[superbad]

I've been working for a while on coming up with a room setting that I can use to replace our product photography (which costs a couple thousand dollars every time we take a picture of something). I think I finally got there. This is SL16.39, which took 14 hours on a MacPro. I think that noise on the walls is down to the bump map I used there, but I'm letting it go to SL18 to check. Only post is a tiny level adjustment. I think the curtain noise is from the SSS, which I'm not sure is worth the trouble. Either way, it looks like I'm going to have to buy a couple Clovertowns to drop into this thing.

This image (actually an almost identical version with different furniture that I can't show) is being used today in a sales presentation at one of the 10 biggest US furniture retailers.

[Maxer]

Nice image, very realistic.

In my experience interiors don't really start to clear up until you reach SL16 or higher. The problem is that to get to higher SL's you have to wait a very very long time. I render interiors out over night at resolutions similar to yours in less than 16 hours with 10 dual core machines and I never reach SL levels above 16.

[superbad]

It's saying it will be another 11h (for a total of about 26h) to get to SL18. I'll post that image too. If going to 8 cores cut that time roughly in half, I'd be down to an overnight render, which would be fine for something like this. Of course, I need to at least double the resolution (probably more) to use this in our printed tear sheets, so I'd be right back to multi-day renders.

[michaelplogue]

Very nice setup. I don't know if this will help, but if you use a very, very fine clipmap instead of SSS on the curtains, you should get a similar effect, and hopefully reduce your noise.

[superbad]

I tried the clipmap method first, but it kept coming out too "meshy". The mullions in the glass were leaving sharp shadow outlines on the curtains, which they wouldn't do in real life. I also tried making the curtains out of dielectric material with a lot of attenuation and high roughness, and that worked pretty well. I'll probably give that a go again.

[jurX]

...the woodmaterial is still nice

[michaelplogue]

I tried the clipmap method first, but it kept coming out too "meshy". The mullions in the glass were leaving sharp shadow outlines on the curtains, which they wouldn't do in real life. I also tried making the curtains out of dielectric material with a lot of attenuation and high roughness, and that worked pretty well. I'll probably give that a go again.

Maybe you could try a rough AGS material.....

To get the SSS on the above image - Did you have to give the curtains some thickness in order for it to work?

[superbad]

Maybe you could try a rough AGS material.....

To get the SSS on the above image - Did you have to give the curtains some thickness in order for it to work?

I'll try the AGS, didn't think of that. The curtains have zero thickness- I wasn't expecting the SSS to work, but it does. The absorption parameter doesn't seem to do anything though (probably because of the lack of thickness).

Also, when you make it a plain dielectric, the attenuation parameter has a really small range where it doesn't do anything, then suddenly kicks in (it has to be really small- 109 um). I would have thought zero thickness glass wouldn't attenuate at all, but it almost looks like Maxwell gives it some nominal thickness internally.

Looking at it now after several more hours, the only remaining noise is on the walls and a little bit on the bottom of the curtains. I'm thinking the bump map on those materials could be causing that.

[superbad]

Strange, I don't know what happened to the post, but there was a very good tip to try rendering at a much higher res and resample. So this is the same scene rendered at 3500x2800 and resampled down. Only this one got to SL10.36 in just under 10 hours- half the time of the original. The only substantive difference I see is that the light coming in through the door and windows isn't as intense. Those areas get brighter as the SL increases. Apart from that, the noise and sharpness are actually better on this one. Now I just need to buy some more RAM so I can leave ML turned on. Thanks for the tip.

[Frances]

Strange, I don't know what happened to the post, but there was a very good tip to try rendering at a much higher res and resample. ...

That was me.

I got worried that it wouldn't work with Maxwell, which of course it does because it's a general concept. And I couldn't find the thread I referenced where Mihai explained it. I'm glad it worked. It looks really good.

[Hervé]

hey this is a really good render... very realistic to me.. superbe wood mapping.. and nice carpet.

[superbad]

Yeah, I forgot to give give credit- the carpet and its normal map come from the excellent Insideko DVD.

[Hervé]

ahh that's ok, you don't have too..., I am impressed with your furniture.. really nice.. !

[w i l l]

Yeah that is some quality wood.

[michaelplogue]

As an experiment, I did a couple of test renders - just to see if there is any difference in render times and quality when comparing the use of SSS versus a fine clipmap. I was very surprised that both types reached a set SL at almost the exact same time and Benchmark. And even though I didn't get the transmittance levels on the SSS to match that of the clipmap version, I have to say that to my untrained eye, the noise level is pretty much the same as well. On both of them, I set the ND value to 2 in order to help blur out the shapes behind them. I ran both at ND 1 as well (which neither looked good) and got pretty much the same benchmarks and times. I used a large enclosed room with windows on all four walls to make it a more difficult test for the physical sky and sun to clean up.

I've got another one rendering right now with a single BSDF layer, 100 roughness, neutral grey transmittance (128), and ND 2. The lit portions are really picking up the blue from the sky though....


Both of the below images were rendered to roughly SL 15.85 and 7h24m

SSS version




Clipmap version