Is XSI instance function same as (LW) HD Instance...

[alvin_cgi]

Is XSI instance function same as (LW) HD Instance plugin…

I want to do some basic crowd simulation in XSI, but wondering XSI instance(not Behavior) can do similar thing like LW HD instance.

Thanks

[ViCoX]

= D oh, we use also lw:s HD instance in crowd stimulation. : )
Im afraid that the answer is no. : (

[alvin_cgi]

thanks... i guess have to export it back to LW... :(


= D oh, we use also lw:s HD instance in crowd stimulation. : )
Im afraid that the answer is no. : (

[mocaw]

= D oh, we use also lw:s HD instance in crowd stimulation. : )
Im afraid that the answer is no. : (

And why is the answer no? Just asking...because I don't use HD instance...but see nothing in the features set, nor the HD instance PDF that leads me to believe you couldn't do it in XSI.

Please elaborate so that someone with a fair amount of XSI experience can either thankfully prove you right or wrong on it.

[ViCoX]

Theres some neat features like you can instance objects to polygons and have random animation offsets with it. Overall all kind of random values like diffuse ect. are easy to setup
Off course xsi has particle/and hair instancing, but hd instance is bit different.
I would quess that alvin_cgi has done some crowds with simple objects and he wants to instance more complex characters to those at rendertime? Im not 100% sure that xsi can`t do it, though : )

[thiago]

- make a particle cloud or a hair object.
- make all the variations you want and put inside a group.
- Use the particle or hair instance and select the group with all the variations.

EDIT:
This is a way of organize your project and still have control of all instance you'll use to mix in the crowd thing.
Put this with Refmodels and you'll have a good setup.
example: load many copies of a refmodel, load each one with a mixer animation.
Each time you update the refmodel you'll update all "instances" of that model in the crowd.

[franky]

this is a bit theoretical thiago.

the problem starts if you wanna have ONE model (ref or not) instanced 100.000 times but animate it with a different offset.
you need for each offset a unique model in your instance group.

that will soon lead to render problems, as each model comes with its own overhead.

i would wish for a own mixer on instance models, that would solve lots of problems in this regards.

[Adrian Lazar]

i agree with franky, this would be a time saver in many occasions.

[thiago]

the problem starts if you wanna have ONE model (ref or not) instanced 100.000 times but animate it with a different offset.
you need for each offset a unique model in your instance group.


Well if you wanna do 100.000 instances than it's not a "Basic crowd simulation" anymore.
100.000 is wat you put on a digital battle sequence using Massive. No software would handle that amount of objects. You would have to use a geoshader or something to output this to mentalray or renderman.

This workflow I explained work for small groups where you can have up to 200 different animations (or more, but that was my last try).
Lets say you do with 40 variations (which is already reasonable), then load this clips on a Refmodel. You'll just have ONE model x 40 offsets.
In your final scene you load 40 times the same refmodel, and load one offset for each refmodel (you slide it in time as each refmodel have it own mixer).
Now you put this 40 different guys in a group and use it as a instance with a particle cloud or hair object, spreading 1000+ particles.
XSI wouldn't think twice to render this.

I bet you can go much more than that before have any problem at rendertime.
If your characters are not moving in space, you can use hair instances and still count that as a variation to deform the thing.

So 40 variations spread between 1000 guys would make 25 duplicates of each model.
In many cases for basic crowds I bet you wouldn't see that 25 guys are playing the same animation within 1000 people in the frame.
Count also that in particle simulation you have the scaling and the particle motion itself. That will create even more variation.
With a bit of code you can do a lot more too.

That said, XSI still have a lot of work to make the standard "instance" more usable.
I totally agree that an instance alone can't survive if you need to animate something... or do a massive 100.000 army.

[mocaw]

100,000 X?

Aren't we starting to ask a lot of certain things? Besides, wouldn't any sane person at some point just render in separate passes/groups in the crowd and then combine them in post?

I'm small fry, but to me it seems a little "brute force" to keep expecting the 3D software to accomplish this all in one frame once you get above a certain number.

[franky]

i just filled a stadium (European championship, anyone? ;) ) with low rez characters that have about 600 polygons each.

as they're not moving they are instanced via hair and it works terrific. its of course split into passes of about 18.000 ppl in each pass. in each pass its 3 hair instances with each 4 characters in the ref group. you cant split into passes as much as you would need, because at some points you run into masking problems when ppl stand behind other ppl...

thats actually what i complain about. i cant just add characters to the instance group, as then the memory starts to melt away with the crowd. its NOT a problem to render 50.000 ppl in one go as long as its just ONE model that gets instanced.
if i just had a mixer on the instances and could randomize that when the geometry shader kicks in, it would solve a big pile of problems and would make you able to render crowds much more conveniently.

in the end, mocaw, rendering grass with 500.000 strands is no different then rendering 100k characters. its already working right out of the box. so why not just going one step ahead?
i am not asking for the impossible here, the instances already have their own kinematics.

in general i am very much surprised how easy it is to generate a huge mass of characters in xsi without much hassle. it renders smooth and fast. 30sec - 2 minutes for 720p resolution, including shadows. :)

[grahamf]

Well if you wanna do 100.000 instances than it's not a "Basic crowd simulation" anymore. 100.000 is wat you put on a digital battle sequence using Massive. No software would handle that amount of objects. You would have to use a geoshader or something to output this to mentalray or renderman.

