Render differences between luminous sphere and light

Started by paulcurtis, December 20, 2021, 07:28:30 AM

Previous topic - Next topic

paulcurtis

So i've been battling this for a while and produced an image to show

Upshot is that the rendering of a luminous sphere is vastly different (and incorrect) when compared to a light source.

In the montage image, i first noticed something off in a spherical - the sphere isn't illuminating anything near it but a distant cloud only. The i did a normal rectilinear render, and at 5000 lum it brightly illuminates one cloud, at double that suddenly the mountain is lit and i'd argue that same cloud is lit *less*

Then pop a light at the same position and i get what i expect to see - at 5000 and 50000 units. 

Scene and render set up is identical. It's the recommended settings - so deferred shading.

Odd huh?

It just means that i will use light sources for what i need so it's no showstopper, It was a method to see whether i could use luminous objects to light the scene.

tgen01.jpg

cheers
Paul

Hannes

Have you tried rendering with the path tracer? As far as I can tell, lighting with self illuminating objects work better with it.

WAS

Additionally, luminosity is approximated. Its not very accurate. Its good to use high cache values make a cache file so the results are the same each render.

Hannes

#3
I did some tests as well, because I was curious.
It seems small (!) selfilluminating objects don't have any lighting effect on clouds, neither V2 nor V3. Neither raytracer nor path tracer.
However on surfaces you can get some result, but actually it doesn't look very good. If you use the path tracer the illumination of the environment by the selfilluminating sphere is incredibly noisy. And with the legacy renderer you'll have to use incredible high GI settings as Jordan already said, and it still looks bad and takes ages to finish the prepass. So, in your case using a lightsource would be the best way, I'd say.

But you can use huge selfilluminating objects (a large plane for example) to illuminate a scene. Or even the background shader. This also affects the clouds.

I hope, what I said is correct so far, but as I said, I did a few tests...

pokoy

Did you try increase the voxel count in the cloud 'Optimistation' settings?
In most renderers, lights are more optimized compared to self-illumination - self illumination is calculated through GI samples only which makes it render longer than lights and is prone to 'missed' rays when caching of any sort is involved or has a too low resolution. With v2 clouds, this is the GI cache (I suppose), with v3 clouds it'll be the voxel grid.
I think increasing the resolution of the cache involved should lead to a more consistent result and picking up lighting from self-illuminated objects more accurately.

WAS

V3 seems to pick up no lighting even if voxels are high. Voxel resolution doesn't really apply much effect inside TG, and really just effects shadowing as far as visible effects.  Voxel resolution does help with V2 clouds though.
Interestingly, hiding a emission object within the main object seems to obscure all the lighting, even though the main object has no shadows enabled. Once I make the sphere 1.01 in size rather than 1 like main colour object, I get more lighting in the environment, like the clouds, but it's ofc just garbage results.

I've always said the illumination in TG is just simply terrible. Even rasterized game engines do this better, including voxel interaction. Blender too. Even without GI like Unity, they have had amazing area lights. The demonstration with the android coming to life in the haze environment is all done with self illumination.

WAS

When you look at the results of area lights on surfaces, it seems clear that GI is calculating the light "somewhere" with its space, but not literally accurate lum surfaces positions. Than it seems to try and calculate intensity of said light, than it tries to paint the environment in light which looks like huge low resolution  soft circles randomly put about in the areas in thinks should be lit. There is absolutely no shape data or ray data imo. Just approximating where light should be and doing some stranger effect to achieve lit surfaces. If they were rays we'd have speckled lit stuff and far more accurate results. Unless those huge soft circles are a single ray somehow? Don't think so. I feel this is faked more than other engines. These weird artefact surface lighting of huge blobs I've never seen in any other engine.

WAS

One thing you could try, I am not able to at the moment, is create a constant scalar of 1, and then a add scalar, and in the add scalars input do a Simple Shape Shader with it's edge colour disabled. Make it a circle with 100% edge. Now use it's colour and set it to maybe like 10. Feed this into the clouds Enviro Light Tab. Set the location of the Simple Shape Shader to the location of your illuminated object and adjust size for light falloff.