Just recently started seriously testing 3Delight and I have to say I'm really impressed and it's so easy! So much so that I may not use MR much from now on, this is great! I don't have anything really worth posting yet but just wanted to say if you haven't checked it out yet, it's worth it. The point based GI alone is worth a look but there's so much more. Big thumbs up to the guys at 3Delight and so glad we have it for SI

Hi,

a few more of indirect lighting, this time without photons. For two lighter pics, in baking pass I've used some mix of classic lights and spherical "light objects" (with exaggerated incandescence in material). In final render, no any classic lights, they were substituted by "light objects". Blue pic is just a zero bounce, all lighting comes from baked incandescence. Render times are about 5-10 minutes on QuadCore, using 4core 3delight.

A few observations:

Global illumination shading rate, this is used twice, as baking shading rate, also as "irradiance shading rate" in final render. So seems it's good idea to split the rendering into baking pass and final pass. For baking pass, I've used separate baking and rendering camera. Rendering camera in baking pass is pointed to "nowhere", just to avoid unnecessary render time.
To speed-up the process, I've used high-detail shading rate for baking, low-detail for final render. About 1 for baking, about 2-16 for final render. High-detail in baking, this seem to help to avoid artifacts in final render. I guess there's smaller chance for "missing" the small cylinders in point cloud. However, displacement seems to be "blurred" by irradiance shading rate (in first pic, I've painted displacement's density, irradiance shading rate is 2).

Max Solid Angle was 0.05, PTC bias 0.05.

A few about sequences... Mentioned settings seems to work fine for camera animation. For moved/deformed objects, I think a lower Max Solid Angle is a way to go, to avoid noise - at least in complex scenes. Simple scenes, just one character or so, works fine. Also I'd believe some problems I've noticed, were caused by default, view-dependent dicing in final render.

Anyway, all that PTC stuff seems to be enough fast and predictable for pleasant playing



1280x720 pic

big thanks to your tests. your last ones looks really good.
i wish i had some time at the moment to play more with 3delight.

Mathaeus,

nice, and thanks for posting your details. Im still a little confused with the baking and rendering side and your posts are helping. I'm currently trying out the demo of 3Delight and impressed with the speed of it all (and the amazing Softimage integration of the shaders!) even though 2 cores are used. Ive been playing around with the SSS parameters but still rendering tests, will try and post something when its finished rendering

Essentially I want to find out about brick maps. and find a way to make / bake a low res lighting SSS geometry spheres into a high res fractally displaced geometry. a cloudy kind of thing Have tried using SSS on the displaced geometry and it is too prohibitave, so am exploring the brick map / ptc route.

anyways , keep the tests coming

cheers

afaik, brick-maps aren't featured in 3dfs plugin. However, subsurface scattering is baked into point cloud (at least with ptc based color-bleeding). So, if you're using "hitsides" "both", it seems there is a really dirty trick: use displacement and subsurface in baking pass. In final render, disable subsurface and apply a slightly negative Push SI operator, use only radiance in some standard SI shader (that is, diffuse to black, disable 'filter radiance with diffuse'). Now surface takes a 'self' radiance.
Of course, this is a really dirty trick, I have no idea what problems this could do on more than my simple test....

Btw, if you didn't already, you can install a 'big' 3delight and preview the ptc files with ptcview - ptcview works with free license too.

heres that render test I promised, was rendered over a few days with the free 2 core vorsion. really like the 3delight stability and consistency in SSS. no flickering or crashing ! definately going to try some things out with rendering pointclouds. Anto, I really like the idea of 'baking' shadows and sss data in before rendering but unfortunately in this instance it would not make much difference as the particle objects are changing in size & position every frame.

Really nice to see how 3dl subsurface "melts" all that overlapping volumes. Something that only renderer can do ...

A few in Christmas style , using 3dfs 3.1.0


First two are variances of point-based color bleeding. First is one standard light, pointed from back to camera, together with color bleeding. Second is solely point-based IBL. Render for each image took about 25 minutes on Quad Core machine, using 4core 3delight, for 50K hairs and the rest of geometry. I did only basic optimization, like very high GI shading rate for hairs, about 64. Point-based bias is about 0.02. Baking is performed in separate pass, with shading rate 2 for all. Didn't tried additional trickery for speed-up, like using GI falloff, low-res hair geometry for hairs, higher point-based bias.

Generally all this was possible before, but this time everything is done with out-of-the-box stuff. No hacking of sl files.

Third is just a pair of classic lighs on 100k hairs, together with a new, polygonal aperture DOF. This one took about 4 minutes to render.

the third looks the best, alot because of the direct lights with shadows i think. spec is also great which i miss in the ibl rendering alot. diffuse illum of the ibl is best i think, like the gi look of it. scattering effect is low, it needs some additionel lights here. rendertime is good for all images for 4 threads only.

Hi,

this one is about 3delight "hit mode" property, when is set to ‘Use SurfaceColor3Delight Vertex Property’. In this "hit mode", if object has a CAV property with this name, ray-traced GI takes a vertex color, instead of shader evaluation. Name should be exactly SurfaceColor3Delight, it's case sensitive. Ray-traced GI renders at speed similar to ray-traced AO, still with ability to catch the color of another object. To get the desired render time, all involved objects should carry the mentioned property.
This is how it looks like with "light objects". Light object has a bright vertex color, the rest is plain black. Render time on Quad Core, 4 core 3delight, is about 3-9 minutes - this one took about 9 min. I've set GI shading rate enough sharp to catch the displacement.Something about GI shading rate 0.5, 256 samples.



Now a bit of hacking. It seems, 'hit mode' applies fine to "gather" function, used for glossy reflections. So, by adding a few lines to mia_material.sl , like:



we will get something like this pic, using glossy reflection from MIA material. Now, with even sharper shading rate, about 0.15, render time is about 4 min.
Of course this is just one-level reflection, for more than this, "next ray" should be baked into CAV, somehow. Probably with some fancy ICE solution. Anyway, I think the render time makes it really interesting.