PDA

View Full Version : Thoughts on nParticles in Maya Version 2009


mayaartist
08-18-2008, 02:17 PM
Hey everybody I was wondering what your thoughts were of nParticles based upon Maya's Nucleus solver? Anyone used it yet?

Duncan and Yos, those guys are brilliant, and I must say I am looking forward to using the new system.

oxygencube
08-18-2008, 02:33 PM
Haven't used it but I am excited to have particle/particle collision.

DJ_
08-19-2008, 10:10 AM
I think that the features of nParticles are looking really awesome and I'm really looking forward to trying it out, BUT I'm a bit disapointed that Maya still doesn't have a node/event based particle system (like Houdini, Ice, Particle Flow, Thinking Particles, etc...).

lazzhar
08-19-2008, 11:12 AM
I'm wondering as well, if there are any speed improvements comparing to what it was available by hacking the nCloth system.

Duncan
08-19-2008, 08:19 PM
I'm wondering as well, if there are any speed improvements comparing to what it was available by hacking the nCloth system.

The nParticle collisions are optimized to particles, which unlike the cloth can change topology dramatically, so it should be significantly faster in most cases.

Duncan

xcomb
08-19-2008, 08:25 PM
Yes i totaly expected to see node/event based particle system (like Houdini, Ice, Particle Flow, Thinking Particles, etc...). But unfortunatly it's not :( will that ever be done in Maya?

DAZZAN
08-19-2008, 08:37 PM
Hi duncan,could not find the video showing a render of the volcano you did,but as you know there is a thread just on Fliud shaded particles,are n particles in the new version helpfull in creating realistic clouds in the sense of afterburn in max.Im sorry, im a push a few buttons kind of guy,and fluids and clouds just seemed a bit hit and miss to get right.Could you elaborate a little on this and the volume rendering in mental ray,or do you still have to use the particles and fluids together?

It seemed that n particles are up my street with easier setup, but is cloud building and shadeing any different.

great stuff.

thanks

dazzan

hkspowers
08-19-2008, 08:54 PM
Yes i totaly expected to see node/event based particle system (like Houdini, Ice, Particle Flow, Thinking Particles, etc...). But unfortunatly it's not :( will that ever be done in Maya?

Well if it's not broken don't fix it. For years studios have used Maya with great success and it has been through countless trials of the toughest demands. I for one have been trying to get used to Houdini for quite a while, and while I am better at it than when I first started I find it extremely frustrating to get anything halfway decent looking out of it without having to have a dedicated shader writer and renderman expert on hand. yeah sure it has the nice node based non destructive editing of particle systems, but I have never once felt the need to have nodes as a must have. In fact the new maya features have included most of the things I have always complained about in Maya. Like self colliding particles, and particle meshing, etc. To me this is a very useful and content rich Maya release and I am very excited about it. If you want a node based package then just learn Houdini or XSI but for me, I will take Maya anyday and learn Houdini at my own pace.

xcomb
08-20-2008, 05:57 AM
Well i've been with Maya for many eyars and it still is my main application, but i've dealt with Houdini for couple of years and it has lots of nice stuff which may take long time to make in Maya.
I was actully excited to hear that now Maya has Asset, which is similar to Houdini Digital Asset, and it's a good stuff!

DJ_
08-20-2008, 08:36 AM
The nParticle collisions are optimized to particles, which unlike the cloth can change topology dramatically, so it should be significantly faster in most cases.

Duncan

Since nParticles are running "under" the nucleus core, does it mean that all it's dynamic calculations are multi-threaded?

Duncan
08-20-2008, 08:02 PM
Since nParticles are running "under" the nucleus core, does it mean that all it's dynamic calculations are multi-threaded?



No. There is a bit of threading in nucleus, but it is still largely unthreaded. We have some structural optimizations we need to do before threading will speed things up. The good part is that as we improve general nucleus speed all the various components like nParticles and nCloth will benefit.


Hi duncan,could not find the video showing a render of the volcano you did,but as you know there is a thread just on Fliud shaded particles,are n particles in the new version helpfull in creating realistic clouds in the sense of afterburn in max.Im sorry, im a push a few buttons kind of guy,and fluids and clouds just seemed a bit hit and miss to get right.Could you elaborate a little on this and the volume rendering in mental ray,or do you still have to use the particles and fluids together?


That video ended before I showed the volcano render. Basically the technique is to use a fluid shader on the particles, however this is now supported in Mental Ray and supports shadowing between particles, not just within a particles. As well it is more efficient and there is built in UI for setting up the general shader( one uses the the "thick cloud" particle style). At some point I would like to create a simplified shader( the fluid node has a lot of simulation stuff on it that is not needed for simple textured volumetric rendering), but at any rate one can now create decent looking volumetric particle renders.

