BMW 8 - Caustic + Dispersion experiments

Our forum gallery for displaying your finished renders.
PDRA
Posts: 51
Joined: Fri Sep 27, 2019 8:49 pm

BMW 8 - Caustic + Dispersion experiments

Post by PDRA »

Hi there!
I'm Pablo and I've been doing the switch from 3ds max to the open-source blender universe for about a year approx.
While trying blender, I bumped with this amazing render engine and I decided to try it and test its capabilities. I've been playing with it and below you will find some renders I've done.
BMW 8 - Caustic + Dispersion experiments (1).jpg
BMW 8 - Caustic + Dispersion experiments (2).jpg
However, I want to learn more about it! I've encountered some things that I will thank you if you can help me understand how I can solve them better.

- The first thing that I noticed is that after the denoising process the image kinda looks blurry and some details get lost. I mainly trying to enhance some details like the tire sidewall and the carbon fiber bump textures.
For the image of the BMW with the purple background I used 800 samples, Bi-dir (metropolis) engine, 1,25 px as the filter width and even though the final render is at 1920 x 1080 I rendered it at 2400x1350 to see if the resolution is the cause, but it wasn't. It took 1h 10m to render it. Should I try increasing the samples and deactivate the denoiser? How many samples should I try? Should I consider other parameter to make a sharper image? This kind of blurriness also happens at close-ups. (Image A)
A.jpg
____

- Apart from that, I've noticed that if I use a cycles material it doesn't render/show properly (Image B). You will find that I'm trying to create a carbon fiber texture. The pattern repetition doesn't match and the bump strength does not have any effect. It doesn't work either in 2.80 nor 2.81. Are cycles materials that incompatible? The only reason why I'm using that as a cycles material is that the cycle nodes offer the "Box" mapping option with the "Blend" parameter which is very handy for this kind of textures and I couldn't find anything similar in the Luxcore nodes. Is this kind of option available in this render engine?
B.jpg
____

Finally, I think that I found a general bug.
If I apply a material I have to move from render mode to solid view and go back to render mode to refresh the applied materials to the objects to the scene. If I don't do that after applying a material, it will look like nothing has changed. Sometimes it also happens when doing major changes in with the material nodes.

Thank you in advance for any critic and info you can provide me!

Here I will leave a link to my artstation porfolio if you are interested :D
www.artstation.com/pablodra

Regards, saludos
Pablo
User avatar
B.Y.O.B.
Developer
Developer
Posts: 4146
Joined: Mon Dec 04, 2017 10:08 pm
Location: Germany
Contact:

Re: BMW 8 - Caustic + Dispersion experiments

Post by B.Y.O.B. »

Wow, this is some outstanding stuff!
Can I add these to our gallery?
If yes, should I use your full name and artstation link to give credit?
- The first thing that I noticed is that after the denoising process the image kinda looks blurry and some details get lost. I mainly trying to enhance some details like the tire sidewall and the carbon fiber bump textures.
For the image of the BMW with the purple background I used 800 samples, Bi-dir (metropolis) engine, 1,25 px as the filter width and even though the final render is at 1920 x 1080 I rendered it at 2400x1350 to see if the resolution is the cause, but it wasn't. It took 1h 10m to render it. Should I try increasing the samples and deactivate the denoiser? How many samples should I try? Should I consider other parameter to make a sharper image? This kind of blurriness also happens at close-ups. (Image A)
You could try to set the filter to "None" and see if that helps.
If not, it might be a good idea to report this to the OIDN authors at https://github.com/OpenImageDenoise/oidn/issues
- Apart from that, I've noticed that if I use a cycles material it doesn't render/show properly (Image B). You will find that I'm trying to create a carbon fiber texture. The pattern repetition doesn't match and the bump strength does not have any effect. It doesn't work either in 2.80 nor 2.81. Are cycles materials that incompatible? The only reason why I'm using that as a cycles material is that the cycle nodes offer the "Box" mapping option with the "Blend" parameter which is very handy for this kind of textures and I couldn't find anything similar in the Luxcore nodes. Is this kind of option available in this render engine?
LuxCore currently doesn't have triplanar mapping like this.
However, it's in the works and will be part of the v2.3 release: https://github.com/LuxCoreRender/LuxCore/pull/252
In general, features that aren't available in LuxCore can't be supported by the "use Cycles nodes" option. Also, this option is very new and still missing a lot of features at the moment, you can find more info here.
Finally, I think that I found a general bug.
If I apply a material I have to move from render mode to solid view and go back to render mode to refresh the applied materials to the objects to the scene. If I don't do that after applying a material, it will look like nothing has changed. Sometimes it also happens when doing major changes in with the material nodes.
I have opened a bug report here: https://github.com/LuxCoreRender/BlendL ... issues/358
I guess you are using the stable release v2.2? It might be that the issue is already fixed in the latest dev. builds, I'll check.
PDRA
Posts: 51
Joined: Fri Sep 27, 2019 8:49 pm

