This is meant as a help to people using Superfly or that simply want to get some more information about it.

Calling the topic "Complete guide to Superfly" is probably a bit exaggerated, but hopefully it will be helpful. :)

One of the most time consuming tasks when using Superfly is the render times, especially when dealing with noise in the images. Changing various settings in Superfly is not always logic in regards to what effect these will have. So the following will serve as examples of some of these settings and what effect they will have on your renders and also as a learning experience for myself.

I will also link to a denoising tool later that can help with noise, which is highly recommended.

The test scene that I will use is a small bathroom which have 6 small mesh lights in the ceiling, 2 small windows and 2 roof windows with 1 infinite direct light outside. This is done because limiting light in a scene seem to cause a lot of noise in PBR renders.

Example of scene.

This was rendered using the following settings:

Looking at the image, first of all it still has quite a lot of noise in it. Secondly it is fairly dark, despite there being 6 lights there. Obviously I could crank of the strength of the emission, but lets see if we can't solve it with some of the render settings.

So will start by lowering the settings to something very low, only using 15 primary samples.

Which as expected will resolve in something extremely dark.

Diffuse samples (Bounce light)

This setting seems control the indirect lighting in the scene, which means that if this is set to 0 there will be no light bouncing from one surface to another. So if we increase this to 1.

Now we can see that the light bounce starts to bounce around the room. Increasing this to 15, we can see that some of the noise goes away, but obviously also increase the render times. But still it doesn't seem like it affects the actual bounce light.

So will turn it back to 0 for now.

Also it doesn't appear to do a lot on its own, when it comes to reflections, as the mirror is completely black and there is no highlight on the sink faucet.

Glossy samples

This is what controls the reflections of light and therefore must work along side Roughness, which determine how reflective a material is. So rendering the image again with 1 Glossy sample we can see that we get all the reflective light.

With 15 samples setting.

As can be seen the quality is increased and less noisy especially in the reflections, but also seem to reduce it slightly on the wall next to sink. Another issue is the shower which can be seen in the mirror. The glass is all black, the is caused by Max transparency bounces which is currently set to 1. So will increase that to a maximum of 8 to make sure that light can travel through it and we will leave it at that as a standard setting.

Obviously we need both the diffuse and the glossy setting for all scenes, since bounce light and reflective light is almost impossible to not have. So rendering the image with both of these at 15, to see how that affects the image.

This allow us to now see the texture on the table and even the floor. Yet there is still a lot of noise and the render times a lot higher. Will reduce both of these to 5 for now, as we will need them.

Transmission samples

Changing this doesn't seem to have any effect, which most likely is because there is no such object in the scene or the effect is so small that its barely noticeable. Even though one would assume that it would somehow affect the shower glass. But it might be because the scene is to dark, so will get back to that.

Subsurface samples

Since this is how light scatters inside a material, such as skin and there is no such think in this scene. I will leave that one to 0.

AO samples

Also seem to do little, most likely because the scene is to dark, but should improve the areas where objects interact with each other. But will leave it at 0 for now.

Mesh lighting

This controls the quality of the mesh lights and since there is 6 in the scene, this should have a significant effect. I assume that setting this to 0 means that mesh lights won't be calculated in the render. So will increase this to 1.

As expected this have a big effect on the scene as these lights comes into effect. However it is still fairly dark. Increasing the samples to 15, get rid of a lot of noise, but also have a very big impact on render times (For some reason the mesh light renders much faster when through the mirror than on the walls themselves, maybe because it uses the glossy setting, im not really sure).

The quality get better, but again did little for the actual lighting in the scene, lowering this to 5 samples.

Doing so, does not seem to reduce quality a whole lot compared to 15 samples, yet the render speed is a lot faster, so tried to reduce it to 3.

Which again seem to have little to no impact on quality. So would probably keep this around 3-5 samples, Ill keep it at 3 for now.

Sample all lights direct (checkbox)

Tried to figure out what this setting does, but weren't able to find an explanation. Turning it off, does seem to reduce quality, while not really giving a lot of render speed if any (Might be because there is only 1 direct infinite light in the scene which is outside the windows). So would simply leave this on.

Sample all lights indirect (checkbox)

Does seem to increase render speed slightly at the cost of quality. However since I want quality, I would suggest leaving this on as well.

Filter glossy

This is used to blur highlights (reflective light) on objects if you think they are to strong. (Default: 1).

Value of 350.

Will leave it at the default of 1.

