like to see how that works on highly-modular scene build from a kit where each piece of the kit is small (2.5 meters cube, for example).
* I'd like to see how that interacts with dynamic actors.
* ...Or with destructible geometry or with levels where doors can open, curtains can raise, etc.
* ... or with stained glass/semi-transparent surfaces
* ... or with situations where you need to have sharp shadows from some pointlight.

So, in the end it is not "The" GI solution. If it handles modular scenes AND actors AND scales well, it might be useful as enlighten replacement for certain situations.

 

I recommend you post your feedback in SE's thread. Better chance of him seeing it that way.

 

Also, I don't think fully dynamic and realtime GI will ever be "the" solution. High quality + high adaptability is going to be more expensive than a lighting solution that sacrifices one or the other. In time, that gap will probably narrow to a point where the difference is negligible because of advances in hardware capabilities, but we're obviously not there yet.

That all being said, I'm excited by the possibilities of this implementation and how it might develop over time. Even with its current limitations and use case constraints, it sure is purdy.

 

You can post a link to that in the original thread, if you feel like it.
Otherwise both threads will be mixed up and we'll provoke wrath of the hippo.

----

I can say that I'd be mildly interested in a solution that doesn't add random tint to modular objects AND can handle scenes of decent size (think scyscraper with interiors). Not sure if that one fits those kind of requirements.

 

The original thread is here and has quite a few pages. I would stick toward the end though, since a lot of what is discussed early on in that thread was scrapped/re-worked. There is, somewhere in there, I'll make a link if I can find it, another interactive demo with kind of a "Grow Home" looking environment. You're able to see the exterior and interior effects and even throw point light-esque objects (Icosahedrons) that dynamically light the world in random colors.

He's talked quite a bit about shadows and it's clear that is still a big limitation. But the limitations have been steadily narrowing down as he's working on this. He does an amazing job of working around issues and coming up with new ways to handle things.

I mean, it's still way too early to tell one way or another. There's the possibility he will run into an issue that simply cannot be solved for, either due to limitations of Unity or just where we are right now in hardware. If he's able to work all of these things out (fingers crossed/knock on wood) then I could see it being something used as an alternative to Unity's current lighting system.

I for one would just like to see us move more toward dynamic solutions rather than baked ones. Sure they're wonderful for mobile devices, but for PC at least, we can do so much better.

The demo I was talking about: SEGI Alpha Preview Demo

 

Demos or not, it wouldn't wonder me, if this ends like previously showed GI solutions. As long as he doesn't put this up on the Asset Store, I'll still have my doubts.

 

I'm not sure about that, though. I'd be more interested in solution that can be used quickly without overly long baking times AND that doesn't produce seams and tints on modular environments.

Both UE4 and Unity sorta insists on modular mesh-based level construction. However, both have issues with modules. In case of unity issues are significantly worse, though. Lightmass produces less-noticeable seams, and in forward rendering mode (in unity) it is very easy to accidentally make modular pieces "light up" when dynamic light passes by.

So something that can address THAT would be more interesting.

Fully dynamic solution is a cool technical issue, BUT not every game needs that.

 

I believe UMA was funded by Unity at one point? I wonder if they could do the same with one of the gi solution. Pay the creator, then release it under MIT/BSD license onto github. THAT would be great.

 

That's totally true, and I do believe there is room for both. But both need to be addressed. I'm sure I'm a bit more biased toward dynamic lighting as we are making heavy use of that in our game. Lights can go out, or shift, or animate and while the deferred rendering system is wonderful, it does feel (again most likely due to my own bias) that the lighting system is more focused on creating lightmaps and "pre-computed" systems.

I know you mentioned in another thread that you were investigating or had personal experience with lighting issues when working with procedural generated geometry. Are you currently using lightmaps or pre-computed "real-time" GI? If you are using the former, I might like to discuss with you your methods on that as I might be able to apply it to my own situation.

Either way though, being able to eliminate those gritty details (seams, bleeding, etc.) would be a higher priority.

 

If they can sort out the problems with VXGI's performance, that would sort out the need for realtime GI.. It looks beautiful!.

Although SE's implementation does look good, IMO better than Enlighten from some of the scenes I've seen.. So yeah give him the access he needs to finish it off.. Won't say no if it works and he plans to update / improve it for larger games etc.

 

The issues with GI were described here. It was not "procedural", but modular geometry, but you'd hit the same issue with procedurals too.

Spoiler

Also, I'm not currently working on that project, but I was royally pissed when I found out that enlighten is unable to support modular design after I wasted time making and adjusting that mini-building kit.

As far as I can tell, the only ways to get rid of those problems are to completely describe ALL gi AND reflective cubemap too. Then you'll get nice flat surfaces everywhere without tints. You could also try using bigger objects, BUT if you're in forward rendering mode, you'll quickly hit dynamic light limit per object which will cause objects to light up revealing their modular structure when dynamic lights approach or change intensity/radius.

Also, due to insanely slow baking times I marked enligthen as being unusable for lightmapping. Which means next time I'll be making something with unity, I'll be either making a 2d project or something with non-realistic lighting without any prebaked component. Cell-shaded or low-poly. Or both.

 

I also dropped Enlighten baking, for Unity mobile i import lightmaps from Blender, if i had a bigger game and need GI ligghtmaps it would be UE4 with help of Lightmass and Ray Traced Distance Shadow Fields.

SonicEther plugin does not fit your game project ,but it will fit other people projects that don't need hyper realistic GI, but a good approximation , real time and no baking , fast enought to run well on small old 3D cards.

 

He...

He...

He...

...

HE DID IT.

http://forum.unity3d.com/threads/segi-fully-dynamic-global-illumination.410310/

 

Must say the demos are beautiful