Re: BMW 8 - Caustic + Dispersion experiments

Post by PDRA »

Hi B.Y.O.B.!
Thank you very much for your kind comment about my works and for helping me out!
I don't have any problem if you upload these renders to the gallery. It will be an honor! :D And yes you can use my full name and artstation link.

You could try to set the filter to "None" and see if that helps.
If not, it might be a good idea to report this to the OIDN authors at https://github.com/OpenImageDenoise/oidn/issues
Ok, I will be playing with the filter settings and check if I can improve it.
Apart from that, I checked the github page you sent and there is a post called "loosing texture details since V1.0" which describes a very similar issue like the one I have. Do you think that I should post something there?
LuxCore currently doesn't have triplanar mapping like this.
However, it's in the works and will be part of the v2.3 release: https://github.com/LuxCoreRender/LuxCore/pull/252
In general, features that aren't available in LuxCore can't be supported by the "use Cycles nodes" option. Also, this option is very new and still missing a lot of features at the moment, you can find more info here.
It is great to know that!!
I have opened a bug report here: https://github.com/LuxCoreRender/BlendL ... issues/358
I guess you are using the stable release v2.2? It might be that the issue is already fixed in the latest dev. builds, I'll check.
Yes, I've been using the 2.2 version in 2.81. Thanks for posting the bug. Can I post there if I find something or is it for developers only? I'm not a programmer.. just in case.

Something more that I would like to know if it is a bug is that when preparing the scene to render is very common for me to start checking the image by regions with the Ctrl+B option. When doing that, sometimes the image gets lighter or darker depending on which area I crop. I supposed that it was related to the light strategy parameter and the lighting calculation (I been using the log power option). But I tried by switching it to uniform and it kept doing the same thing. I don't know if it is something that only happens to me, a bad setting I'm doing, a bug or something else.

Please let me know which is the place to report this kind of thing or where I can post suggestions to help somehow!

Regards, saludos
Pablo
User avatar
Sharlybg
Donor
Donor
Posts: 3101
Joined: Mon Dec 04, 2017 10:11 pm
Location: Ivory Coast

Re: BMW 8 - Caustic + Dispersion experiments

Post by Sharlybg »

First congratulation for the outstanding work. ;)

About the oidn blurring issue : maybe switch to path tracer GPU + light tracing and a reasonnable amount of clamping can give you faster render high sample level while reducing denoising need. You will also keep all the render quality.

Welcome.
Support LuxCoreRender project with salts and bounties

Portfolio : https://www.behance.net/DRAVIA
User avatar
B.Y.O.B.
Developer
Developer
Posts: 4146
Joined: Mon Dec 04, 2017 10:08 pm
Location: Germany
Contact:

Re: BMW 8 - Caustic + Dispersion experiments

Post by B.Y.O.B. »

PDRA wrote: Tue Dec 31, 2019 4:59 am I checked the github page you sent and there is a post called "loosing texture details since V1.0" which describes a very similar issue like the one I have. Do you think that I should post something there?
It might be related, but I'm not sure. Better create a separate issue, if it's a duplicate, they can close it.
PDRA wrote: Tue Dec 31, 2019 4:59 am Thanks for posting the bug. Can I post there if I find something or is it for developers only? I'm not a programmer.. just in case.
You can post there, it's open to anyone.
The github issue tracker is also the place to ask questions and open feature requests, not only for bug reports.
PDRA wrote: Tue Dec 31, 2019 4:59 am Something more that I would like to know if it is a bug is that when preparing the scene to render is very common for me to start checking the image by regions with the Ctrl+B option. When doing that, sometimes the image gets lighter or darker depending on which area I crop.
Maybe you are using an automatic tonemapper?
PDRA
Posts: 51
Joined: Fri Sep 27, 2019 8:49 pm

Re: BMW 8 - Caustic + Dispersion experiments

Post by PDRA »

Hi there!
I've been trying the different things that you suggested.
First congratulation for the outstanding work. ;)

About the oidn blurring issue : maybe switch to path tracer GPU + light tracing and a reasonnable amount of clamping can give you faster render high sample level while reducing denoising need. You will also keep all the render quality.

