I have an idea for a sticky thread - bear with me. A lot of this information exists in little hints and clues in the wiki, but -
What [qualitative] effect do different settings have on render times?
What are the trade-offs (memory, quality)?
This may seem very amorphous, but it would be great to look at a render that’s taking inordinately too long and say “oh, if I just turn down AO bias slightly, it will make a huge difference”. I think most of us would be happiest with the bare minimum that would produce visible quality, with a reasonable rendering time, but trial-and-error takes a long time.
I think the way to discuss this might be to take discrete settings and say what our experience has been. Things like:
Octree resolution?
Different AA filters?(is Mitchell really faster than Gauss?)
Different levels of OSA (how low do you go?)
How much does command-line rendering help?
Spotlight vs Area Light vs Point Light? Settings?
A while back when I was researching a lot about Elephants Dream I stumbled onto one of their lighting and compositing workshops that shed a little bit of light on the issue of spotlights in Blender, and using them in a production environment (AKA efficient rendering times).
The video’s a bit long winded, but is very informative. All in all they talk about using spots with shadow buffer’s efficiently being that when used properly they take up very little resources. They also get into things like faking ambient occlusion with negative lights and things like that.
Downside to these approaches is that most of them are only available in the blender raytracing engine. But, perhaps that is a lesson that the blender engine is capable of producing results just as realistic as things like yafray with the right skills and techniques. Not to mention that it is more efficient under a production environment.
To avoid time-consuming ray-traced shadows, use spots with shadow buffers. The larger the size of the buffer, the better the quality of the shadows.
You can use negative lamps with the “sphere” option and quad attenuation to simulate corner and proximity shadows (or nostril shadows) and avoid using Ambient Occlusion.
Using a soft spot (with cast shadows turned off) underneath your ground plane you can simulate a bit of GI bounce off the ground plane.
Keep 'em coming!
As the page indicates, the octree resolution is something that requires a bit of experimenting to finding the right value for your render. But it gives a nice general chart of various settings that is quite useful. Furthermore, the octree resolution does not affect the Yafray renderer as I had discovered recently.
And I agree RocketShip, I would like to hear more about AA filters as well.
Here are a few tricks that served me well when I was stuck on a much-too-small machine …
(1) Break the problem down into layers. Sandwich the layers together back-to-front to produce the final image. This allows you to “tweak” the image to get exactly what you want, without starting over.
(2) Focus on what you want the final image to “look like,” then cheat your way shamelessly to get there. The greatest graphics-processor in your pipeline is always “the viewer’s eye.” You don’t have to actually model physical reality as long as you can somehow come up with a good-looking frame that looks like you did.
(3) Computationally-extravagant techniques will never beat a really good lighting setup.
(4) All the computation in the world won’t salvage a poorly-composed or poorly-lit picture, in any medium.
Can you provide any concrete suggestions on settings? The layers suggestion is a good one, but are there specific features or settings that you avoid or minimize to make things run faster?
My issue working like this has always been dealing with shadows. For instance, say I broke a scene down into 3 parts: aniamted obj1, animated obj2, and a plane that they both interact on.
Well, if I broke that down into 3 separate layers, naturally, the shadows on the plane caused from the lighting of this setup does not carry over because it is rendered separately. That leaves me with 3 alpha images I sandwich together in the nodes editor that cast no shadows. Is there a technique to work around this?
p.s. Maybe I could add an instanced obj1&2 to the plane layer, and give them shadeless properties. The drawback is that it would take a lot longer to render all 3 layers than it probably would to just render it all together, but you would have a tweakable final product in post editing.