[WIP][Asset] Dynamic Light Probes for Unity Indie, Pro, Mobile

Hello everyone. I am a member of the team developing games for mobile platforms (or at least trying to do it..). While working on lighting, I invented next wheel and named it Dynamic Light Probes. It may be used for the lighting of movable objects. Works with all platforms and licenses, in all rendering paths. Does not break dynamic batching. Just leave all lights in the scene, the system will do the rest. You can change light's parameters (position, range, intensity and color) in runtime.
Stress test on Nexus 5. 1156 animated lights (and dynamically batched barrels).

This system allows you to do dynamic lighting without increasing the number of drawcalls. And without a big performance hit. It bakes all light's data into a sort of "light volume", which is available to any object in any time. It can be a static geometry with precalculated beast-lightmaps, which can be illuminated by dynamic light sources further , or it can be dynamic objects.

Advantages:
+ Editable and movable "light sources". Theoretically unlimited amount
+ Dynamic objects, static objects.
+ Dynamic batching.
+ Different types of "light sources".
+ "Working Area" around the camera or other world point. You can control balance between quality and perfomance.
+ Should work with all versions of Unity (tested on PC and Android). In all modes of rendering.
+ No blending or alpha test (mobile friendly)

Disadvantages:
- Problems with the height difference (although changing lighting supported in some way). Requires sophisticated algorithm. Working on it.
- Lights not respect obstacles. Requires sophisticated algorithm.
- With a small number of light sources, the benefits are not obvious (special shaders for objects used, not "free" for GPU).
- Large areas require big amounts of data (Can be adapted by changing light density / quality).

I'm wondering what you think about this. Will it be useful for you and how much you think it's worth?

 

This is clever, (even though not many people can find a use for it)
But it would be cool to use those lights on a ufo . Which I think I will use it for eventually. And I think 10$ seems like a fair price

 

Do you mean there are no people who want to use tons of lighting sources at one scene (lanterns, flashlights, fireplaces, alternating with the time of day, etc.)? I think its not true. But with "real" lighting you drown in the sea of drawcalls (actually you have to tell it "Goodbye" on mobile platforms, if you are targeting on the wide range of devices).

 

Maybe you can say more about what this does and how it can be used? And what its limitations are? I'm interested in dynamic lighting and solutions that will work for higher level mobile.

 

Here is another example. At that screenshot you can see the scene with free assets from Asset Store. It is illuminated by 17 light sources. As you can see, all barrels batched in 1 drawcall, as they have the same material (men does not batch because of skinmesh). You can move lights and change their parameters, the number of dc and performance is not affected.
p.s. #1 message updated.

 

This could be really useful, anything that saves drawcalls but still have dynamic light is important to me.

 

Working on optimization. Stress test video added (check #1). 1156 animated point lights (and barrels).

 

Looking very impressive, any chance of a webplayer demo ?

 

any updates on this ?

 

- Spot light type added

- Still working on correct height algorithm

- Optimizations

 

Looking really good, the spot lights great. Will this support any form of bump mapping?

 

Well, as I wrote earlier, I'm interested in mobile gamedev and complex per-pixel lighting is not very good choise for it. Especially considering that in DLP getting a light direction is not so easy. Nevertheless, it is possible, by use of a few tricks. Although for the tangent-space normalmaps, you'll need gles3 support, but for world-space it's much easier, and gles2 is enough. As you can see, this trick also allows to light objects almost correctly, based on vertex normals.

Spoiler: Image

 

Can lights be created at runtime? What are you actually doing on the assets, vertex color?

 

Any time scale for a release ?

 

No, you can't do it for now, for the performance reasons. But you can turn it on / off (and use something like ObjectPool).

Actually, I do not touch it. Perhaps in time, i'll try to keep some information in the attributes of vertices for optimization.

I work on demo scenes that will help to represent system abilities in details.

Spoiler: Image

 

How are you doing this? IT LOOKS AWESOME!!!

