first post here - so let me quickly introduce.
My Name is Lasse, i am heading a small studio in Berlin, Germany where we produce mainly architectural visualisation along with a lot of other 3d stuff.
Well - for 3d work we are normally based on a 3dsmax - vray and corona - pipeline.
There has been a lot of upshake lately when the provider of our main software, autodesk, announced
radical changes to licensing policies and cost.
This (but not only this) for many collegues from our field it became became a reason to “now really” look out for a different solution.
Long story short: after frequently looking into Blender I recently spent a bit more time with it.
What can i say: Love it.
All my anxieties regarding the UI etc just vanished. Blender is awesome
I did a little test with an interior scene first in blender with corona renderer and then with cycles.
Well - I posted the results of my little study on the corona forums and someone said i should share it as well in the blender-world. So now I do.
Also wrote little blog-article on it:
some pictures of the series in combo Blender With Corona Renderer
I am a new blender-user, coming from 3dsmax.
Tried my hand with it in combination with corona, later re-done the scene again with cycles.
see pictures with blender / corona first
and I also tried out the cycles.
I have learned that branched patchtracing can help to get the grain out, well i tried bruteforce method, 5k pixel with 10000 samples.
I feel that Corona is a little more contrasty and Cycles a little more dull (e.g. glasses) and desaturated.
It would be nice if you could give us some insights of your scene (Lighting, set-up, sampling, machine specs, render-times).
I’ve tested Corona on alpha stage for a couple of hours only since they don’t have a linux version and I felt that it was at the time the easiest render engine out there, super fast, with great results , not to mention that it runs in CPU.
I was so impressed that I could even buy it if there was a linux version.
What’s your opinion?
For some reason the images are sharpened to death by this forum software
Behind the link to the blogpost http://xoio-air.de/2017/blender-corona-renderer/ you see them in their original state, also a bit bigger.
Well, rendertimes for the corona pictures range between 6-8 hrs in 5000x3750 resolution. On my Workstation with 6900k that is, 20% longer on 6-core I7.
Cycles the same resolution, same time - but GPU. i have put 2x1070 gtx cards into my rig for this “test”, same rendertime for 10000 samples. I have set the bucket-size to 256x256 because I have read that is the size GPU performs best with cycles.
Unfortunately the cycles renders have some grain that does not go away (after 50000 samples maybe, means 2 days of rendering… bit too long) so I left them like that.
I have learned in the meantime that with branched PT i could get better/faster results. problem is my scene: light comes only through the very small windows. most of the grain is “diffuse” or GI grain - there is very few reflective noise.
I guess biggest problem is that cycles has no lightcaching - but someone also pointed me to this technique, which is what is basically missing: http://real3d.fr/how-to-render-interior-a-lot-faster-with-cycles/
vray would call that lightcache, in corona it is called UHD cache i think. with this techniques it would be very more efficient
nevertheless faster, but also with much less noise. but 8 hrs is OK for that resolution.
As you ask for the light:
It is “just” white sky/backgroudn with a sunlamp with a slightly orange color. I put portals in front of the winfows.
Sorry that i have less descritions in OP, I had before. but for some reason i deleted it accidentally.
opinion on corona:
I love it. I started using it also in the early days after working many years with vray. still using vray though - each have theirs strengths.
But yes, it is very very good software and also the team behind this are very nice guys. it is a good thing! The ease of use was revolutionary those days and everyone in my studio was totally amazed when i introduced corona. finally you could work on the picture more concentrated. more art, less tech
The truth is that you could never get read of noise in Cycles but the de-noiser is in the way for 2.79.
I’ve been testing de-noising builds since day one and the latest build seem quite promising (though don’t expect miracles).
I think that the only real solution would be some kind of adaptiveness in Cylces so samples are concentrated where most needed.