Welcome.
Thank you Sharlybg for your kind comment! I've tried setting the engine to path but it had the same result than the Bi-dir after denoising. Also for the renders I did with the BMW, as far as I tried and understand, it is necessary to use Bi-dir with metropolis to obtain that kind of caustics rendering.
I've also tried to denoise it with the 2.81 node and I found that the one done with luxcore was better.
I suppose that the best way to sort out this problem is on postpro by combining the denoised and predenoised image. With that I can get the best of both.

Denoiser differences
Luxcore denoiser
Denoiser Luxcore.png
2.81 denoiser node
Denoiser 2.81.png
Main differences
Areas.jpg
____

Bidir vs Path
Bidir
Noisy - Bidir.png
Path
Noisy - Path.png
____
It might be related, but I'm not sure. Better create a separate issue, if it's a duplicate, they can close it

You can post there, it's open to anyone.
The github issue tracker is also the place to ask questions and open feature requests, not only for bug reports.
I will try to post there then!
Maybe you are using an automatic tonemapper?
Regarding to this issue, I attach to images to explain the effect better. You will see a difference in brightness. I don't know exactly where I can find that automatic tonemapper option you are suggesting.
Cropp issue.jpg
Cropp issue 2.jpg
____

Happy new year to everyone!

Regards, saludos
Pablo
User avatar
FarbigeWelt
Donor
Donor
Posts: 1046
Joined: Sun Jul 01, 2018 12:07 pm
Location: Switzerland
Contact:

Re: BMW 8 - Caustic + Dispersion experiments

Post by FarbigeWelt »

PDRA wrote: Mon Dec 30, 2019 7:40 pm Should I consider other parameter to make a sharper image? This kind of blurriness also happens at close-ups. (Image A)

A.jpg

Regards, saludos
Pablo
Hi Pablo
Depending on the amount of noise the results of OIDN look blurry or sterile smooth. As already said in this post by someone else, pixel filter should be off (0.0) to get crispy picture.
About the over smoothed result render a little series with different samples, e.g. 25, 75, 250, 1000 and compare the images in relevant closeups.
My, including others‘, way to deal with the de-noiser is to setup Blender composer color mix node to mix raw and de-noised picture with 0.75 to 0.95 amount of de-noised image. Adjust this value according to your taste. It is a good compromise between fine noise and details and almost noiseless but also missing some details sometimes.
Light and Word designing Creator - www.farbigewelt.ch - aka quantenkristall || #luxcorerender
MacBook Air with M1
PDRA
Posts: 51
Joined: Fri Sep 27, 2019 8:49 pm

Re: BMW 8 - Caustic + Dispersion experiments

Post by PDRA »

Maybe you are using an automatic tonemapper?
Thanks BYOB! I found that option and it was that.
Hi Pablo
Depending on the amount of noise the results of OIDN look blurry or sterile smooth. As already said in this post by someone else, pixel filter should be off (0.0) to get crispy picture.
About the over smoothed result render a little series with different samples, e.g. 25, 75, 250, 1000 and compare the images in relevant closeups.
My, including others‘, way to deal with the de-noiser is to setup Blender composer color mix node to mix raw and de-noised picture with 0.75 to 0.95 amount of de-noised image. Adjust this value according to your taste. It is a good compromise between fine noise and details and almost noiseless but also missing some details sometimes.
Thank you FarbigeWelt,
I've been trying some different samples setups and they seem to end in the same result. However I will pay more attention to the samples and pixel filter option for future renders.
Regarding to the mix technique, I will be doing something like that, but I believe it is easier to do directly in photoshop to enhance certain areas rather than others.

If you wouldn't mind guys I have some others questions to ask you. Your comments help me a lot.