We investigated going towards a more nodal based system( like Houdini ), although the most prudent course was to leverage the current particle system, which is actually quite powerful, but much of that power has been relatively hidden(actually a lot still is... In future I hope we can expose more of this functionality through the UI). I think that for things like gas, dirt explosions and such one will be able to get to the final rendered frame fairly easily with nParticles. For things like crowd simulation and flocking one will still often need expressions and for this a nodal based system would prove useful. The nodal systems are often basically visual programming languages, which are sometimes nice, but can get rather complex. However I particularly like some abilities in Houdini like being able affect construction history per particle instance. If we add nodal capabilities to maya particles at some point I see it taking the form of a network plugging into a particle system in a hybrid approach rather than having a fully nodal particle system.

With nParticles we have added a lot more behavior and rendering control to the base particle node, which in turn makes attribute presets much more meaningful in terms of defining general particle look and behavior. One thing I would like to do is use the new container nodes for creating simplified UIs for various preset particle setups. This would be sort of like the old particle effects menu, but with custom UIs that allow one to vary the high level settings after creating the system. We did not have time for 2009 to create any such presets, but it is something TDs should find useful.

Duncan

pixelAffairs
08-21-2008, 08:30 AM
One of my first toughts was how succesful you could blend between the different "states"/solvers, or if you could even transform them.
What i am talking about is nMesh becoming nParticles or nFluids (once they arrive :) ) etc. based on events/rules.
Optimized mutual influence and shared attributes are nice, but i can see some real power coming from a "arbitrary Nucleus".

republicavfx
08-21-2008, 04:18 PM
nParticles look like a great addition. Can't wait. Its true a node based approach would be useful at times and for reuseablity. Although i had heard Thinking Particles would be ported to Maya at some point it would be nice to keep things native and make fx development faster.

DAZZAN
08-21-2008, 08:20 PM
I wish to thankyou for telling me these facts,its nice a that perhaps a future simplified shader is in your mind,

Looking forward to the new implementation.

regards


That video ended before I showed the volcano render. Basically the technique is to use a fluid shader on the particles, however this is now supported in Mental Ray and supports shadowing between particles, not just within a particles. As well it is more efficient and there is built in UI for setting up the general shader( one uses the the "thick cloud" particle style). At some point I would like to create a simplified shader( the fluid node has a lot of simulation stuff on it that is not needed for simple textured volumetric rendering), but at any rate one can now create decent looking volumetric particle renders.



Duncan[/QUOTE]

azshall
08-22-2008, 03:11 AM
With nParticles we have added a lot more behavior and rendering control to the base particle node, which in turn makes attribute presets much more meaningful in terms of defining general particle look and behavior.

Hey Duncan, this all sounds awesome. I can't wait to play with nParticles, however I am curious about Hardware Particle types and rendering.

Is mental ray more integrated? I understand that in 2008 mental ray can render sprites, etc but it seemed to lack on some things like some per particle attributes. Have these things been addressed at all? If so, do sprites in Maya possibly render in the same fashion, say using Renderman/3Delight with deep shadows (which give you a very fast and very nei volumetric effect)?

Duncan
08-22-2008, 04:49 PM
I don't think the MR sprite rendering has changed too much. For a more volumetric "sprite" one could texture the software cloud shader in a similar fashion to sprites. The particleSamplerInfo along with the current frame can be used to define the file texture sequence in a similar manner to sprites. Note that for rendering in MR one can't use mel expressions in a shading network context, so the logic for determining the sequence needs to be created with utility nodes( the sprite wizard creates expressions for this, which won't work in MR).

ParticleSamplerInfo(the main shading control for particle sw renders), now works well in MR and I would say that most of the volume and blobby particle rendering is significantly better than with the mayaSW render.

Duncan

azshall
08-22-2008, 05:24 PM
I don't think the MR sprite rendering has changed too much. For a more volumetric "sprite" one could texture the software cloud shader in a similar fashion to sprites. The particleSamplerInfo along with the current frame can be used to define the file texture sequence in a similar manner to sprites. Note that for rendering in MR one can't use mel expressions in a shading network context, so the logic for determining the sequence needs to be created with utility nodes( the sprite wizard creates expressions for this, which won't work in MR).

ParticleSamplerInfo(the main shading control for particle sw renders), now works well in MR and I would say that most of the volume and blobby particle rendering is significantly better than with the mayaSW render.

Duncan

I see.

