Recommended Posts

Hello everyone,

I'd like to understand picture resolutions and pixelation when baking textures to meshes in blender. Here I have a very simple node setup where I'm using a repeating pattern texture onto an unwrapped cube. The pattern image is 2220X2220 and I'm baking onto a baking image that has the same dimensions (I'm using "combined" as baking option).

On the left is the rendered mesh up close, where you can see the nice detail. On the right is the finished bake, where the pixelations shows.

 

Here is the node setup:

 

Could someone explain what is going on during this baking process that transforms the bake into this pixelation and what is the correct way to fix this? Also, I'm using a cube for this example since it's easy to unwrap but my goal is to texture clothing items so the unwrap is sometimes not as neat as this.

Thank you all for your help and advice.

nodes.JPG

Pixalation 2.JPG

Share this post


Link to post
Share on other sites

It looks like unfiltered antialiasing. Most likely @Chic Aeon has encountered this issue too, as i remember from a past thread, but i don't know how she managed to fix it. In Maya this would be fixed by using a higher Anti-Aliasing value in combination with a better filter type (most common is the Gaussian, but the best in my opinion is Blackman-Harris)

 

Good luck =)  

Share this post


Link to post
Share on other sites

In Maya? So I have to fix the pattern image in an image software before bringing it inside blender? Or do I have to fix the baked image? I was hoping to get a nice quality bake directly from blender.

(I use photoshop by the way)

Share this post


Link to post
Share on other sites
1 hour ago, Spinell said:

In Maya? So I have to fix the pattern image in an image software before bringing it inside blender?

Ehm no, Maya is another 3D software. You should look for the Anitialiasing settings and filter types available in Blender in order to get less pixellation.

Share this post


Link to post
Share on other sites
Posted (edited)

Thanks for Optimo for tagging me :D

Maya is "the" 3D modeling program :D, so no the problem is within Blender.

 

A couple of very nice folks tried to help me fix this issue. They used more industry standard software so they didn't exactly know the answer but they found areas in Blender that corresponded to what would fix things in their programs. You can likely find that thread searching for something like "blender bake jagged edges" or the like. I can never seem to find threads.

The bottom line was that I tried all their much appreciated suggestions. They were complex (to me anyway). I couldn't see that they helped. There are things that Blender just doesn't do as well as the high priced alternatives and this is one of them. I have ALMOST made something akin to what I call a "Maya bake" -- that lovely mouthwatering look with the soft shading, but I know I will likely never get things as good looking as they could be in Maya. That's just how it is. 

 

Other thoughts but really I think you did as good as you can do. 

You baked to an odd size which MIGHT have made the problem worse. So 512 - 1024 - 2048 are usually recommended. I bake to 1024. Since the largest size you can upload to SL is a 1024, you will need to resize in a graphics program. This will automatically reduce some of that pixelation.  You can try different compressions and sharpenings and hopefully get something you will be happy with.  Note that your texture will be COMPRESSED again on upload (imagine you know that) making it softer yet again. If you try and upload a 2048 texture the uploader compresses to 1024 and not all that well (so resize before uploading). 

Since we can only see a portion of the bake it is hard to tell how much of the texture plane it takes up. You COULD perhaps make the piece bigger?  I am not advocating that really as I doubt you would see much difference after upload. 

 

One thing you can do in general (this for folks reading this thread later) is to make what you can  linear since Blender does a really awful job with pixelation on angles. But sometimes you will just have angles - no way around it. In that case I soften the edges AFTER baking in my graphics program. I use a soften brush in Corel.

That of course doesn't help with the pattern you are trying to reproduce. I am guessing though that after it is resized and uploaded that you won't see the pixelation. 

 

edges.png.d5ef415aa10d698309e5ddc3d4aea1bc.png

5b1c2c54a7c29_edgesafter.png.84083d6fb3ea21beb7d8048b3b552696.png

 

Inworld close up:

image.png.5d78b56c8f6b01ede16f7b618335e18d.png

Edited by Chic Aeon
spelling

Share this post


Link to post
Share on other sites
Posted (edited)
6 minutes ago, OptimoMaximo said:

Ehm no, Maya is another 3D software. You should look for the Anitialiasing settings and filter types available in Blender in order to get less pixellation.

Yes, she should (took me a long time to do all that posting with photos LOL) but while it SHOULD help, it doesn't SEEM to help. If someone really has the answer, I would love to see it on this thread.  :D.

 

Edited by Chic Aeon

Share this post


Link to post
Share on other sites
3 hours ago, Chic Aeon said:

Thanks for Optimo for tagging me

Well i knew you were working on it from the thread you mention in your post, but i didn't know how it turned out for you. There should be a feature request to the Blender Devs for a baking improvement to cover this then.

@Spinell Chic's suggestions for manual post processing is the best way to go, at this point.

Share this post


Link to post
Share on other sites
1 hour ago, OptimoMaximo said:

Well i knew you were working on it from the thread you mention in your post, but i didn't know how it turned out for you. There should be a feature request to the Blender Devs for a baking improvement to cover this then.

