PDA

View Full Version : Talking about Hypervoxels



Dan_Ritchie
09-19-2014, 01:27 PM
There are places where LW is fairly cutting edge, and there are places where we haven't seen updates in a while. Hypervoxels is certainly a very powerful tool, but it's not always powerful in the way users want to use it today. I'd like to talk about how it can be improved. There are many ways it could be improved, but as always, my hopes are to try and open it up or lift some of the burden off of it so others can pick up and carry it where it needs to go more easily. Hope some of this might help.


https://vimeo.com/106620342

prometheus
09-19-2014, 02:03 PM
interesting...

but...huhh skytracer...drop it and implement a full volumetric layer with dponīt sunsky instead...but you have some programming insight here..which I donīt understand, so really not sure
why skytracer would help aiding a volumetric hypervoxel shape object?
As mentioned skytracer is slow, canīt fly over it, it canīt cast shadows, and the sky looks worse than dponīt sunsky...for clouds itīs better to throw in hypervoxels with dpont sunsky.

I still need to check your new software out..the shading is intresting, but it doesnīt look good without some sort of hypertexture..so maybe you are planning on adding that?

We do have a thread dedicated to what could be...
http://forums.newtek.com/showthread.php?139560-INeed-Your-feedback-for-the-next-generation-of-Hypervoxels-features-improvements-etc&highlight=hypervoxels



hereīs my basics...
New features..

1. distance between particles..gradients..see previous showcase from erikals and Mr rid.
2. Blending mode tension enhancement..like dynamite plugin or modo.
3. geometric item volumetric Like modo/houdini...particles, vertices is to limited as of todays standard.
4. make hypervoxels available to use null groups in the same hv shading group, as it is now you canīt ..which means each null is adding double the calculation of hv shading and ligth for each null, for
directing and setting up cloud groups..itīs common to use nulls, or you would have to wrestle with points and switching between modeler and layout each time for editing..not very workflow friendly.
5. density or softness curve editor.
6. secondary advection noise.
7. black body fire temp shader.
8. Volumetric handling improvement with volumetric lights (huh..try to use volumetric lights on clouds) speed it up!!
9. infinite volumetric cloud planes...maybe based on hypervoxels...or implement new stuff like ogo taiki.
10. access to node materials.
11. openGL display of volumetric mode, like dynamite had..the display of opengl hypvoxels in volume mode is very poor and donīt give you a volume or 3d sense at all.
12. sprite normal rotation of hv sprite clips...would make it easier to acheive a solar flare cascade with particles on a sphere and hv sprites with premade particle simulations renderd to image sequence clips.
the rotation in hv clips are not allowing for a normal rotation of the clips.

Fixes...

1. falloff should work in basic falloff, as it is now it actually only works with hypertextures in node editing.

2. a thickness local density gradients needs fixing..it doesnīt work, local density works in dissolve channel though.

3. hypervoxels are distorting procedural surface textures other objects.
4. texture patterns in hypertextures need an ovierview..some have to much circular cutoff patterns, It would be nice if the texture coud have some offset falloff within the hypertexture itself so itīs boundary edges
can be blank/black...that will avoid sharp cutoffs when blending with other particles or nulls.

OFF
09-19-2014, 02:03 PM
I think Dan should be taken in Newtek ))

Dan_Ritchie
09-19-2014, 03:32 PM
Skytracer is slow because computers are slow. Volume marching is just going to be slow on a CPU. Memory architectures are slow -- very very slow.
Most of the realtime stuff we are seeing out there is done on expensive GPUs. GPUs can mask the slowness of ram (somewhat) by working on a lot of threads at the same time, but there are limitations there to, because with raytracing a lot of threads, they may not all be doing the same thing at the same time. Most GPUsIt these days are SSE in nature.
Writing a robust and stable gernal purpose volume renderer for GPU is still very difficult for several reasons. GPUs do not currently handle multitasking very well. GPUs do not generally work in system memory, so there are memory limits. GPUs are bad at things like random numbers. Passing pointers back and forth isn't something you regularly do. Multiplatform options are very limited as well, so there are all kinds of limitations compared to a CPU still.
Give it another year or two and I think we'll start to see some of these problems resolved and you'll start to see GPUs that work a lot more like traditional CPUs, only with massive parallelism. Also, there should be DDR4 to help speed memory up a little.
Anyway, I"m a proponent of having options for both doing things quickly, like sprites, and high quality, like ray marching. There's a need for both.