So, instead of using, say a runtime expression to iterate through spriteNumPP you would need to use utility nodes or an Array Mapper to do it, and then it would work? ... Ooof, I guess one should consider this route before starting anything and tailor the effect to strictly mr rendering :)

DavidLessel
08-22-2008, 05:30 PM
The new particles seems very powerfull indeed, can't wait to get my hands on them. As I usually rely heavily on the use of particles in my work it is very nice to hear that attention has been paid to integration of the particle sampler info node in MR :)

Duncan:
Can you say anything about wether or not the instancer has been given an overhaul too? One of its greatest shortcomings is that it isn't possible to control shaders on the instances on a PP level. If for instance one needs 100 different colored boxes, then, with the current implementation one would have to create 100 boxes with 100 different shaders on them... Can you disclose any info on this subject? :)

Duncan
08-22-2008, 05:36 PM
So, instead of using, say a runtime expression to iterate through spriteNumPP you would need to use utility nodes or an Array Mapper to do it, and then it would work?

No...Runtime particle expressions in MR are fine(these are computed before the raytrace of the raster commences). The problem is expressions inside shading networks (one could create an entire shader in maya sw with expressions only, but it would be dog slow). In particular the sprite code creates an expression that sets the file texture extension value per particle.

Probably the easiest way to handle this would be to create a userPP attribute for the frame extension per particle and then stuff it in aruntime expression. One could then simply connect particleSamplerInfo.userPP to the the frame extension on the file texture that is assigned to the particle shader color and transparency. ( it does not make sense to do complex logic to determine things like this at shade time... it is much better in particle runtime expressions)

Duncan

Duncan
08-22-2008, 05:47 PM
One of its greatest shortcomings is that it isn't possible to control shaders on the instances on a PP level.

Agreed. ParticleSamplerInfo should work with instanced particles but currently it does not. I pushed for this, but we were not able to get it working for Maya2009(it is more complex a problem than appears at first blush). For now if you want different colored boxes you should instance from a set of boxes of different colors instead of just one box. (still doesn't address things like changing the color of a box based on the current particle age, etc) There are lots of issues with instancing that I'm hoping we can address in future releases.

Duncan

Als
08-23-2008, 10:29 AM
Since MR already works with switch, would be possible to hack particle sampler with switches to get instanced particles geometry to work ?


Thanks


Als

DavidLessel
08-23-2008, 10:44 AM
There are lots of issues with instancing that I'm hoping we can address in future releases.

Duncan

I'm glad to hear that, controlling shaders on instances on a per particle level would make the instancer a very powerfull tool. For now I can't wait to get my hands on the stickyness feature of the nParticles :)

Duncan
08-23-2008, 04:48 PM
Since MR already works with switch, would be possible to hack particle sampler with switches to get instanced particles geometry to work ?


ParticleSamplerInfo does not get per particle values for the instance render so that won't work. Currently there is no way I'm aware of to associate the particle with it's instance at render time.

Duncan

MikeRhone
08-23-2008, 07:33 PM
I just cheered when I read this: "however this is now supported in Mental Ray and supports shadowing between particles, not just within a particles."

Booyah!

Als
08-24-2008, 12:07 PM
ParticleSamplerInfo does not get per particle values for the instance render so that won't work. Currently there is no way I'm aware of to associate the particle with it's instance at render time.

Duncan

I miss this feature for a long time. If I remember well on some thread someone showed that this can be done in final render already. I really hope that this issue can be addressed soon.

Thank you for your replies


Als

drGonzo
08-24-2008, 01:55 PM
Yes i totaly expected to see node/event based particle system (like Houdini, Ice, Particle Flow, Thinking Particles, etc...). But unfortunatly it's not :( will that ever be done in Maya?

I hope never. I for one feel great dread and suspicion towards this "No need for expressions", or "Code less, achieve more" Zeitgeist as of late.
There is no way around it, FX is the most technical field of 3D and to be good at it, you need a firm knowledge of physics and (if you are doing them in Maya) MEL/expressions. No "visual programming system" will every replace that. Houdini does it good indeed, but they have been at it for 15 years. And - contrary to the urban legend - to get something decent out of Houdini you do need (simple) expressions; not to mention years of experience.

It is with a snicker and a whimper that I watch demos of ICE and Thinking Particles where conditionals (if...else) and "get data calls" are done by connecting one of a gazillion nodes in a workspace. This looks all very nice in meticulously prepared demos, but I can guarantee you a world of pain (not to mention massive dependencies) if you want to create something of feature film level.

Similarly, there is no way you will be able to create advanced, photo-real effects with nParticles without expressions. The ramps and sliders and all are very handy and slick, but to get realistic particle effects, you will need vector math and expressions. Always have, always will.

