Hi everyone,
I trying to obtain the same detail like on the volcano demonstration of houdini 12 but no succes. Anybody have an idea how to obtain this king of detail into the smoke ?
This detail appear if I subdivide the resolution ? the shader used is it the bilowysmoke ? I think I miss something

thx a lot.

I'm interested as well. Actually would be nice to see some serious examples bundled with software, because at the moment i'm just trying to start from another end, learning how all those fields/advections/math/etc works, instead of trying artistic side of H12.

Question to devs: is it possible to post some serious scenes, for example aforementioned volcano source scene? I mean, i don't see why not.

You right Mudvin, that's will be nice…

I believe the dense smoke cloud was generated using the Pyro solver + some shape operators.

The most important parameters to play with to get a dense pyroclastic cloud are:

Temperature Diffusion (relates to the blurring of temperature over time)
Cooling Rate (How fast the temperature in a simulation disappears)
Buyancy lift (The rate of which the smoke rises based on temperature present.)

As you can see they are all related to temperature. The more temperature in the beginning, the faster it will rise at the bottom. Making it cool down relatively fast makes the smoke slow down as it rises. Keeping the diffusion low makes the actual rolling effect more present, as there will be less of a uniform temperature volume present.

To get the speed right, without simming for hours, let the sim speed be normal in the beginning, lowering the timescale when you've got a decent volume. The timescale parameter is very useful for slowing down or speeding up simulations.

You could use combustion to generate smoke, or start by simply emitting density. By creating some nice base flames the initial smoke distribution will probably look better. Including some nice velocity to pump into the cloud.

Use collision objects!

Create nice initial velocity sources!

Disturbance adds a lot of detail. It's there to simulate small external forces like shifts in temperature. These will break up the cloud. Shredding will try to push parts apart based on the temperature…

Sharpening helps refine detail.

Maybe up-res your sim to play with the speed, look, or source…

Cheers Coen

Nice explanation, thank you very much for this one.
I'm also digging pyro masterclasses from this site, but still example scenes would be nice. Much easier to tweak ready scene than building your own from zero.

For example i'm still cannot understand how to achieve that level of detail like in that volcano image - is it very dense grid simulation running for a week or distributed on ferm, or it's some upres, and what is the proper way to do upres, and does all those shelf tools actually will work if i will introduce some custom changes into AutoDOPNetworks, and lots more questions.

I understand what you're aiming at. And I can't give you that file because I didn't make it. I believe Scott at sidefx did, but I wasn't at sidefx when they made the example.

The new pyro solver doesn't need to run for a week to get that sim out. It's not clustered as well. Up ressing isn't necessary for most cases. It used to be pre H12 but you can get to a higher resolution using the default solver.

I am planning on doing a sort of tutorial showing H12 pyro soon. Including example files and renders. But I have to find the time to start.

In the mean time, I've recently done a masterclass in Seoul and posted the material online. It's rather in depth and shows you how volumes in Houdini really work. From a sop and dop perspective.

http://www.apileofgrains.nl/category/tutorials/ [apileofgrains.nl]

Maybe it will be of some help!

And I agree, it would be great if those files could be posted online. I remember working on the collapsing bridge when incorporating the clustering algorithm, those files should be around!

Regarding the DOP network customizations, shelf tools should still work. They don't store the internal state of the network but iterate over what is present. based on what is found the most optimal choice is made. Most likely this will be correct!

Cheers Coen

Mudvin
Nice explanation, thank you very much for this one.
I'm also digging pyro masterclasses from this site, but still example scenes would be nice. Much easier to tweak ready scene than building your own from zero.

For example i'm still cannot understand how to achieve that level of detail like in that volcano image - is it very dense grid simulation running for a week or distributed on ferm, or it's some upres, and what is the proper way to do upres, and does all those shelf tools actually will work if i will introduce some custom changes into AutoDOPNetworks, and lots more questions.

Mudvin,

For the volcano simulation, the final simulation was quite high resolution which accounts for alot of the detail you are seeing. I believe the final simulation time for 400 frames was around 8 hours. Also, Coen is correct in saying that most of the detail/shape is coming from the shaping tools. Additionally, in order to get the correct ‘column’ of smoke for a volcano, you need to control the temperature field so that most of the heat stays inside the smoke ( low temperature diffusion ). Also, in this case the sim wasn't retimed since I wanted the cloud to develop slowly and build naturally.