HD Instance on LightWave handles it just fine, even with randomized animation offsets. You could do a billion or a trillion characters instanced over a terrain, but then it's just getting silly because if you put them all on screen they'd be smaller than a pixel (but they would still render).

There are 8,000 elves in the scene for the movie on Happy Digital's main page: http://www.happy-digital.com with full radiosity, and it's not even starting to sweat.

- Graham

[xvivo]

HD instance for XSI Mental ray ...perhaps? Graham?

Mike

[thiago]

You could do a billion or a trillion characters instanced over a terrain


There are 8,000 elves in the scene for the movie on Happy Digital's main page: http://www.happy-digital.com with full radiosity, and it's not even starting to sweat.

- Graham

lol thats a big difference then...
you said you can do a trillion or a billion, lets get from the average then:
500 billion - 8 thousand (thats what you said you rendered) so you are missing just 499999992000 to make half a trillion lol

But don't try to go that far, tell me when you put this amount of actual polygons in Lightwave without make it explode:
http://www.xsi-brasil.com/temp/gygapolygoncore.jpg

Now seriously... you can do conscious 100.000 particle/hair instances in XSI using one refmodel and many different mixer animations and shaders without make mentalray sweat.
The thing that Lighwtave will fall apart is that eventually the Geometry core (and the way you manage these instances) will go nuts (or will make you go nuts).
This will start to have serious pipeline problem when you get to manage 100 different characters in a crowd of 50.000 people (to fill a digital stadium shot for example).
Although I've never used HD-Instance... I'm an ex-waver too :) ... and I surely know when Lightwave will make you need something where the software is not well designed.

So considering we have Motor, Delta, Animation Mixer, Gigapolygon core, flushable BSP and a multi-core everything in the near future... my opinion is that XSI will naturally jump to a very powerful Instance system... one day.

[grahamf]

lol thats a big difference then...

You're right, 8000 elves is not a big enough test. Here's one trillion (1,000,000,000,000) elves working:

http://www.happy-digital.com/trillion_elves_9m29s.jpg

I made a 1,000,000 x 1,000,000 meter plane and put one elf per meter. It took 9m29s to render on my 8-core machine, with full radiosity and antialiasing.

I have no intention of belittling XSI here. XSI is a great tool. I fully plan on supporting it in future Happy Digital products, whether it's instancing tools or something else. XSI really looks like it has a great future ahead.

- Graham

[thiago]

lol man this almost looks like an add noise effect!
Now seriously, it Looks great! Maybe you should make an XSI version then... Or it's hard coded to Lightwave?

Let see what will come in the future for XSI. I hope will can discuss this thread again.

[thiago]

you got the 499999992000 that was missing in your previous render!! Congratulations.

[ViCoX]

If I have understaded correctly HD instance renders then volumetrically somehow? We have rendered 17000 statues with 2k normal, diffuse and reflection textures with no hassle at all. : )

[mocaw]

OK, so this is not the old HD instance in anyway! Very impressive indeed. If you can get it working in XSI not only will you sell to all the LW users now using XSI, there are a lot of XSI users that would make use of this.

OR, maybe soft should contact you directly about the possibility of direct integration with XSI. I imagine there are quite a few other tricks that this might be able to do if it were able to tap deeply into XSI...

[CiaranM]

What about the new proxy function in the latest Mental Ray? This may have similar capabilities to HD Instance (at least as far as heavy instancing, I've no idea about animation, offsets etc.).

http://forums.cgsociety.org/showpost.php?p=5023838&postcount=61

Although this doesn't seem to be exposed at all well in others apps., one hopes this will be available in XSI in a usable form.

[ThE_JacO]

HD instance uses a volume shader to run instancing internally, because LW's engine for the longest time hasn't supported anything decent in those regards.

"All" it does outside of the shader domain is take animation offsets and different materials and all, and generates permutations before creating a master of each permutation, filters eventual duplicates, then that list is actually instanced.

To do the same in XSI, which already benefits of MRay's solid enough instancing, all you need is something to generate the permutations for you.

So if you have a model and you want it to have 10 different starting points for its animation you just create one for each offest step and feed that to the particle instancing group, done.
Same goes for materials and all.

The whole basis of efficient instancing is addressing the same cache for each instance of one master, that's why you can't have animations offsets if you just reference one master, because it drastically changes what you're supposed to look up.
Committing to models the permutations of whatever variations you need solves the problem.
I've done exactly this for a number of clients and projects and it works just fine.

For the sake of testing take a 10000 frames animation and ask HD instance to source your trillion elves by a step of one. Chances are it will have to internally create those masters and you'll see your memory footprint going up by several thousands times.