Back on the topic of nParticles, I wonder if it will be able to do real liquids, as seen in Fedkiw's stuff (http://physbam.stanford.edu/~fedkiw/ (http://physbam.stanford.edu/%7Efedkiw/)), Scanline, or even RealFlow. Because the liquids I saw in the Siggraph videos looked very blobby.

Als
08-24-2008, 06:57 PM
I have to admit this is same in realflow, unless you really get enough particles for definition, which means quite a lot. I guess for demo this would be way too slow to show, so it's really a mater of it everything works the same if you change number of nparticles.
This is only major issue I have with ncloth, that it's hard to change number of polygons and get the "same" simulation.
I'm certain it looks better with more nparticles, but how many you need etc. I can only guess.

Als

Aikiman
08-25-2008, 03:23 AM
Back on the topic of nParticles, I wonder if it will be able to do real liquids, as seen in Fedkiw's stuff (http://physbam.stanford.edu/~fedkiw/ (http://physbam.stanford.edu/%7Efedkiw/)), Scanline, or even RealFlow. Because the liquids I saw in the Siggraph videos looked very blobby.

That guy Fedkiws should win the noble peace prize or something for the stuff he can do especially cracking the holy grail of all FX - crashing waves. Im not sure nParticles will ever do that because Fedkiws uses completely different algorithms - SPH and particle level set methods. Nucleus would have to be rebuilt.

azshall
08-25-2008, 05:00 AM
That guy Fedkiws should win the noble peace prize or something for the stuff he can do especially cracking the holy grail of all FX - crashing waves. Im not sure nParticles will ever do that because Fedkiws uses completely different algorithms - SPH and particle level set methods. Nucleus would have to be rebuilt.

Not to mention that LucasFilm and ILM have Ron Fedkiw on the payroll as a consultant. He writes PhysBAM, with his students and other dudes at Stanford. That is ILM's fluid effects tool, among many other things... Shits pretty insane.

Aikiman
08-25-2008, 09:19 AM
Not to mention that LucasFilm and ILM have Ron Fedkiw on the payroll as a consultant. He writes PhysBAM, with his students and other dudes at Stanford. That is ILM's fluid effects tool, among many other things... Shits pretty insane.

Theres no way Maya can compete with that technology, im guessing all that stuff is patented also so you cant just take it and use it for your own pipeline or have it integrate into Maya development, what a shame.

azshall
08-27-2008, 08:13 PM
Theres no way Maya can compete with that technology, im guessing all that stuff is patented also so you cant just take it and use it for your own pipeline or have it integrate into Maya development, what a shame.

EH.. I guess, proprietary software is great and all but as an artist it can be a bad thing.

What happens if you get amazingly good with said proprietary software that nobody else has? You leave ILM which uses PhysBAM to go to another house that only uses Maya Unlimited. ...Quite the contrast, and not by features but by methodolgy and workflow.

Do a shot on your reel with said software, ..sure.. looks amazing. Go to that little house and try to do the equivalent without it, might not be as good as you thought you were.

Aikiman
08-27-2008, 08:44 PM
EH.. I guess, proprietary software is great and all but as an artist it can be a bad thing.

What happens if you get amazingly good with said proprietary software that nobody else has? You leave ILM which uses PhysBAM to go to another house that only uses Maya Unlimited. ...Quite the contrast, and not by features but by methodolgy and workflow.

Do a shot on your reel with said software, ..sure.. looks amazing. Go to that little house and try to do the equivalent without it, might not be as good as you thought you were.

true dat...

mayaartist
08-27-2008, 10:51 PM
azShall,

That's true for the most part, but if you have been at a big house such as ILM then someone at Dreamworks or Sony for instance will usually have a training period where you can switch over to their in house software.

I agree very much if you are to go to a small house they won't have time to train your butt.

azshall
08-27-2008, 11:44 PM
azShall,

That's true for the most part, but if you have been at a big house such as ILM then someone at Dreamworks or Sony for instance will usually have a training period where you can switch over to their in house software.

I agree very much if you are to go to a small house they won't have time to train your butt.

Yeah, and usually those studios do train you for their in-house tools. Just kinda sucks when you're acclimated to said tools and are unable to use them. However, there is more than 1 way to skin a cat I suppose :)

mustique
08-28-2008, 03:35 PM
Fedkiw's work is nice. Sure it would be nice to recreate those effects with our apps, but those things probably need insane amount of proccessing power.
So it might only be feasible for a couple of big studios.

CGTalk Moderation
08-28-2008, 03:35 PM
This thread has been automatically closed as it remained inactive for 12 months. If you wish to continue the discussion, please create a new thread in the appropriate forum.