Clamp direct / indirect samples

Limits the amount of light these sources can contribute with in the scene. Will leave these on default.

The settings so far, have mostly dealt with the quality of the renders, some settings are needed for some of the lighting to even show up in the first place.

Diffuse - For the indirect lighting or bounce light.

Glossy - For reflections.

Mesh light - If any mesh lighting is used in the scene.

However the scene still appear fairly dark and there is still a lot of noise. Given that im only using 15 primary samples, it won't be possible to get rid of all noise, it is simply to low an amount.

But will move to the other side and look at some of these setting.

Min/Max amount of bounces

This determine how many times a light ray can bounce in the scene. So far I have used min (1) and max (2), but will try to increase the max bounce to (50).

Seems to do little for the light in the scene, despite that the light should not be able to bounce around a lot more.

If we increase the diffuse bounces to 3, but change the maximum number of bounces to 1, the mirror appears black.

If we switch them around so now we have a maximum of 3 bounces and only 1 diffuse, the mirror seems to get lighter.

So lets try to reduce all bounces (except transparency) to 0 and leave the maximum bounces on 1.

Now we have lost all light from the mirror. So increasing the Glossy bounces to 1.

Now we get it back, increasing the Glossy bounces to 5 and the maximum allow bounces to 5 as well.

Give us the reflective bounces on the table. So it seems that the maximum amount of bounces is a bit weird setting, as it control the maximum allowed bounces for the other settings, but seem to do little on its own. So will just leave this at 50 and control the bounces through the other settings. Will also set the Glossy bounces to 3 so we are sure to get the reflective bounces from the windows on the table.

Given that we now know that the diffuse control the indirect light or bounce light. Increasing this should increase the amount of light being bounces around the room. So will set this to 5.

This as expected will cause light to bounce around the room more and make it lighter, but still its fairly dark, so tried to increase it to 25. Which did nothing in regards to making the scene lighter. Which makes me wonder if the mesh lights simply ain't strong enough. So I tried increasing their strength and render again at 25 diffuse samples to see it would get even more lid.

However it doesn't seem that diffuse bounces work that they might be capped at 5 bounces as it didn't seem to make any difference, so will leave it at that.

So finally I increase the primary samples to 45 with the new settings.

Which does give a fairly clean looking image, however the render times are still very high, even longer than the original image. Which leads to the use of a denoiser.

Denoiser

The denoiser in itself doesn't reduce render times, but rather allows you to render at lower settings at the cost of details once your image have been through the denoising process. So how many details are acceptable to loose is obviously a matter of opinion.

But throwing the image above through the denoiser, we get this:

Rendering the same image but this time I have reduced the primary samples back to 15.

After denoiser.

Comparing the two images after the denoising, its clear that the one rendered at 45 samples does how more details. But the render times are also significantly higher.

The denoiser I used is an external one that can use both Nvidia and Intel. I used Nvidia's as it seem to do a better job.

It is extremely easy to use, all you have to do is simply drag and drop you image unto it once you have rendered it Poser and it will make a copy of it to where you have told to save it.

The program looks like this:

The denoisers themselves are also available on this persons website and you just download them and extract them to a folder. After that you just tell the program where they are located as shown in the image above.

The program and denoisers can be downloaded here and also seem to be used by Daz users.

https://taosoft.dk/software/freeware/dnden/

Conclusion

I hope this was a bit helpful, at least I learned some from going through it in details. But most importantly, I would say that a denoiser is an absolute must when rendering in PBR, take the quality of the render as far as your patience goes and then throw the final image through the program above will save a lot of time. Hopefully the speed of PBR will improve in the future or we can get some Realtime rendering like in Unreal, but until then I think this will be the best solution.

How does this denoiser compare with the built-in app in Poser?

hborre posted at 2:46PM Wed, 15 September 2021 - #4427503

How does this denoiser compare with the built-in app in Poser?

I don't know.

This one is based on this one:

https://www.youtube.com/watch?v=l-5NVNgT70U&ab_channel=NvidiaGameWorks

https://www.posersoftware.com/documentation/12/index.htm#t=Poser_Reference_Manual%2FRendering%2FSuperFly%2FApplying_Post_Effects_to_Renders.htm%23h&rhsearch=denoise

Obviously Poser uses Intel's Denoiser.

But all denoisers worth mentioning are publicly available (Intel's at Github as open source, NVidia provides its own SDK which is also basically freely available). The advantage of Intel's denoiser (and probably the reason why Bondware doesn't use NVidia) is that - unlike NVidia's software - it can be used with CPU only.

