BLENDER(2.93) | RENDER SETTINGS
Below are a couple images showing the current general render settings I’ve tested for final/production ready renders. Here are some added notes:
Square Samples:
I'm still deciding if square samples is necessary. I used to use these in a production environment - however I honestly can’t recall the benefit. I did a test with the samples at 16 with square samples turned on and another test with square samples off and the samples at 256(equivalent) and I got the same results with the same general render time. I’ll keep square samples on for now, but that may change later.
Caustics:
I left caustics on here even though I didn’t have any glass or water or anything really that would utilize caustics. I did this because in the test I did anyways it seemed to effect the light bounce a noticeable amount and the render time stayed the same anyways. I probably could have turned off caustics and just increased the light intensity slightly to up the light bounce, but since the render time didn’t change I just left caustics on. If glass is present or something that may push the caustics more the render time may go up, in which case this may need to be turned off if it isn’t needed to add to the desired look.
Extra Notes:
Aside from that I did a number of tests with the light path bounces at various values and these values seemed to provide a consistent/desired look while not drastically effecting render times in a negative way. Other then what I have setup here I may try 512 samples(equivalent) using either the built in denoiser or denoising in comp to see if faster output would be achievable since the sample count here is what’s primarily driving the render times. Also if you are using GPU Compute to render(which you should if you have a GPU), you should set your tiles to a higher pixel count(something around 64px x 64px I think should work). I’m using my CPU, hence 16px x 16px.
Update:
I’ve been denoising using the comp denoise node and with that in place 512 samples(or square sample equivalent; 22 or 23) seems to be plenty and it cuts the render time in half from what it was at 1024 samples(in my test it went down from 58 minutes to 27 minutes - with no noticeable quality change, at least on the denoised still frame).
So I just did another test at 256 samples with comp denoise and it still holds up quite well, once denoised at least(the noise is very noticeable undenoised). It doesn’t seem useable for final settings however as it’s pretty splotchy in some areas, but it’s not that bad. I’d say 256 samples with comp denoise is at least good for look dev/tests. Mind you these tests I’ve done are with untextured objects(alley scene) with just a light grey bsdf material. So 256 samples may hold up better with textured objects. But as is I’d only use it for tests/look dev stuff since it still looks good and renders much faster - again quality does noticeably go down though, at least with untextured objects so wouldn’t use for production. 512 seems to be the sweet spot right now for final working sample settings.
Originally posted: September 17, 2021