We plan on releasing most of the demonstrations as tutorials along with .hip files in the near future.

Coen, you rock (once again)

thanks for sharing that.

thanks Coen! That's a huge package!

Glad to hear that, thanks for info!

I applied the volcano shelf tool to a default sphere. I set the shader to be Basic Smoke, which defaults to a 50 density/shadow density, which gives very dense smoke. I then rendered with PBR and one environment light set to the DOSCH_SKIESV2. Sampling on the output driver was set to 15x15, volume step size 0.025 (to match the finest divisions)

I then simulated using division sizes 0.025, 0.05, and 0.1. (The default from the shelf tool, for comparisons, is 0.15)

The detail you see in the volcano simulation is all voxel detail. So if you don't have the voxels, you can't have the detail. In the attached the coarser sims you can clearly see the voxels.

There are two shaping tools most affecting the volcano shelf tool. The first is disturbance that is jiggling the velocity field on a per-voxel level. Being per-voxel, however, as you decrease your grid size (so increase your voxel count) you are going to make this jitter finer scale which changes the look. Note the high res doesn't spread as much as the low res.

The second shaping tool is local sharpening. This keeps the boundary of the smoke from smearing out at all, even when slow motion is being performed. But it will also accent individual voxels, possibly giving a more checker or block pattern than desired.

You will also notice the default tool runs at full speed for 100 frames before suddenly slowing down. This is because to get a good sense of scale for a volcano it has to move ssllloowwwllly. But if you tried to sim from frame one at that speed you will be waiting all day. So this is a sort of pre-roll.

Finally, the voxel counts at frame 141 pictured below are 67.8 MVoxel, 10.7MVoxel, and 2.1MVoxel. A glance at top on the machine simming the fineset resolution shows 5Gb of RAM used at frame 208. (Peak may be higher as it goes up and down during the sim)

*meditating on picture*

Btw, that's just a thought, but i think such an explanations/guides should exist in documentation. Learning by example is N1 source of knowledge, at least for those like me.

Thank you very much Coen, Very nice learning material!!!

Volcano preset

Thanks jlait and Coen.

Sim details:

division sizes - 0.02
voxel counts (frame 200) - 160mil
simtime - 5h 40min
rendertime - 22h 30min

LoL, now why on earth do my tests look nothing like that coupled with nothing but crashing. Using 2 x 6 core 2.4 xeons and 48 gigs of ram.

Rob

Hi circusmonkey,

My steps:

- add sphere (shelf default)
- add volcano pyro preset
- change division to 0.02
- sim 200frames
- for rendering - jlait workflow (PBR was 9x9)

Yup that's what Ive been pretty much doing > Ive even optimized the step size etc. I get ok stills like this 1080i
division size of 0.06
200 frames sim in 32 mins
Frame 144 8.34 mins render 1080i
40million odd voxels from the default volumes

Its when I drop lower division sizes that I get a lot more crashing. On a side note is there a expression to give the sum voxel count ? .

Rob

talking about detail, I think it be great if users could share some of their experience in rendering lots of smoke and how to optimize render times.
(because mantra can make things awfully slow even though everybody will tell you micropolygon rendering is swell when it comes to rendering volumes)

here's what I usually do:

- try to avoid billowy smoke shader, that guy can be a huge slowdown, rather build your own shader or use the basic smoke
- blackman as a volume filter
- try to lower filterwidth if possible
- lower opacity limit to a point where you get no artifacts or flickering
- stochastik transparency on with a small number of transparent samples
- try to keep volume stepsize as high as possible (at some point you have to bring it down for final renderings or make compromises which I hate)
- use depth map shadows in your lights (try to have only one light casting shadows) with a resolution as low as possible (1024x1024 and 5x5 pixel samples is what I mostly go for to avoid flickering)

anybody wants to add settings on top of that list?
anybody using PBR ?

let's hear some opinions,

cheers,

Manu

For PBR tips

http://www.sidefx.com/index.php?option=com_forum&Itemid=172&page=viewtopic&t=21548 [sidefx.com]

Rob

so what's the advantage of rendering volumes with PBR ?
I just read in the docs that stochastic transparency is ignored when you use the micropolygon renderer for example …