h's blog

New page: a useful wrangle for when you need to render lots of tiny glowing particles, sparks, fireworks, UI stuff, star-fields etc.

Helping the renderer with tiny particles

Thought I’d better try writing this up as it took me bloody ages to get my head round it.

Situation: you have an object that you want lots of instances of, all with different looks. Say, one tree object, but you want lots of instances with variations in their colouring.

The approach differs depending on whether your tree object is in the scene file somewhere (ie instanced using s@instance=”/obj/my_tree”) or if it’s a file on disk (s@instancefile=”$HIP/geo_trees/my_tree.rs”).

Instancing a scene object

You can use either point attributes or stylesheets to poke new values into the material parameters: see https://docs.redshift3d.com/pages/viewpage.action?pageId=12419786&product=houdini for guidance.

Instancing a proxy file from disk

This is slightly more tricky: you must use stylesheets, and you can only poke new values into parameters that are exposed at the material’s top level VOP node. I’ve created a hip file to illustrate:

Download: RS per-instance colours_001.hiplc

To test it, just make sure you “render” the Proxy ROP first, to create an .rs proxy file for the scene to use. It’ll create an RS proxy containing this single plane with some nasty coouring:

Then you can go to Render view and render an image.

You should see this lurid mess:

But that means everything is working. Every plane has a different hue shift going on, even though they’re all identical instances. If it wasn’t working, you’d see this:

This is how it works: to vary the material, I’ve stuck an RS Color Correct node in the material, and an RS Multiply node to multiply an incoming 0-1.0 value by 360 (turns out the hue shift parameter on a colour correct node wants a value in degrees, which caught me out for a while):

and exposed the Hue Shift parameter—promoted it to the top level VOP—so it now appears as a parameter on the material itself:

That parameter name is important—hue, in this case—as that’s the thing we can bind and alter on an instance-by-instance basis using a stylesheet override.

So—into the stylesheets: to get to them, open a Data Tree pane, and choose “Material Style Sheets” from the dropdown. I’ve added a Style Sheet Parameter, and a Style, a target (“Point Instances”) and two overrides:

The first override sets the material to the one that’s in the scene (as opposed to any that may be saved in the proxy file itself). You can only override materials present in your scene, not ones within the proxy (as far as I can tell).

The second override is an “Override Script” – even though it’s not really a script: you can choose “Material Parameter” and “Attribute Binding” from the dropdowns. The Override Name is the parameter that’s exposed on the material—”hue”—and the Override Value is the point attribute you want to bind it to. In this case, my Instance object has a bunch of points to instance to, each with an @hue_shift attribute.

And that’s it. No reason to only use this for shifting hues: you could create a material with a shader switch, or a texture, and as long as you expose the switch parameter or the texture filename string in the top VOP level of the material, you can poke new values in at render time like this.

One caveat/limitation: Redshift is blazingly fast at rendering loads of instances. But if you try tweaking 1000s of instance’s materials like this, you may find that the pre-render processing, as Redshift builds the scene, could become quite time-consuming. I’m guessing that behind the scenes Redshift is creating a new shader for each tweaked material. So for huge scenes with loads of instances, you may need to take a different approach to adding variety. Stay tuned.

The node editor snaps nodes to an invisible grid, by default, to help you keep things tidy.

But enabling the grid lines (Node editor -> Tools -> Show Grid / Grid Lines makes the grid show up. For some reason it makes the node editor feel much tidier to me, and much more ordered. YMMV.

Should be called “Point Looker” really. Super simple really, just a time-saver if you keep need certain looks, which I found I did. Plug some points in the top and this will create Alpha / Cd / pscale attributes. Set colour from a ramp, or from a base colour and then hue/sat/val variance; Alpha over age or from a custom attribute; Size based on mass, or a ramp based distribution.

And a “Create specials” option, which takes some proportion of your particles (best not too many) and gives you some extra controls. Handy for making a few odd differently coloured or extra bright particles to spice up the mix, like in the top image.

