New approach to dynamic GI

Just saw this:

Maybe it’s a feasible technique Epic can integrate into UE4 ?

So, what do you folks think ?

I would imagine that a lot of open world games are using a similar method

Why wouldn’t Epic implement this in UE4 ?

Also found this (not GI, but might be useful for many real-time lights)

I don’t think you know what dynamic means.

No moving objects, no moving lights. completely precomputed.
From the Video I can say that Lightmass does a better job in precomputing.

If I have to place those “probes” by hand… then thanks, but no thanks.

Theire’s now some pre-backed Global Illumination in open word games that give an effect and “half-fake” a full dynamic GI (as enlighthen). I saw the one used in Ghost Recon Wildlands, and its impressiv (interior and exterior in the same time).

  • CryEngine use now good quality dynamic GI.
  • AnvilsNext 2 use a pre-backed GI technique usefull for verry large scale and small scale working in the same scene.
  • Unity use Enlighten that is full dynamic RT GI, verry impressive.

And the dynamic RT GI used in the UE4 is really far to give the result of all that engine…Epic is for sure developping it to be better. I don’t think they will let the competition steal theire place (i mean, they don’t occupy the same place that much but they have to be updated with technolgy and tools), kowing that futur video games will more and more use RT dynamic GI !

Enlighten is not fully dynamic, it’s precomputed and only works on static objects, all it does is allows you to move lights with GI.

Ho yes it work on statics objects only sorry for the mistake. But it work with movable spot light, directional and omni light.

Anyway, i just mean that lot of motor now start to propose powerfull RT GI technology (or pre-backed), and UE4 will for sure propose a technology like this in the close futur. It already porpose it in beta (but doesn’t work properly really… and is really not effective in different scales in unic level…but : it work with movable objects !)

This technique is only for stationary lights (including arbitrary emissive materials), and dynamic objects can only receive light not bounce/occlude it. Though they mention that in theory the probes could be updated in the same way as Geomerics Enlighten, slowly over time. All they did in their samples was turn all indirect lighting on and off. Using it for a path tracer is possible as they showed, and it runs faster than normal path tracers, but that is still not suitable for games.

This is basically a replacement for lightmaps, light probes, and reflection captures. It uses one system to do the job of all of those with a single lookup that can compute all indirect lighting (specular and diffuse) at any point in space from any direction with high quality, fairly cheaply. In particular it allows for very good geometrically correct specular reflections, accurate like screen space reflections but able to work with stuff that is off screen and even multiple reflections (but only static stuff). It also doesn’t care about surfaces, you can get as close to a static object as you want and the lighting will still be high resolution unlike a lightmap that has a fixed texture density per object.

Their referring to it as “real time global illuminatoin” is a bit misleading, its more like “realtime rendering of baked global illumination”.

Saw it, dismissed it. It’s not actually dynamic, nor does it make good use of memory, nor does the specular trace come anywhere close to useable in performance. So, interesting try, always good to see a paper based on what seemed like a good idea but did’t end up as practical so no one else tries the same thing. But not very useful.

I think there’s huge compromises in any lighting system made for use in real time applications. Either a bias for pre-baked, high quality global illumination or extremely coarse dynamic global illumination from either baked probes or something like Light Propagation Volumes or Voxel Based Global Illumination. Nothing in that video looked appealing from an aesthetic point of view and unless it’s designed for mobile rendering then I’m not sure I’d use that over light maps either. Even my LPV tests had results similar to that video (though perhaps using more memory etc etc)

Some of the best work I’ve seen on a shipping product recently was Quantum Break. There’s a great write up on that system somewhere.

Matt Hermans

looks fair

It´s a bit hard to say if the quality is sufficient. For an indoor architectural scene as shown in the example it ´s way to bad. For a game it might be enough.

The best realtime GI solution I have ever seen is still imperfect shadow mapping, wich is now nearly 10 years old. I still wonder it was never implemented in a big game engine like Unreal ? My technical knowledge is to limited to judge. So why does no one use it ? Is this just a nice demo that won´t work a real production environment or are these license issues ?