Warrior and her's friend

[amm]

Hello,
made this thing as an home exercise along line of "how many of different things, including a long blonde hair, is possible to put into two hours render on home i5 quad core with 16 gigs or memory".
So that's it, original is 2560 x 1280. It's rendered with Mantra, of course. Character, plasma guns and such, is an old Softimage work, here I've just imported the snapshots of animated character.
Houdini is all the rest. For more details, here is a couple of screenshots. Sedimentary rocks are from - scratch node work, a bunch of deformed boxes converted to VDB and VDB booleans. This is used to filter the array of high resolution grids via volume sample.  Then, chain of polysplits, polyextrude, cleaning the manifold geo, turbulent deformations at the end. A lot of nodes and around ten minutes to execute.
Hair is more ambitious work. Just for exercise, wanted to use simulation for styling, a bit more interesting styling than plain flat falling. So, these curls are product of additional geometry VOPs. Entire sim is blended, around roots, with 'static' procedurally based interpolation between two NURBs surfaces. Such relative small number of simulated guides wasn't enough to develop the desired look at roots. As this part worked better than I've expected, felt free to built the entire interpolation system with custom VOPs, instead of Fur SOP and Mantra Procedural. Basically, each guide has six auxiliary guides around, where Sweep SOP helped me to do not create my own Frenet frame - style interpolation. Later, instead of Fur SOP, there's plain Duplicate SOP, duplicates are just interpolated between originals, inside triangles formed by auxiliary guides and 'main' one. I think all that is several times faster than Fur SOP, as there is no any searching for guides, interpolation is performed from in-advance created attributes. Also, thing is 1:1 predictable - while, well, it is not generic anymore, as Fur SOP. Finally, converted the lines to NURBS curves to get them smooth, and converted NURBS curves back to lines, as Mantra seems to do not like NURBS curves that much.
Probably there are unnecessary steps here with hair, but, allowed me to render the hairs directly after the last SOP, no unpleasant hair rendering setup, no jumping to procedural shader and back. Also, had a continuous, plain 'forward' structure af attributes, enough 'friendly' for a lot of detailing trickery, later.

Anyway, I think I spent most of time with this really fascinating Mantra renderer, so further improvements are along different lighting, let's say some cloudy atmosphere.

Thank you for reading all this

 

[Sierra62]

That is beautiful work. 

[jrockstad]

Very cool piece! I love all the little details in the rocks with the cracks and layers, and the hair is beautiful as well. My only critical comment would be, and perhaps this is just an effect of the lighting, but it appears to me that the skin shader is looking a little flat - perhaps you left out subsurface scattering to get your render going a bit faster? If you go back to this piece that might be something to consider. In any case, nice work!

[amm]

Of course, there is subsurface shader, applied on entire object, including the parts covered with small fishnet structure. In Mantra terms, multiple scattering, based on point cloud. This 'point cloud' makes it relative fast to render, most expensive part here is hair, as usually. By the way, rendered the similar shape - lighting configuration before, using 3Delight REYES with many deep shadowed lights - looked similar. Me too, had impression 'so what, is this all to get from subsurface side?'. My small 'explanation theory' is based on somehow sad fact, that even in 2015, a long hair able to cover the face, actually is rare thing in CG. So artist have habits to build the entire lighting around what they have, like subsurface scattering with more exaggerated contrast in lighting - not in almost uniform, shadowed side like here, where probably there is no that much to expect from subsurface effect.

Anyway here is small comparison, subsurface (still deeper than it should), and plain diffuse.

 

Edited June 3, 2015 by amm

[amm]

This one has more of smoke in front. Took 3 hours for 2560 x 1280 on i5 quad core. This is probably last time for "render everything in one go" approach, really has no sense in case of smoke rendering. Anyway, still been able to find the compromise between relative high primary ray sampling, required for hair, and volume rendering behind. Turned the stochastic transparency to something really low, just 2 samples, got reasonable render time. With a bit of noise on 2560 resolution, anyway.

 

[Tyfx567]

Very nice!