Tweet
Ihavenofish on the Splutterfish forum:
My question is, for people who worked with Vrays displacement, is it that bad?[/i]
hi,
i'll probably get yelled at for feeding the trolls, but i would like to give my opinion about brazil's lack of those 2 particular features(displacement specificaly).
i have been evaluating displacement recently in every renderer available to me (basically everything ever made) for potential use in our pipeline at work(feature film vfx company). i still have a few renderers to go, but so far, a few things are very clear to me. i'll share some with you.
there are 2 things in particular i have been trying to achieve as reference for production usability.
the first is low poly (box primitive) displaced buildings. i was shown an image of displacement for a movie (day after tomorrow) using this method. for reference, it was rendered (i've been told) in prman. the image showed no flaws that i would find unacceptable. note, this type of displacement is "special" because there are many sharp corners and 0 degree draft angles. this puts the renderer to the test, unlike softer gradual displacements which most renderers handle easily.
using standard max displacement (scanline, brazil, vray all have, of course, the same result) i was able to achieve a decent result only with a very high initial tesselation. this you should agree rather defeats the purpose of using a low poly building with displacement vs a modeled one.
soo... on to the spiffy new(and old) sub pixed displacement type renderers. i'll just mention the 2 most people have access to, mental ray and vray.
- mental ray works, but yeilds extremely poor quality at low settings, and exteemly slow(compared to normal displacement) at higher ones. oddly, it also uses more ram!? but, in the end, it is infact capable of achieving the desired result.
- vray in 2d mode "seems" to work. its fast, and looks nice... in the parts that dont have nasty errors. unfortunately, no matter what settings i chose, and how i tweaked my maps, i could not produce a result that was not riddled with unacceptable errors. in 3d mode, i was introduced to a whole new set of glitches, and again, in the end could not create anything remotely usable. i wont get into the details of the errors(this is a brazil forum after all), but i will be reporting these as bugs shortly.
my second test is trying to use displacement maps created in Z-brush on a low poly mesh. many of you might have been playing with this software, and likely have similar (or maybe different) experiences. anyway, on to the results...
- standard max displacement (scanline/brazil) seems to project the displacement along the smoothed normal, from the face. this is not good for low poly meshes because your displacement takes the shape of the low poly mesh. the only solution is to meshsmooth. in standard displacement, this is acceptable, since you are going to end up with tons of polys anyway. however, this utterly defeats the purpose of using a low poly proxy and displacemnt(well, not utterly, you still have the advantage of fast viewport and skinning performance). to see what i mean stick any displacement map on a sphere with segments set really low(8 or 10)
- mental ray performs perfectly in this situation, displacing as if the low res mesh were actually smooth, however, like on the buildings, its is either crappy quality, or painfully slow, and for reasons unknown to me, does not save any ram. again, you can test this with the same 8 segment sphere
- vray behaves exactly like the scanline in terms of not respecting smoothing, forcing you to meshsmooth, negating much of the point. it also exhibits many of the errors it did on the building if the map is not sufficiently "blurry". so, for the purposes of low poly displaced characters, it offeres little to no benifit over brazil/scanline, and this only if you manage not to get any errors.
lighting:
here's a little insight into how (MR/prman, not sure on vray)displacement works when combined with ray tracing. in order to raytrace the displacement, like in brazil, an acceleration grid must be set up. in order to set up said grid, you need to pre calculate all the displacement. wait a minute, doesnt that negate all the benifits of sub pixel fancy schmancy displacement techniques?. well, yes. but theres a twist(in mr and vray anyway, have not got to prman yet), combining displacement and raytracing absolutely decimates render times(moreso in vray than MR). adding a simple area light to a displaced building increased the rendertime 10 fold over just displacement, or just an area light. this does not happen in normal max displacement.
alrighty then, i'm done spouting off. but now the question is...
how important is displacement as a feature if the displacement you have is either no better, no faster, no more memory efficient, and worst of all full of show stopping bugs.
in the end.. and in keeping with the fishies philosophy of not releasing things till they are release worthy, i personally would rather wait for something good, than have a feature just for the sake of having the feature even though i can never use it. displacement that works WILL be a "development".
now, having read this, you might also understand why i, and others here, are completely uninterested in the challenges posted comparing renderers. its not that we dont like chalenges, or are close minded, or hate you. its because the particular challenges posted dont reflect anything anyone would ever do in real life. if you would like to post a scene of say, a condo, full of furniture and props, ready to be lit and textured AND a photo of the real thing it's supposed to match, you may get a more receptive response since the outcome actually shows something practical. it will also kick into your brain just how much of this outcome relates to artist SKILL rather than hiting render buttons.
