Of course, it is possible to have a sequence of baked vertex colors, by ICE. Not vertex color directly, instead you create a custom 3d vector attribute, per point, bake and load this 3d vector, convert it to vertex color at the end of ICE tree. Something like: get node location > get 3d vector attribute > convert to scalars > convert to RGBA > set vertex color. Once you have this 3d vector attribute, loaded through load cache node ( not through mixer), it's possible to smooth with the neighbors, blend across frames, apply it partially, so on.
It's also possible to create entire ICE based lighting setup, affecting the vertex color.

To prevent ICE to kill the attribute during cache process, my own recipe is something like: get the attribute in first row, do something with attribute in second row, set the attribute in last row.
All the time, you want a float vertex color, not 2 bytes.

1 Mb of ICE cache, having only this attribute, contains about 100k vertices. For scene like this, that would be about 30 mb per frame, for all.

How hard it is to manage all that, well, just like any other ICE caching. I'd say it's 90% about skill of user. If I would be a measure, let's say I've been confident with this after two months of ICE-ing.

Still, I don't explaining and defending a workflow theory here, I just saying what I've used.

Afaik, method belongs to 'old fashion', I'd believe the "proffered" way is point-based GI.

Looks like a new version just became available for the public.

http://www.3delight.com/en/modules/PunBB/viewtopic.php?id=3514

Looks like a number of nice fixes and upadates.

One noticeable, which is in commercial version for a while, would be a stronger support for deep shadow maps. If volumic shadow map is enabled, slider for sub-samples has effect. Before was only from up to down mip - mapping, 3dl optimized shadow map of high resolution, Now it's in both direction, generally some smaller initial resolution + sub- sampling should be enough.
Easier to get detailed shadows for tiny objects, leaves, grass, hair or like.

Hi

just to warm up version 3.1.15 a bit. Boots are here instead of people I've planning to put into scene, one day It's closed scene, rendered with a help of shader modifications, mentioned above in the thread. So it's *not* OOTB 3delight for SI, not especially related to last version. It took about 30 minutes on QuadCore with 4core 3delight for SI, using ray-traced GI and reflection. Probably it will take a twice to get smooth sampling, or I'll switch it to point-based GI, if ray-trace become boring.

Hi Mathaeus,

The scene is looking good! I specially like the fine detail of the fabrics bump (or disp?) on the seat, it does read well although its pretty small..
metal is also looking nice, although you may want to add some glossiness variation to some spots to make a bit less *clean*

only thing bugging me a bit is the boots modeling, shape is not bold enough I think, might be the top ending looking without some sort of cap all around,
also a light shell to the interior might help to gain some weight.

cheers.

Yes this is displacement on fabric, variable reflection's glossy spread as well. Once I'll put everything I want into this scene, if I ever, probably I'll see what to tweak.
BTW if you're coming from Max, maybe a good start with Renderman compliant renderer, is to consider it as an highly boosted Max default scanline renderer - means, shadow map rules... of course plus a bunch of additional features.
Regarding light falloff question from your another post, in 3delight it's the standard SI solution, by enabling falloff in light's PPG, you getting a default exponent of 2, usually considered as 'realistic' one.

Hi,

a few tests of lightweight particles. When particle shape is set to "point", particles are rendered as circles using a specialized algorithm which is suitable for such small primitives. This allows to use much higher count of particles, but drawback is default billboard orientation. Circles are always aligned to rendered camera.

Here's small trick for more interesting shading - desired shading normal is converted to color. In render tree, color is converted to vectors for bump map. Particles are still aligned to rendering camera, but shading normal is modified. ICE tree and Render Tree are something like this. For some reason it seems 3Delight wants a negative of shading normal, here.



And how it looks like. I've used ICE setup which tries to do even distribution on mesh, following polygons. Still it needs work.

Looking good, how much faster is it compared to your Knit Strands Pro plugin with these woven patterns?

That SSS shader looks sooo nice. What's the rendertimes and resolution?

yeah I forgot rendering times. First woven is longest - without subsurface on skin, it's about 1 minute, with subsurface, it's about two. Using 4 core 3Delight on QuadCore machine, 1280*720 resolution, 8 light with deep shadow maps, shading rate 0.5. In image, there's really sharp shading rate of 0.1, so this is about twice longer.

Anyway, I wanted effect of first one, ability to add dust or small specular dots, then to add this in post over existing image, somehow. Especially these specular dots are hard to do with displacement, as displacement needs to be "continuous". Another advantage is resistance to shading rate, each particle have own "contour". Woven is some kind of occasional product, maybe it could be faster than mesh, with detailed pattern. But once the particle is moved out of emitter's surface, or it's rendered alone, there's a good number of issues - it does work with ray-tracing, but circles seems to be still aligned to camera, supplying an unwanted target for the ray. It works with point-based GI too, but it's not so fast, nicely to say.
Would be interesting to try this for clouds, or short high-dense fluffy fur, like on teddy bears.

Subsurface is more along line of misss_fast stuff, preset is default, almost monochrome "marble". Subsurface gradient belongs to two Car Paint shaders, adapted to work with 3delight subsurface in similar way to 3dl misss_fast shader (two times stronger subsurface from back side, scattering from back side too, or "Nn" instead of "Nf"). Car Paint is there because of diffuse falloff, to get difference between layers. Base layer is original skin color, using diffuse falloff of 1. Top layer is inverse hue of skin, some kind of blue-purple, using diffuse falloff of 2-6. Idea is to get a "whitening" on high light-surface angle, in additive mix.
Finally, mix is what I call screen with pedestal, which allows output higher than 1, gradually looking like plain additive.
In short, subsurface shader is more for blurring, it could be replaced with some simpler shading model. Skin color is almost uniform.

Here is with - without subsurface comparison.