Also, a free version would be great. Means I could spruce up one of my mobile projects for free. all I really need though is a good (and Unity-free compatible for $0) light probe system.

 

This cant come soon enough.

 

@dirtybassett: please state what you mean! I have no idea what you're talking about!

 

I ment that a system like this will greatly improve the look of my environments. Can only get so far with static lighting.

 

@dirtybassett: Agreed!

 

Outside of mobile I would be interested in seeing what this could do to allow for a pseudo global illumination effect when combined with directional light, much like CryEngine. That could be extremely effective if you took it in that direction or if that's already viable. So from what you have said this will require custom shaders? Any chance it can work with existing shaders or shader packs?

 

@Tristan Moore: From seeing how this is done, what you might do with such a system would be to add one more layer of functionality, and perhaps do some sort of final gather calcuation that generates a set of 3D points with lighting information. This way, that info can be used to generate a set of indirect lighting probes.

@BelfegnarInc: Would such a baked GI method like the above ^^ be possible with this system?

 

Guys, I apologize for delay. Some problems with AssetStore publisher registration (on my side, if you know what I mean ). At the same time, I continue my work on the system. Making it more user friendly (closer to designers than programmers).

By "custom shaders" I mean that we need to provide information about light to the shader, as it is done in vertex lighting, pixel lighting or lighting from native lightprobe.
Out of the box DLP will include multiple shaders with different lighting quality / performance. Basically it is simple diffuse, but there are a few special shaders such as this, which allows you to fake some interesting lighting:

Spoiler: Image

It's "hemisphere lighting": DLP + sky light + ground light and / or directional sun light.
If you need an existing shader to support DLP lighting, it is no more difficult than adding native lightprobe lighting to your (custom) shader. Is it a problem? I think this point needs to be discussed, I want to know how close designers are to shader code.

In fact, I really have been working on a kind of extension (and you guys read my thoughts). It looks like something between FarCry 3 and Assasins Creed 4 GI. You bake indirect lighting information offline for every time of day (however I use native lightprobes for baking it to my probes, so it may be a problem for indie users and I need to rethink it). Then at runtime, you can dynamically blend that info based on sun direction. I would also like to add a dynamic gi relighting from dynamic sources, but do not know what platform will handle it.

I do not want to make loud statements (GI is a sore subject on this forum, and many people have been disappointed more than once already ). I think, I'll write some post on my strange blog, if I get some serious results, and give here a link.

Spoiler: Here are a couple of screenshots

 

Hmm... looks like GI's coming along nicely!
All you really need to do is fix the weird green reflection coming off the blue cube. (Technically, the light should be blue, but being WIP, I won't push it! )

And, unless one has asked already, and in that case, just quote the post for me, will this be free? or will it be a nice low price?

 

In fact, it's a blue reflection, but floor's shader used it incorrectly
The sad part: all visual issues depend on lightprobe's positions (this is typical for gi in above mentioned games and for any lightprobe system, I think).

What do you mean by "nice low price"? I planned to talk about it in this thread actually

 

@BelfegnarInc: By "nice low price," i mean't not ridiculously expensive. I would prefer something like $2, but $10 is okay as well. And of course, having it free would be asking for too much!

And your GI system... how come it's doing it incorrectly? Isn't there some sort of way to grab the colour form the nearest probe, and use it to colour the GI?

 

Release date?

 

Is there any updates on the asset. I am actually eager to test this out as we have many overheads in the light issue. I am sure this would be very helpful.

Please let me know when are you planing to release it and what would it cost.

 

Any news ?

 

Maybe I should PM the guy...

But didn't he say this light system was going to be featured in a game too? if so, maybe his absence means awesomeness may erupt soon...

 

i will buy it !! if you released about the price i can t tell ...but i will buy it

 

As far as I know, this system was either going to be a free asset, or even just a couple of dollars: From what I remember either way, it wasn't much at all.

 

Are you still working on this or is it dead?