You have to process everything at a higher resolution (4k+ ) and you will be able to get rid of the noise without losing detail and have a crisp clear render. Start rendering the scene at 2,400 pixels wide at least... then enhance it to 4,800 pixels wide... now remove the noise and blur then rescale it back to 2,400 .... if you must lol

TwiztidKidd posted at 4:45PM Wed, 15 September 2021 - #4427512

You have to process everything at a higher resolution (4k+ ) and you will be able to get rid of the noise without losing detail and have a crisp clear render. Start rendering the scene at 2,400 pixels wide at least... then enhance it to 4,800 pixels wide... now remove the noise and blur then rescale it back to 2,400 .... if you must lol

Sure – but bigger render size means longer render time (double size, double time). That's what you avoid with a denoizer.

TwiztidKidd posted at 4:45PM Wed, 15 September 2021 - #4427512

You have to process everything at a higher resolution (4k+ ) and you will be able to get rid of the noise without losing detail and have a crisp clear render. Start rendering the scene at 2,400 pixels wide at least... then enhance it to 4,800 pixels wide... now remove the noise and blur then rescale it back to 2,400 .... if you must lol

Sure – but bigger render size means longer render time (double size, double time). That's what you avoid with a denoizer.

But let stop discussing this – maybe the OP has other interesting things to publish.

Superguide: use Superfly for super images I'll see myself out...

adp001 posted at 5:19PM Wed, 15 September 2021 - #4427510

But all denoisers worth mentioning are publicly available (Intel's at Github as open source, NVidia provides its own SDK which is also basically freely available). The advantage of Intel's denoiser (and probably the reason why Bondware doesn't use NVidia) is that - unlike NVidia's software - it can be used with CPU only.

Using the one I posted allow you to do it on any image. From the Daz forum it seems that the Nvidia one normally requires you to have a Nvidia graphic card, but this one allow anyone to use it. Whether that is true or not I don't know, since I have an nvidia card.

TwiztidKidd posted at 5:22PM Wed, 15 September 2021 - #4427512

You have to process everything at a higher resolution (4k+ ) and you will be able to get rid of the noise without losing detail and have a crisp clear render. Start rendering the scene at 2,400 pixels wide at least... then enhance it to 4,800 pixels wide... now remove the noise and blur then rescale it back to 2,400 .... if you must lol

"Sure – but bigger render size means longer render time (double size, double time). That's what you avoid with a denoizer.

But let stop discussing this – maybe the OP has other interesting things to publish."

But as Adp001 say you still have to render more. With the denoiser you can render fairly noise images and still get something fairly good out of it. But since the program works externally, you could probably combine both method and might give you a good or even better result, worth testing :)

Personally I have to say that it does a good job in removing noise, obviously it would be nice if there were some setting for how aggressive it is.

Thank you. And thanks for the link to the external denoiser.

I think this type of scrutiny is worth pursuing with stricter, incremental control. There are many variables to try yet there should be a standardization of material shaders that would compliment the testing. I don't know how much of an impact Material shaders have on rendering but I would imagine if your object material shaders are complicated or poorly assembled that calculations may affect how they render.

primorge posted at 6:19PM Wed, 15 September 2021 - #4427523

Thank you. And thanks for the link to the external denoiser.

Sure no problem. :)

3D-Mobster posted at 3:11AM Thu, 16 September 2021 - #4427518

adp001 posted at 5:19PM Wed, 15 September 2021 - #4427510

But all denoisers worth mentioning are publicly available (Intel's at Github as open source, NVidia provides its own SDK which is also basically freely available). The advantage of Intel's denoiser (and probably the reason why Bondware doesn't use NVidia) is that - unlike NVidia's software - it can be used with CPU only.

Using the one I posted allow you to do it on any image. From the Daz forum it seems that the Nvidia one normally requires you to have a Nvidia graphic card, but this one allow anyone to use it. Whether that is true or not I don't know, since I have an nvidia card.

The one you posted is just a wrapper around the original commandline utility :)

Yes, NVidia works only with NVidea cards. As I sayed: The big diffence with Intel's Denoiser is that you can do it with CPU only. Or with any Graphic card.

Here is Intels command line utility. Make your own script or batch to use it the way you like.

https://github.com/DeclanRussell/IntelOIDenoiser

Personally I have to say that it does a good job in removing noise, obviously it would be nice if there were some setting for how aggressive it is.

Mix your original image with the denoised result.