You have a scene that you're just about to start rendering in Redshift/Mantra/Arnold, but you're keen to give Karma a try? Don't wanna learn all that prim/reference/sublayer/delegate stuff all the USD nerds talk about?
Hopefully this will get you started. USD jargon is unavoidiable at some point, but I think its possible to dive in without getting too caught in the jargon weeds. The aim here is to take a /obj, /mat, /rop setup, and port it to Lops as painlessly as possible, while picking up some basic USD concepts on the way.
This whole thing should be 20 mins at the max. Lets gooooo!
What you know vs the end target
If you've rendered anything in Houdini before, you're probably used to a setup like this:
Familiar sure, but if you think about it, doesn't really play to any of the strengths of houdini's node network way of thinking. One of the goals of Lops is to unify this, so that you can combine everything into a single network:
Neat right? "Whatever," you say, "It's the same nodes in a different skin, big deal". But think about the sops tricks you're used to, and how you can reapply them in this context. Eg being able to have multiple cameras, and easily change between them with a switch node:
Or create a copy of your materials, change all their properties, and swap them out instantly:
Or add extra lights that will only be used by 2 'rops', and ignored by the other:
A network view of those operations is easy to understand and edit, while it's possible to do those things in /obj /mat/ /rop, its much harder to edit and understand.
Here's a hip with some rbd, ground, volume, lights camera, we'll port this to lops.
Here's what we'll do, playing to the strengths of Lops to start with, and talk about other options later:
- Import geo from sops
- Create lops lights
- Create and assign lops materials
- Create camera
- Create a render node
You can either jump to the /stage context, or do what I and the other sensible people do, create a lopnet (people who use the top level contexts are savages).
A sopimport node is like an object merge, it grabs geo from sops, and pulls it over to lops.
Create one, and set the path to the pig. Tada, your pig is in lops.
The scene graph tree
The one bit of Lops jargon you HAVE to understand is the Scene Graph Tree. Super quick mini tutoral follows, sorry if this is dry, we'll get back to our foggy pig in a sec.
Jump back to /obj, and hit shift-w in the network view. You can collapse it down to look like this:
You can see that each node in /obj, /mat, /rop has a matching single entry in the tree view. One node = one tree location.
Houdini gives no tools out of the box to procedurally create or modify nodes at this level. You can't copy-to-points a bunch of cameras. You can't pack a bunch of lights down into a single node. You can't procedurally create and destroy materials. (Ok yes TD's can with python, but thats another story).
In other words, the shift-w tree view is just a different way to display the network.
Now look at this, I've got a single file lop, and I'm viewing the Lops Scene Graph Tree:
A single lops node has created an entire structure of objects and materials!
But that's not all. A lops node could remove one thing, or many things from the scene graph. Or rename one or many things. Or reparent one or many things, or transform one or many things.
Unlike /obj and the shift-w tree, Lop nodes do NOT have a 1:1 correspondance to the Scene Graph Tree.
This seems unintuitive, but don't stress; you already have a similar analogy in your brain with sops. You know that 1 sop node doesn't mean 1 piece of geometry, but each sop node could create a single point, OR hundreds of polys, OR modify polygons OR delete them.
Sop nodes modify geometry, you keep an eye on the geometry spreadsheet to see what the 'state' of your geometry is, the viewport shows you what it looks like.
Lops nodes modify hierachy and objects, you keep an eye on the scene graph tree to see the 'state' of your objects, the viewport shows you what it looks like.
As such, you need to see your Scene Graph Tree at all times, so lets do that. Jump back into your lopnet, and change to the 'Solaris' desktop. You should see the Scene Graph Tree in the lower right. You're gonna get very friendly with this panel.
See how the node appears in the scene graph tree as 'sopcreate1'? It's called this beacuse in the sopcreate parameters, it uses $OS to define its own name. Lets be lazy here and rename the sopcreate node in the node network to 'pig', so that the scene graph is correct.
Append another sopcreate, and use its parameters to bring in the ground object. Note that when you attach the two nodes together and display the second, you see both the grid and the pig? Connecting sopcreate nodes appends their results, as opposed to sops which does chains of geometry modification, or connecting /obj nodes which does parenting operations.
Repeat the operation for the fog, you should now have all your geometry in lops and a nicely named scene graph:
- Append a light lop
- Use the transform parameters to move the light to the top left.
- Select the 'base properties' tab, set colour and exposure (I set exposure up +4)
- Can also hit enter to get its interactive handle to move in the viewport
- Append 2 more lights, do the same
- To hide the transform gizmo for the selected light, hit escape
- In the viewport menu where it says 'persp', select Karma. You're now viewing the full quality renderer in the viewport. You can move stuff, change lights, it will keep up.
- If the light icons are distracting, toggle the 'display light guides' toggle on the right stowbar of the viewport.