@Spinell Chic's suggestions for manual post processing is the best way to go, at this point. I am in 2.78C since the newer version didn't work for me at all (many features that I use all the time would no longer work).    

Looking at the report on 2.8 I see at least two things that will help us SL folks but nothing about the really bad diagonal bakes.

 

AGREED but those of us that need bakes seem to be in the minority. The newer version of Blender MIGHT be better, but there have been a variety of bugs (no way they are 'features') for a long while that haven't been address -- so I suspect not. 

I am in 2.78c since the newer version did SO not work for me ^^.  I doubt the diagonal baking has been improved though. I didn't notice any great change when I was testing it last year. 

The info on 2.8 shows a couple of new features that will be helpful for SL folks (multi object editing and a new "layers" system that isn't layers) but nothing that I see on better baking. I think the addition of Cycles long ago was the last big change for baking.  

 

There are a fair amount of workaround-everyday bugs in Blender but they aren't showy and we CAN work around them. So once you are used to always checking -- it works. 

In the OP's example though there is no real fix in the post processing unless you like "blur". Some of the more popular designers use that feature (or perhaps a choice within Maya).  Of course their texture plane  project count is about four to eight times mine LOL  --- soooo :D 

And a note:  I was thinking later this afternoon when talking to a friend that there IS a post processing area INSIDE of Blender. I only used it once to make some gorgeous killer (Maya) shadows. But it took me so long and then I again had to post process in Corel, it didn't seem worth the effort. But maybe that is a good thing to explore. It isn't the "change a setting and get a good bake" answer but it might be helpful -- and in theory I should be smarter now :D. 

And I am off for the day a bit late.   Hope this thread helps some folks along the way. 

Share this post


Link to post
Share on other sites
9 hours ago, Chic Aeon said:

In the OP's example though there is no real fix in the post processing unless you like "blur". Some of the more popular designers use that feature (or perhaps a choice within Maya).

Antialiasing is basically edge detection and blurring along the line, and the filter type is how it gets blurred (plus how many pixels get involved around each sampled pixel) Indeed:

9 hours ago, Chic Aeon said:

And a note:  I was thinking later this afternoon when talking to a friend that there IS a post processing area INSIDE of Blender. I only used it once to make some gorgeous killer (Maya) shadows.

That is the post processing for video editing usually, with nodes that operate like photoshop filters

However, if you're willing/able to invest a few hundred bucks, VRay is a good choice and supports Blender. Takes a bit of learning but it's definitely a great option.

Share this post


Link to post
Share on other sites
2 hours ago, OptimoMaximo said:

That is the post processing for video editing usually, with nodes that operate like photoshop filters

Yes, but it does work for stills too -- at least on the project that I did. I am just not sure the time involved is worth it. 

 

You CAN blur the whole baked texture  in graphics software of course but that takes away the texture DETAILS in the rest of the texture -- those that I like to keep. So I do the soften thing with a small brush on the edges which can sometimes be VERY tedious.  But I am one who cams in to see those details.   

There are plenty of folks who just leave those jagged edges, so that is a choice. Just not mine. 

Share this post


Link to post
Share on other sites
2 hours ago, Chic Aeon said:

Yes, but it does work for stills too -- at least on the project that I did

Sure thing, you work on one frame to apply the editing to the whole sequence, no question about that

 

2 hours ago, Chic Aeon said:

I am just not sure the time involved is worth it.

Can you export nodes setups for that in order to reuse it in another project? You would have just to tweak it instead of re-doing all the nodes piping

Share this post


Link to post
Share on other sites
11 minutes ago, OptimoMaximo said:

Can you export nodes setups for that in order to reuse it in another project? You would have just to tweak it instead of re-doing all the nodes piping

 I am not sure about exporting node setups. I have mine "saved" in files but of course they lose their paths to the texture image(s) from time to time. 

***************************

A quick search in Google found this page which looks like a great way to save the node groups I use most often (mine are texture based).  This same idea could likely work for lighting set ups which would be helpful too. 

https://bwide.wordpress.com/2012/12/08/how-to-useinstall-node-groups-in-blender/

 

So thanks for that suggestion. I use append often but mostly for materials and now and then objects.  Will see how that works when I finish (or get tired of) the project I am just starting. 

Seems like this COULD solve the path issue that we have talked about. 

Since summer is SO VERY SLOW and I have no GIANT project like last year, I am going to be working on improving textures which of course in Cycles has a lot to do with lighting. So this will work right in. 

 

 

 

 

  • Like 1

Share this post


Link to post
Share on other sites

I often save node setups myself, especially when I had materials with particular setups like a nice gold texture, or a polished shoe.

Well... after reading all of your awesome comments, I'm still a bit unsure on how to proceed to fix this loss of detail when baking. I'm going to search for some antialiasing techniques to see what I can find but I wish there was a way to improve this either in Blender, with the nodes setup, or by doing something in Photoshop before bringing the image into the nodes.