later
i'll probably get yelled at for feeding the trolls, but i would like to give my opinion about brazil's lack of those 2 particular features(displacement specificaly).
i have been evaluating displacement recently in every renderer available to me (basically everything ever made) for potential use in our pipeline at work(feature film vfx company). i still have a few renderers to go, but so far, a few things are very clear to me. i'll share some with you.
there are 2 things in particular i have been trying to achieve as reference for production usability.
the first is low poly (box primitive) displaced buildings. i was shown an image of displacement for a movie (day after tomorrow) using this method. for reference, it was rendered (i've been told) in prman. the image showed no flaws that i would find unacceptable. note, this type of displacement is "special" because there are many sharp corners and 0 degree draft angles. this puts the renderer to the test, unlike softer gradual displacements which most renderers handle easily.
using standard max displacement (scanline, brazil, vray all have, of course, the same result) i was able to achieve a decent result only with a very high initial tesselation. this you should agree rather defeats the purpose of using a low poly building with displacement vs a modeled one.
soo... on to the spiffy new(and old) sub pixed displacement type renderers. i'll just mention the 2 most people have access to, mental ray and vray.
- mental ray works, but yeilds extremely poor quality at low settings, and exteemly slow(compared to normal displacement) at higher ones. oddly, it also uses more ram!? but, in the end, it is infact capable of achieving the desired result.
- vray in 2d mode "seems" to work. its fast, and looks nice... in the parts that dont have nasty errors. unfortunately, no matter what settings i chose, and how i tweaked my maps, i could not produce a result that was not riddled with unacceptable errors. in 3d mode, i was introduced to a whole new set of glitches, and again, in the end could not create anything remotely usable. i wont get into the details of the errors(this is a brazil forum after all), but i will be reporting these as bugs shortly.
my second test is trying to use displacement maps created in Z-brush on a low poly mesh. many of you might have been playing with this software, and likely have similar (or maybe different) experiences. anyway, on to the results...
- standard max displacement (scanline/brazil) seems to project the displacement along the smoothed normal, from the face. this is not good for low poly meshes because your displacement takes the shape of the low poly mesh. the only solution is to meshsmooth. in standard displacement, this is acceptable, since you are going to end up with tons of polys anyway. however, this utterly defeats the purpose of using a low poly proxy and displacemnt(well, not utterly, you still have the advantage of fast viewport and skinning performance). to see what i mean stick any displacement map on a sphere with segments set really low(8 or 10)
- mental ray performs perfectly in this situation, displacing as if the low res mesh were actually smooth, however, like on the buildings, its is either crappy quality, or painfully slow, and for reasons unknown to me, does not save any ram. again, you can test this with the same 8 segment sphere
- vray behaves exactly like the scanline in terms of not respecting smoothing, forcing you to meshsmooth, negating much of the point. it also exhibits many of the errors it did on the building if the map is not sufficiently "blurry". so, for the purposes of low poly displaced characters, it offeres little to no benifit over brazil/scanline, and this only if you manage not to get any errors.
lighting:
here's a little insight into how (MR/prman, not sure on vray)displacement works when combined with ray tracing. in order to raytrace the displacement, like in brazil, an acceleration grid must be set up. in order to set up said grid, you need to pre calculate all the displacement. wait a minute, doesnt that negate all the benifits of sub pixel fancy schmancy displacement techniques?. well, yes. but theres a twist(in mr and vray anyway, have not got to prman yet), combining displacement and raytracing absolutely decimates render times(moreso in vray than MR). adding a simple area light to a displaced building increased the rendertime 10 fold over just displacement, or just an area light. this does not happen in normal max displacement.
alrighty then, i'm done spouting off. but now the question is...
how important is displacement as a feature if the displacement you have is either no better, no faster, no more memory efficient, and worst of all full of show stopping bugs.
in the end.. and in keeping with the fishies philosophy of not releasing things till they are release worthy, i personally would rather wait for something good, than have a feature just for the sake of having the feature even though i can never use it. displacement that works WILL be a "development".
now, having read this, you might also understand why i, and others here, are completely uninterested in the challenges posted comparing renderers. its not that we dont like chalenges, or are close minded, or hate you. its because the particular challenges posted dont reflect anything anyone would ever do in real life. if you would like to post a scene of say, a condo, full of furniture and props, ready to be lit and textured AND a photo of the real thing it's supposed to match, you may get a more receptive response since the outcome actually shows something practical. it will also kick into your brain just how much of this outcome relates to artist SKILL rather than hiting render buttons.
later
Comment