One more thing I suppose you are using Filmic in your CM.
no, i didn’T use filmic (i think, hehe). Sorry for unprecise answer, but i think i de-installed the filmic again. i found it really hard to deal with, somewhat unpredictable
I pushed gamma to 2.2 and added some curves in post to get the depths in again. i know it is a bit wonky - but that’s what you would do in vray together with srgb. I’ve heard later that it is also the small dynamic range in blender is a “problem” or the reason for the cycles behaving a bit weird => but with the gamma push you get similar result like a reinhard/exponential cm when getting down the exposure and anhancing the darks.
i know - not perfect: and also maybe the reason why i had a hard time to actually tweak shaders. they look pretty dark in the preview because of my gamma stunt
yes denoiser sounds good: but i am a bit affraid of using them. the corona denoiser is one of the best around and even that is producing artifacts when used on too noisy image.
i think best solution would be to get some lightcaching, and yes: adaptivity
but it’S a cool engine, and obviously my setup is very challenging for it (but it is on purpose. this kind of scene i always use to find out weaknesses of renderers. small windows, low angle sun. hehe)
I seriously wonder why no one so far have ever attempted about developing it in cycles.
I’ve seen bidPt, adaptive sampling, denoising attempts and other ideas, but no one think about caching.
Not a fund rising, not even a GSOC proposal so to be payed a bit.
Too much obsession with the “unbiased” philosophy I guess.
Or just that the interests of developers doesn’t match with the interest of audience on it.
It’s like in filmic stuff… you just say meh, it’s not worth until you see someone who posts results and buzzes the word.
The only hard to deal with thing is to change your mind to work with Scene Referred Data (the raw output of cycles).
If you try to “clone” a technique used in VRay’s display referred data it’s obvious that you’ll end up with unexpected results.
yep, maybe that’s why i was a bit confused
in meantime i tried it and funnily the output looks similar, it is just that the material preview looks much more like it would in end-result.
hm: maybe go tat way next time. Think it was just in the middle of the process everything looked very screwed up, so i switched to a technique/workflow i am familiar with.
So and what do you do then with that “data”? you convert it to display refered one, no?
Filmic makes the transform and gives you a view.
You just have to change your workflow so to not work with “classic” operations that assume as input the range of values 0-1.
Also apply a gamma correction is wrong cause you’re working on linear.
It’s not that hard, it’s actually almost a “magic button”, but you have to trash out or rethink many of the techniques used in the past.
Can you see for example how much data is lost in your images where you have burned out data and at the same time dark areas? With filmic this thing would have been “fixed”.
Also it provides a set of “starting” contrast look with constant 0.18 set as middle gray, and a false colour look useful to measure and easily spot middle gray points to get a desired exposure.
I suggest you to read and ask there, where Troy Sobotka (author of filmic_blender configuration) uses to explain.
And some things also in this big thread but it’s kinda full of other things so… good read.
So, reading your blog, the gap beetween Cycles and Corona, is going to be filled by Filmic and the upcoming denoiser.
The cool thing is that the denoiser algorithm can be somehow partially reused to get adaptive sampling (at least this is what Lukas explained sometimes ago)
sorry for bump:
also, because it is integrated into corona, it is possible to render these 360°Vr spheres. check the walkthrough i made from the scene here: http://bit.ly/2nAOThv
yes, doorhandles suck - but this is a test. imagine them painted white
Nice tests!
I like the lighting and feel in both versions but the glass material seems a bit basic in Cycles renders :D.
I am also trying to switch from 3ds max/Vray to Blender/Cycles and I can agree that its quite fun to learn something new and different. I chose to stick with Cycles because of the seamless integration. Render time for final image is high but scene development / test rendering is fast and enjoyable. I wonder how does the Corona exporter perform in complex scenes. Is the test rendering smooth enough for fast feedback?
Thank you for putting the weight of your studio’s reputation behind Blender with this article. I hope the decision by Autodesk (et. al.) to move away from perpetual licenses becomes a catalyst for broader adoption of Blender and other well deserving open-source tools.