prometheus
09-19-2014, 04:02 PM
skytracer seemed slower than ozone spectral 1 layers and even slower than ogo taiki.., but anyway I donīt like the fact that though it being raymarching, no cloud shadow casting nor did it seem to produce good volume clouds ..nor any option to fly above it., and I didnīt like skytracer look overall for the sky/air..it had to much of artificial look to it.

Easier to just through in hypervoxels in there with sunsky..

What I miss with sunsky is a true volume marching fog/godray layer like ogo taiki seemed to have, and also infinite cloud textured based volume layers, sunsky is also missing a volume light option for the sunsky which forces you
to use either a distant or point light instead..which makes it harder to use with proper world global scale of the volumetric light ..since it is based on the light radius and settings, it is very difficult to get the same realism in volumetrics as a full fog/volumetric layer.
using backdrop color in the fog setting..thus artifacts/edge cut off of the sunsky is over imposed on the terrain which can be seen on the second image, Havenīt found a way around that for direct renders.

Images are particle fields with hypervoxels, terrain some dem files with procedural mixing and instanced a few times, dpont sunsky and sunlight, and some fog, three gradients in dissolve channel, local density, distance to camera and an image map to dissolve y axis a little.

http://forums.newtek.com/attachment.php?attachmentid=124396&d=1411163731

http://forums.newtek.com/attachment.php?attachmentid=124397&d=1411163763


heres a fiddle and experiment page of mixed ogo taiki, hypervoxels, some houdini stuff too....
https://plus.google.com/photos/100944643113557837045/albums/5978023753291712209


Michael

erikals
09-23-2014, 10:05 AM
related, i liked this modo test >


https://www.youtube.com/watch?v=d9OQoz68xag

erikals
09-23-2014, 10:12 AM
kinda reminds me of the LW5.6 Steamer Smoke, though to me the Modo one looks to have a slightly better algorithm...
could be wrong though, they might be basically the same...


http://youtu.be/IDMTm3bh0DY


it might be that Ogo Taiki can pull off the same result, or better >
https://www.youtube.com/watch?v=cg8w8AVzsO8

shrox
09-23-2014, 11:18 AM
The render times are what kill my projects.

prometheus
09-23-2014, 11:49 AM
related, i liked this modo test >




Yes...I think they have improved the blending mode in modo, similar to what the dynamite plugin had which I think is better than hypervoxels blending..which I donīt believe is working properly between particles in a surface tension style..that is a must .

Modo looks way better at least from what I have seen in some images of the blending mode, it also has a softer edge in the quality of the volumetrics than we
do have with hypervoxels, it also has more advanced light forward scattering, and mie scattering which seems to have a nicer smoother shadow blending than rayleigh?.
I should actually install the modo trial and try it out first hand..but I got so much else to do right now...



kinda reminds me of the LW5.6 Steamer Smoke, though to me the Modo one looks to have a slightly better algorithm...
could be wrong though, they might be basically the same...



it might be that Ogo Taiki can pull off the same result, or better >
https://www.youtube.com/watch?v=cg8w8AVzsO8

Steamer ? maybe...I wouldnīt be surprised it they simply took the tech behind steamer and improved on that.