 

Mind if I PM this guy?

 

It's definitely not dead and i am still working on it.
Here is an example of full 3d light volume (only light color info, without direction and textures).

Spoiler: Image

Specifically what I do now:
1) Optimization
2) Сorrect light direction recovery in object shaders for proper illumination of faces and all bump stuff.
4) Geometry occlusion for light propagation. So the light won't pass through walls, and generally more correctly distributed.
3) Shaders combinations. I think to enable lighting as per object, and as an post-effects/lightpass (in some way it will give a constant predictable performance but will reduce the possibility of combining shaders with different performance). Also I think about the ability of switching some computations between CPU and GPU, depending on what is most loaded. Simply put, the system will look like this: the more opportunities you have (hardware, platform and Unity license), the more choices of algorithms and methods for calculations it will be, i.e. the higher performance and quality. So it's not a one universal algorithm with permanent quality and performance. This is the most slow-developing part.
4) Something like a cascade approach to the volume of light. It helps to distribute the load of calculations depending on distance and improve light quality near the camera. In short, it is LOD. At the moment, I am not satisfied with light quality in FPS view.

Now, about GI.

Spoiler: I got some consistent results with indirect lighting.

It has some limitations. Firstly, the light passes through the object with thickness less than distance between two probes. Second, it's 2.5d-kind, there is only one layer of probes along the Y axis and indirect light just goes into a neutral color with height (but every probe got info about light from 4 directions). This is closer to Assassin's Creed 4, so think of it as the outdoor ambient light for such geometry as city with closed houses or other open world. For indoor you will need to use something else, or greatly increase the density of probes, although the problem with Y axis remain. Maybe someday I will expand it to PRTV (Farcry 3) doped with LPV, but I don't have enough free time for now.

 

Here is a screenshot of a paper about AC4 for you to understand what I mean by saying "closer to Assassin's Creed"

Spoiler: Image

 

So any release date on this?

Looking good, BTW!

 

Could you not release the light probes as in your first video's, then add GI later?

 

any news on this?

 

any news on this?

 

is this release? Or something news?

 

Quite frankly, this guy seems to have either vanished, or discontinued this asset, sadly... I can sort of see why though, judging by the impact unity 5 made on one of my free assets...

If he's just AWOL at the moment though... give him time, it's probably for a good reason. He might even come back with mindblowing changes, who knows?

 

A lot of guys / projects disappear, usually because the dev takes on work with a company. I suspect that's what happened here.

 

Hi guys. I want to tell you that I'm still alive and continue to work on this asset in my spare time. I spent last n-months on optimizations, GPU-version of the system, tests on different platforms such as web and webgl (quite good results, I must say =)), experimenting with the propagation of light with respect to obstacles (low-res). Almost stoped all works on indirect part (so that, most likely, it won't be included in the first asset release version, or it will be released separately). I want to note that Unity 5 update brought me more optimizations than problems for the system. Nevertheless, there is some unresolved points. For example, do I need to add support for PBR gles2 Standart shader (or leave it only with Legacy shaders support). The fact, PBR gles2 shader have almost no free attributes, interpolators and instructions. Thus, to add new features, one will probably have to remove the old part of the shader. Any thoughts on this?))
Furthermore, since this system is a lighting approximation, benefits for pbr from it, is not obvious, I think. Although.. Optimizations, right?)
Summing up, I want to inform you that in a month or two it will become clear what is it: a ready-to-use solution in Assetstore (I'm not sure that it will be approved, given the recent sad forum topics by noname-guys =)) or just my next closed experiment).
Lastly, a few screenshots from my latest experiments (protect your eyes):

 

Approved! I simply like how you've got normal maps working well together.

 

I would buy this in a heartbeat. Looks amazing!

 

I agree here sir.

 

Is this still being worked on? I would buy this without hesitation, I need lighting for my mobile gaming and regular lighting causes way too many drawcalls.

 

will the author come out to say the progress of the plugin? it's useful on mobile