- You can swap back and forth between HoudiniGL and Karma whenever you need.
- Append a camera lop
- Look through it via the viewport menu ( where it says 'no cam', select and change to '/cameras/camera1')
- Click the padlock icon on the right of the viewport, this locks the camera to match the view. Select the pig in the scene graph tree, hit space-f to frame, then the usual space or alt controls to rotate and pan the camera
Did you notice that in the scene graph tree the lights are put under a lights folder, and the camera under a camera folder. Why? Tidyness and convenience. You can see in the camera parameters it's primitive path is set to /cameras/$OS, lights are similar. You can place them wherever you want in the hierarchy, sidefx suggest with these defaults that having a top level folder that collects all the lights and cameras is useful. If you don't like it, you can just change the path, eg make it /$OS so that it no longer lives in a /cameras folder.
- Append a material library
- Dive inside its a /mat context, create maaterials, or be lazy and copy them from /mat, paste here.
- Jump up, look at the scene graph tree. There's now a materials folder (same idea as before, this is nice for tidyness), inside there are the materials.
- Append a assign materials lop
- Ensure the /materials folder is expanded in the scene graph tree
- To assign the pig material to the pig geometry, drag the pig geo from the scene graph tree into the 'primitives' parm, and drag pig_mat from the materials folder into the 'material path' parm.
- click the plus icon at the top of the parameters multilister twice, assign the ground material to the ground, fog material to the fog.
Note that if you select items in the scene graph tree, they'll be highlighted in the network view. If this bothers you, click the 'display selection' button on the right of the viewport to turn it off.
- Ensure you have karma enabled in the viewport (click the first menu where it says 'persp', choose 'Karma')
- Hit d in the viewport to bring up display properties, swap from CPU to XPU
- After a short delay (data has to be buffered to the GPU), XPU will kick in, offering even faster interactive performance
- Not all features are supported, but its being actively worked on
- If the viewport feels like its lagging too much after 30 seconds, try restarting Karma (viewport persp menu, 'restart render')
Enjoy procedural scene building
In sops its often educational to 'walk the nodes', start from the first node, display it, and walk through each node. You can do that in lops too. Note that the scene graph tree operates like the geometry spreadsheet in that it follows node selection, not display flags.
- Select the first pig sopimport. The Scene graph just shows a single pig node.
- Display the first pig sopimport, the viewport reverts to just the pig.
- Select each node down the network, see how the scene graph gets adjusted with each node
- Do the same with by displaying each node, observe the changes in the viewport
- Try bypassing some of the nodes, see how the graph and the viewport react.
Setup karma rendering to disk (wip)
- Append a karma lop
- This acts like a render rop, and should be intuitive to understand; framerange, camera, resolution etc
- Set the framerange 1-24
- Change the output suffix from exr to png
- Change the rendering engine to XPU
- Change pixel samples down to 4
- In the lower section choose the 'Image output' tab, then the 'filters' subtab'. From here you can enable denoising, try optix to start with.
- Click 'render to disk', your machine will render those frames. Render times will be maybe 10 seconds per frame, and Houdini will be locked up while rendering. Lets fix this. Cancel the render and move to the next step.
Setup karma background rendering with husk (wip)
- Append a USD rop (NOT a USD Render rop, thats a different thing)
- Set the framerange to 1-24, write to disk, it should be very quick
- This USD file has everything required to render, like a .ifd, .rib, or .ass for other renderers. We can use tops to render this in a background process.
- Jump up to /obj, create a topnet, dive in
- Create a file pattern top, point it to the usd file you just generated.
- Append a USD Render rop
- Set 'evaluate using' to 'frame range', set the range to 1-24, and enable 'all frames in one batch.
- Select the rop, press shift-v to start tops processing. The frames will be writtn to disk, more efficently and in the background rather than locking up your session.
Ok, that's the basics covered. Where to from here?
Now is the time to tell you about the 'scene import' lop. It's meant to be a shortcut to do all the work I just had you do. :)
That said, while the amount of work the scene import lop does is admirable, it can fail on all sorts of edge cases, and the nature of Houdini and everyones individual crazy networks, its too easy to run into. Besides, if you want to start using Lops effectively, eventually you have to start building setups as I've shown here.
You might've also spotted this lurking in /out, dive inside and you'll see its a mini lopnet using a scene import lop, and all the karma lop parameters promoted to the top of its UI. So the same caveats apply; fine for simple things, will probably break if you push complex old scenes through it, and really, you're not learning.
Sop create vs sop import
There's no need to create your work in /obj and then import it. A sopcreate node lets you dive inside and just create sops directly, saving you a step, and keep your network less dive crazy.
There's lots to cover, but this was the bare minimum to get renders out. If you're interested in more, go check out the other USD and lops guides I've written on this site.