My wet dream is that newtek secretly has bought the plugin or disected it and during these last two years improved it in terms of UI, and air set up and quality settings along with a 10 times speed boost, as it stands now I think it still is too slow for final renders..(vpr decently) and the quality is hard to get rigth everytime, and the UI is to messy along with setting up basic sky properties, If I could simply use dponīt sunsky with ogo taiki godray/foglayer and itīs infinite procedual textured cloud layer
that would be sweet...or at least get the same easy setup and algorithms in the ogo taiki air layers, it would be so much easier to use.

prometheus
09-23-2014, 11:56 AM
The render times are what kill my projects.


50 min to 2 hours per frame for volumetric environment is probably what to expect, and that with basis on vue ..which isnīt really slow considering, the volumetric algorithms are complex and demanding, and with advanced terrain, like infinite procedurals..and
ecosystem..I reckon you will end up with several hours per frame.

my images above consisting of displaced terrain object at over 1 millions of polys, instanced 25 times and a huge particle field of 3000 particles with hv sizes of several km per particle, took around 27 minutes per frame, but with no depth of field and low AA at 3x3, if I were to make volumetrics light with that...boom count on additional hours to the renders.


I think MPC had around 8 hours per frame for itīs landscape stuff in the prometheus movie, due to the high poly amount and subpixeld displacement only, without any volumetrics in there since I think it was mostly footage comping for that.

prometheus
09-23-2014, 12:01 PM
Sorry this was the image that took 27 minutes, just experimental since I didnīt do a proper blending beetween the flat ground textures and the terrain objects, a mix of sculpted sculptris main shape, exported a heightmap out from sculptris, in to
worldmachine for erosion filters, then exported out 16 bit tiff image from world machine in to displacement in lightwave...and rendered with hypervoxels for clouds, just experimental texturing not completed....
I cloned the instances manually for rescaling and rotating and placement with manual variation control.
Dp_sunsky and dp_sunlight set to manual mode for exact control of lighting.
some volumetric fog with background color activated.
also one additional dome light for overcaster sky light, no radiosity.

unless doing postprocessing on the Hvīs you really need to set itīs quality to good, and that of course increases rendertime a lot, I almost consistently turn off textured shadows nowadays.
you can also get a few minutes of speed up by turning of volumetric aa in the volumetrics tab, but you need to have that on if you use volumetric lights in the scene, the impact on the quality of hvīs itself is very small though.
http://forums.newtek.com/attachment.php?attachmentid=124440&d=1411495217

124440

erikals
09-23-2014, 12:03 PM
the more i look at that Ogo example, the more i feel Ogo Taiki is the way to go...

though, got some new ideas recently when it came to HyperVoxels... will test it later on i guess...
Ogo is better for animated clouds though, for sure, as of the blending

hey, did you test some more Houdini btw ? and are you buying it ?

------------------

that render is Nice! http://erikalstad.com/backup/misc.php_files/biggrin.gif

actually bought World Machine way back... http://erikalstad.com/backup/misc.php_files/smile.gif

prometheus
09-23-2014, 12:20 PM
the more i look at that Ogo example, the more i feel Ogo Taiki is the way to go...

though, got some new ideas recently when it came to HyperVoxels... will test it later on i guess...
Ogo is better for animated clouds though, for sure, as of the blending

hey, did you test some more Houdini btw ? and are you buying it ?

------------------

that render is Nice! http://erikalstad.com/backup/misc.php_files/biggrin.gif

actually bought World Machine way back... http://erikalstad.com/backup/misc.php_files/smile.gif


what ogo sample do you mean? not mine? my images here are all hypervoxels..but you ment your own samples?

Houdini is out of the question for the moment, need to strengthen the economy first..and that will probably take a year or more.

Yeah..the render could be nicer though, I spend to little time with each scene and often abonded it to test new things, often I find myself also realizing that I can not acheive the results I desire too...and partly because of hypervoxel limitations and displacement object level limitations, and the general volumetric handling and lack of such in Lightwave, a little frustration within that...since I donīt want to invest in the latest vue versions and work in that environment either, though some stuff in there would get me to my desired goals easier.