Colours can be cycled through the ramps too, to add a bit of life.

No guarantees, may not function as advertised, may burn down your house. But it’s reasonably well documented if you dive inside, and most of the parameters have hover-text help. If you’re happy with that, download: com.howiem__h_particle_look__1.8.hdalc

I use AE for comping stuff, so once I’ve rendered something out of Houdini, I usually need the camera animation data in AE to line flares etc up. This Python script, stuck on a button in Houdini, will copy the animation to the clipboard in a format you can paste onto a Canera in AE. Almost. There’s one additional step: paste the data into something else first—any old plain text editor—and copy it from there into AE. Dunno why, but it makes it work…

 

You can control a POP Source’s birth rate using an emission attribute, but it only works with primitive attributes, not points. Bit of a pain. Often I just want to birth from a single point – but I want to leave the door open to adding more sources later, so using the POP Location node is a bit limiting.

Turns out the POP Source node is actually not too tricky to understand – unlock it and dive in. You’ll find a Solver SOP: dive in there, and you’ll see how particles are birthed. Awww.

The wrangle you need to tit about with is here:

It’s called attribwrangle1, and it’s just above the “random_points” null.

Here’s its existing code:

#include <voptype.h>
 #include <voplib.h>

int npts = ch("npts");
 int seed = ch("seed");
 float frame = ch("frame");
 int firstpointcount = ch("firstpointcount");
 int firstprimcount = ch("firstprimcount");

for(int i = 0; i < npts; ++i)
 {
  float r = rand(vop_floattovec((float)i, frame, (float)seed));
  int sourcept = (int)(r * firstpointcount);
  sourcept = clamp(sourcept, 0, firstpointcount-1);
  int newpt = addpoint(geoself(), sourcept);
  setpointattrib(geoself(), "sourceptnum", newpt, sourcept);
 }

 for(int i = 0; i < firstprimcount; ++i)
   removeprim(geoself(), i, 0);
 for(int i = 0; i < firstpointcount; ++i)
   removepoint(geoself(), i);

It works by taking the total number of points that are supposed to be birthed this frame and spreading it randomly across the source points. But we want to go through all the source points, read the @birth attrib, and birth that many particles on the point. So replace the bold bit with this new code:

for(int sourcept = 0; sourcept < firstpointcount; ++sourcept)
{ 
 for (int j = 0; j < point(geoself(), "birth", sourcept); j++) {
 int newpt = addpoint(geoself(), sourcept);
 setpointattrib(geoself(), "sourceptnum", newpt, sourcept);
 }
}

To test it:

• Set the emission type to “Points”
• Ensure there’s a point attribute called i@birth on the points being fed into the POPNet
• Now, every frame, each point will birth @birth number of particles. Ta-da. You can drive the @birth attribute any way you like – even just using noise makes for some sexy effects.

Note that the Constant and Impulse birthrates are now ignored. For tidyness, you ought to do some more tidying up to the node, and perhaps create a new HDA – but this’ll do for now.

A super-quick test (and hey, gotta love the new Attribute Noise SOP – saves having to build a VOP net every time):

That attribute wrangle above the popnet just says

i@birth = int(10.0 * @Cd.r);

Animating the birth rate on a per-point basis is now super easy. And yep – you can do something similar-ish using primitive / surface emission attributes, but this gives you a whole world of different looks, and, for me at least, much easier control. I love point-based emission. Fireworks, here I come

Sometimes you need each instance of an HDA to have a random seed for internal use. Sometimes you just can’t be arsed to do it yourself. This’ll do it – create a seed parameter in the UI, and then stick this in the onCreated handler to set it automatically on creation:

[code language=”python”]node = kwargs[‘node’]
seed = 6173*node.sessionId()+4139;
node.parm(‘seed’).set(seed % 10000)[/code]

From Leaf on the SideFX forums, who adds: “Explicitly setting the seed parameter on node create is the safest approach. Basing the seed on the node name can quickly cause problems when you go to clean up your scene and re-name everything (Toootally never accidentally done that before :)”