Path Guiding: Show the difference between path guiding enabled and disabled. We’re especially looking for scenes that were difficult to render previously (like a light shining through a door crack into a room). Please note that while Path Guiding can help with caustics, this isn’t its primary use case.
Sampling improvements (especially with volumes and long area lights)
Attribute node: support for using attributes of geometry nodes instances.
You can use your own scenes or take a look at the demo files for inspiration or use them as a basis:
Please upload your material here into this thread or share a link to an online storage.
I know you wanted to avoid caustic scenes, so I had a think about the main area Path Guiding benefits render. Indirect lighting through diffuse bounces. And one area that occurs a lot is interior scenes lit up primarily by bounce lighting. And so I tried creating a scene to test that.
For reference, the hardware used was a Ryzen 9 5950X. And for path guiding I used 128 training samples. This is also with Open Path Guiding Library 0.3.1.
Also, all assets and textures are either downloaded from AmbientCG, Polyhaven, NASA, or created by myself. Both AmbientCG and Polyhaven license their assets under CC0. I’m not sure about NASA.
I also had a question. In what format do you want these release note images for Path Guiding to be?
Comparison of quality for equal render times? (E.G. This is what a render looks like with path guiding off and on if let to render for 1 minute)
Or comparison of render times for equal quality? (E.G. Both of these images are “presentation quality”. But the path guiding render took 5 minutes while the render without path guiding took 15 minutes.)
Hi @Alaska,
thank you for these images! I think both meassurements are interesting but in the end it doesn’t really matter, since both methods show a clear Path Guiding advantage.
I think equal render time is best for comparison and easiest to setup. Quality is not really well defined, especially when different parts of the image have different noise levels.
You can also recreate this “effect” by repurposing the “Barbershop” scene from the Blender demo files.
Here’s a render with 2 minutes each. They’re still really noisy. Light portals, removing glass from the windows, and longer render times would make these images more presentable.
I want to point out how scenes look really good with reflective caustics when using PG. If Alaska puts some mirror or shiny objects in the scene it will make huge difference.
this is very simple test scene, not proposing this, just to show the difference.
PG / 1500 samples
I am hesitating a little bit putting too much caustics in the example image.
That it helps to solve SOME caustics is a nice side-effect of path guiding but not its main intention.
I think we should avoid adverting it ‘a caustic solver’.
If you look a the Cycles_path_guiding tests - Blender Tests - Blender Artists Community thread people are already only looking at caustics when testing PG, even if I tell them PG is not meant for it :D.
@ThomasDinges and @brecht would it also be possible to add two images for the PG feature. One like the barbershop and one with some simple caustics like the ones shown by @Caner-Aslan with an explicit note, PG can help with ‘simple’ caustics but it is not a caustic solver?
hmm I see, I also can’t stop myself luring into reflective caustics. It is very charming. don’t blame us
Maybe a secondary smaller image to say here this is the sideeffect.
@Caner-Aslan don’t worry, I am not blaming you at all. I know how it is with new toys.
I just want to avoid to put the expectations of the users to high when they read the release notes and see the images
Straight glass shader, roughness = 0, 150 samples, filter glossy = 0.1 (for non path guided so you would at least get some colour in the shadow area, 0.01 for path guided)
It probably would be better to demonstrate how these improvements help in the context of an arch-vis scene (as opposed to a demo image of a glass object above a grey floor).
I would think a key bit of reasoning here is to show the universal nature of OpenPGL helping with everyday creativity, an algorithm far from being something for whiz-bang demonstrations.
It does that - it shows that with PG on you can use a straight glass shader instead of the ‘architectural glass’ workaround. Yeh it might not be as sexy as a full arch-vis scene, but it demonstrates a key feature of path guiding quite dramatically (actual transparent glass) - sometimes less is more.
Also - what’s the cottage scene if not arch-vis
Both of these scenes show off quite different advantages that path guiding brings. In the case of the cottage scene, it’s finding hidden light sources that need to bounce around corners.
Whilst not to diminish the value in lower noise levels (particularly when coupled with denoising), in many of the arch-vis scenes posted, all it looks like PG is contributing is lower noise levels (particularly to a layman), but PG brings many more advantages that these scenes don’t explicitly highlight or that get lost in the overall composition without pixel peeping.
It would be nice to include a few example renders that pull out and isolate very specific scenarios to really emphasise just what PG is bringing to the table and what new possibilities it opens up.
IMO the ability to render clear transparent glass using just a simple glass shader setup is a really big thing. I have lost count of the number of threads that get started over at Blenderartists by newbies asking “why are my glass shadows dark” or “why is my render so noisy” (when they are trying to light through glass). The answer is of course “use an architectural glass or light path trick”, but now PG adds a new option to solve this particular problem.