Whatīs nice with true volumetric clouds is that you can design them almost as you want and have real ray cast shadows on the terrain making it more dramatic and realistic, compared to comping footage and fake shadows.

hereīs a mix of ogo taiki stuff and hypervoxels..though it is so much harder to acheive smoother quality looks in ogo taiki ..than it is to set up clouds with hypervoxels.
https://plus.google.com/photos/100944643113557837045/albums/5978023753291712209?banner=pwa

ogo taiki faces a few points that are an obstacle

1. no 64 bit support if you need high amount of poly objects in 64 bit lightwave.
2. the ui is simply to confusing.
3. the setup of air properties isnīt very nice to get started with.
4. the quality settings is a mess and hard to get right.
5. render speed.

apart from that...it could be awesome and kick ozoneīs butt.

erikals
09-23-2014, 01:41 PM
1. no 64 bit support if you need high amount of poly objects in 64 bit LightWave
sometimes one can solve it in comp, sometimes not. agree, i'd really like to see 64 bit support

2. the ui is simply to confusing
agree, this is why i use the Taiki presets and tweak them, much easier

3. the setup of air properties isnīt very nice to get started with
haven't looked into this one, can't recall...

4. the quality settings is a mess and hard to get right
nah,.. you think ?

5. render speed
yep, this will always be a problem though, with HyperVoxels and TurbulenceFD also

prometheus
09-23-2014, 02:27 PM
5. render speed
yep, this will always be a problem though, with HyperVoxels and TurbulenceFD also

Indeed so, though turbulencefd in some cases can render faster, but it is dependent on if you use certain illumination shadowing and how you set that etc, besides, to get really good cloud shadow and lighting, I think
he would need to implement other algorithms designed for cloud illumination specificly..and that would probably further add rendertime.
Apart from that..adding subgrid detail and manage to simulate at large scale for cloud vistas will be problematic.

Houdini cloudfx is funny, awesomely impressive to see it in real time openGL, but you still have to render it, and to avoid grain and noise you have to increase settings so in the end you will get very slow renders anyway.

no matter how you dig in to it in most of the software out there, the problem of slow rendering will still be there it seems.

What dan ritchie suggest is interesting, a new shading method (lambert?) though the samples shown donīt look good without any hypertexture to cut in to the blobby volumes.

Impressive to see lumen from eonsoftware, based on vue.. but crippled in the quality of the cloud shadow/illumination and alos of the detail, just to make sure you can adjust it in Realtime for more archviz realtime presentations.
http://www.youtube.com/watch?v=woSiikekxus&list=UUjJLjOANw3Bwaf-zGqVoakQ

erikals
09-23-2014, 03:33 PM
Lumen is ok, but i'm sure we can do similar inside LightWave...

some comments, see attachment...


http://forums.newtek.com/attachment.php?attachmentid=124449&d=1411507951

http://forums.newtek.com/attachment.php?attachmentid=124450&d=1411507951

prometheus
09-23-2014, 04:08 PM
Lumen is ok, but i'm sure we can do similar inside LightWave...

some comments, see attachment...


http://forums.newtek.com/attachment.php?attachmentid=124449&d=1411507951

http://forums.newtek.com/attachment.php?attachmentid=124450&d=1411507951


No...not really:) itīs quite beyond lightwave realtime capabilities and have realtime shadows in opengl is still something to wait for, and have clouds casting shadows in realtime too.
and thereīs no way to animate clouds in realtime either, unless faking it with animated clips, but then you donīt get any shadows and you canīt change lighting conditions or cloud shapes etc..
so something similar is probably not the right term for it.

I agree with you about the quality though..it is sacrificied in order to get realtime editing and presentation, to blurry is due to the lack of cloud detail and variated shadow cloud density..jitters..havenīt checked.

Dont think it is based on photo, I say procedural driven as with vue...otherwise it wouldnīt be editable in the way you can change it now.