- I've been trying to use a background image to do some compositions. However it doesn't load the image (it appears all black). I'm using the Path method at this moment. Am I doing something wrong? I tried with a smaller size version of the image and it didn't work.
Backround image problem.jpg
- Is there any way to control just the IOR of certain materials (specialy the Disney one, which I believe is like the principled BSDF luxcore version)? I'm trying something with the glossy coatting material, and it kinda works but maybe there is a simpler way. It is to control de Fresnel effect of the car paint.

- The car paint seems to be bugged. Apparently there it doesn't "stop" to bounce the light. I tried by using the clamp option and reducing the amount of max bounces but it didn't have that much of effect and it made the rest of the image look bad. I decided to use a custom material node setup to use for the car paint in this ocassion.
Car paint
Car paint problem.jpg
My custom setup
Car paint solved with custom nodes.jpg
- Is there any way to deactivate shadow casting / reflection / etc of certain objects? The front light plastic is making the interior of the light look too dark and I might want to cheat a bit there.

- And lastly, is there any way to add multiple bumps? also luxcore doesn't support a drag and drop method right?

Regards, saludos
Pablo
User avatar
B.Y.O.B.
Developer
Developer
Posts: 4146
Joined: Mon Dec 04, 2017 10:08 pm
Location: Germany
Contact:

Re: BMW 8 - Caustic + Dispersion experiments

Post by B.Y.O.B. »

PDRA wrote: Thu Jan 02, 2020 7:54 pm - I've been trying to use a background image to do some compositions. However it doesn't load the image (it appears all black). I'm using the Path method at this moment. Am I doing something wrong? I tried with a smaller size version of the image and it didn't work.
It should work, strange.
Does it also happen in a simple (default cube) scene? If yes, can you upload the image so we can check if there's something wrong with it, or if it triggers some bug in LuxCore?
PDRA wrote: Thu Jan 02, 2020 7:54 pm - Is there any way to control just the IOR of certain materials (specialy the Disney one, which I believe is like the principled BSDF luxcore version)? I'm trying something with the glossy coatting material, and it kinda works but maybe there is a simpler way. It is to control de Fresnel effect of the car paint.
The IOR setting of the glossy coating material is just a more physically oriented way of specifying the reflection brightness (higher IOR = brighter reflections). On materials without this setting, you can just make the specular color brighter/darker to achieve the same effect, or in case of the disney material, change the specular value.
PDRA wrote: Thu Jan 02, 2020 7:54 pm - The car paint seems to be bugged. Apparently there it doesn't "stop" to bounce the light.
The colors you have set are too bright (100% brightness). Try making them a bit darker.
This wiki page might also help: https://wiki.luxcorerender.org/LuxCoreR ... Parameters
PDRA wrote: Thu Jan 02, 2020 7:54 pm - Is there any way to deactivate shadow casting / reflection / etc of certain objects? The front light plastic is making the interior of the light look too dark and I might want to cheat a bit there.
Not in v2.2, but in the upcoming v2.3 you will be able to disable shadow casting of objects.
In v2.2, you can try to use a glass material with the "architectural" option for the front light plastic, however this will disable refractions for the material, you'll have to judge if that looks OK.
PDRA wrote: Thu Jan 02, 2020 7:54 pm - And lastly, is there any way to add multiple bumps? also luxcore doesn't support a drag and drop method right?
You can add them with a math node set to "Add".
We can not support drag and drop because Blender does not expose this in the Python API (I would have liked to though).
PDRA
Posts: 51
Joined: Fri Sep 27, 2019 8:49 pm

Re: BMW 8 - Caustic + Dispersion experiments

Post by PDRA »

Thank you BYOB for the detailed reply!
I have tried some of the things you said and it helped me a lot.
I will be trying the car paint and add bumps tips later.
I tried with a default cube and the background image didn't work.
New camera with same post-rot missing background.jpg
I attach a sample image for testing and a few viewport screenshots.
Somehow I'm sorting it out using Fspy but it is a bit odd because on the imported camera from Fspy I get the background but it overlays with the HDRI I'm using when I activate the render mode. To sum up I align it in solid view but I can't properly see how it would look in render mode (because it overlays or it doesn't appear).
FSPY camera works ok with solid view
The attachment FSPY camera works ok with solid viewI.jpg is no longer available
However it overlays the image with the HDRI
FSPY camera works ok with solid viewI.jpg
Positioned a new camera but it doesn't shows the image
FSPY camera overlays with HDRI.jpg
Regards, saludos
Pablo
Attachments
Default cube same problem.jpg
Post Reply