Hello everyone. I just have a simple question. Does cycles support caustics? And if it does, how realistic is it? Like, can I shine a lamp through a triangular prism and get a rainbow out of it?
AmedeoCompositions have a âappetizerâ tutorial which shows how to fake caustics using nodes and masks, (you can produce the same effect with a suitable image map). Itâs not precise, but it certainly enhances the render.
Have attached images showing the difference it makes, without & with.
Article how to get caustic faster
And here is a faked caustic, looks quite good and its much faster (awesome specially if you need an animation 
).
https://dl.dropboxusercontent.com/u/55113627/>Blender/Caustic%20(Fake)/50%.mov
Blend File (its from january 2013 so many changes in cycles, maybe you will need to rebuild a node tree)
https://dl.dropboxusercontent.com/u/55113627/>Blender/Caustic%20(Fake)/test-Fake-Caustic.blend
Inside are two versions (two authors) of node group, that I found in that time:
Ok thanks guys. So cycles isnât great on caustics. Nor does it support bidirectional pathtracing. Alrighty then. I might as well go learn Luxrender.
One more thing. Do the developers of cycles plan to improve the caustics? Because I would prefer to texture objects with cycles and get results similar to Luxrender.
The official word is that itâs not a high priority because caustics are so rarely worth the trouble in production use, especially in animation where Cycles is mainly targeted. Storm_st has been working on a patch for awhile that would improve it, but I donât recall that status of it.
Oh okay. Interesting. Well, thank you for the information. I appreciate it.
Looks like i am only guy who play with that idea . The problem is i have no access to real reliable GPU hardware to tweak/optimise bidirectional patch (NVidia based too closed/proprietary, AMD/ATI just not work on valid OpenCL code, and improvements not fast as i want). So patch in âearly developer only prototypeâ mode, not useful for anything.
I think that this caustic mode is not that needed, as current Cycles âPath Tracerâ cover 99% of typical needs. Maybe replacing Sobol + Creanley-Patterson by Niederreiter-Xing + Owen scrambling will boost Cycles on complex indoor scenes a bit (I suspect it is main reason, despite usual optimisation, that ppl say that Arnold and Maxwell much faster even on CPU). I still have no real code to check it looking at pictures side by side, have trouble to understand that Owen trick, but all that papers with theory say on dimensions ~(4âŚ32) it will be significant, up to order of magnitude improvement. There will be some overhead, but i doubt it big especially on multi bounces indoors.
For now better use Lux or Yafaray (or all that commercial Maxwell, Octane). Or get your hands dirty and optimize/rewrite from scratch my patch :).
1 Like