And you say Maya users don't really get this problem? I'm starting to consider switching if it would get rid of this pixalated mess... (I'll save it for a last resort maybe)

Share this post


Link to post
Share on other sites
3 minutes ago, Spinell said:

And you say Maya users don't really get this problem?

Maya users do get the ability to have filtering applied on bake time, using the appropriate settings. Not using filtering gives you pixellation anyway, and getting overboard with these filters lengthen the bake time and decrease the texture sharpness like applying a quite strong blur in photoshop.

Share this post


Link to post
Share on other sites

While my texture and lighting improvements project is coming along well with minimal issues (always a thrill) I can't say that I found any answers to this. I tried a bunch of settings that sounded promising in the bake section including upping volume sampling and dither. Neither made any impact on 400 and 500 bakes on a 1024 texture. 

And yes, I have seen the overblurriness in some Maya bakes:D.

One other thing that can be done is to bake at 2048 (or even bigger I guess) and resize in a graphics program. This gives you more starting pixels and an automatic softening on resizing. I did this a few times a couple of years ago and there was a difference. I am not sure how much it would help the OPs texture though. There are just some things that don't translate well.  I could do that 2014 bake but it would be so painfully slow on anything complex (sometimes baking to several 1024s at one time) that I would probably hit the computer with a mallet from boredom. But someone with a really smoking machine might try that :D.

 

Onward. 

Share this post


Link to post
Share on other sites
19 hours ago, Chic Aeon said:

settings that sounded promising in the bake section including upping volume sampling and dither

Chic, Volume sampling is related to Volumetric shaders, like a self contained smoke or atmospheric scattering, and i would suggest to turn that always to zero if you're not doing scenery renders. Dithering instead should do something but it's most likely to be unnoticeable if it's applied on <2K render images

19 hours ago, Chic Aeon said:

One other thing that can be done is to bake at 2048 (or even bigger I guess) and resize in a graphics program. This gives you more starting pixels and an automatic softening on resizing. I did this a few times a couple of years ago and there was a difference.

but...

19 hours ago, Chic Aeon said:

There are just some things that don't translate well.

Exactly. This is ground for the type of approach that i was trying to explain in the other thread where i was using 2 UVSets, one for the main bake within the regular UV space (0-1,big image) and a second UVSet using more UV Tiles/texturable faces. The problem with baking a huge texture is that, no matter what you do, after resizing you get some texture degradation as a mandatory collateral damage to the image.

20 hours ago, Chic Aeon said:

I could do that 2014 (edit: I guess you mistyped 2048) bake but it would be so painfully slow on anything complex (sometimes baking to several 1024s at one time) that I would probably hit the computer with a mallet from boredom.

That's why you should try and use the LightPath feature instead of the general sampling i can see you refer to. That way you can tell your software to not waste samples on features you're not using like SSS or Volume and you can also set (if i remember Blender correctly) how many bounces you allow to be calculated before the sample dies off and stops being accounted for. Moreover, the sample bounces values imply that there should be some geometry to bounce off from, so setting up a sort of huge floor with walls at a distance helps avoiding waste of samples that fall in the void to never stop being calculated because they never bounce. I don't know whether lights in Blender have their own samples, but if they do, set those to a high value and keep the scene render sampling relatively lower. These three things help improve render/bake time in relation to a bigger output size.

Share this post


Link to post
Share on other sites

I'm sure this isn't a Blender software specific issue. I used to have the same problem when baking in other software. I fix this issue by double-ing the size of my input image.

In this case 2220x2220 x2 = 4440x4440 baked to 4440x4440.

In photoshop up-size your texture x2 using "nearest neighbor". Nearest neighbor resizes every pixel cleanly with even numbers.

In Krita up-size your texture x2 using "box"

Any blurring introduced by the baking process is on the giant image so when you shrink to your needed resolution the blurring is minimized.

You should be ok if your computer has 8 gigs of memory. If you up-size it further to 8000x8000 your computer, or Blender might crash.

I would try this without a complicated node setup first. Just a simple bake to try this.

Edited by RipleyVonD
Typo

Share this post


Link to post
Share on other sites
11 minutes ago, OptimoMaximo said:

Chic, Volume sampling is related to Volumetric shaders, like a self contained smoke or atmospheric scattering, and i would suggest to turn that always to zero if you're not doing scenery renders. Dithering instead should do something but it's most likely to be unnoticeable if it's applied on <2K render images

Volume Sampling was set at 0.1 by default but I typed in 0 and got 0.000001  :D  

Here is what the Lightpath area looks like (defaults)

image.png.4a90e8aac732c4b1d00495aa3620a2ee.png

So anything in there that I would normally turn off?  I use diffuse and glossy for sure. Volume is at 0 as it.  I don't use transparency but don't bake to a alpha texture anyway (except when Blender does that FOR me on a rebake (VERY FRUSTRATING).  And I do have shadows used.  Not sure exactly what reflective and refractive refers to in this context.

I typically have a floor (well a plane) under objects but not walls. I could try that. Baking isn't really a problem for me until I am doing a house and need to bake to three or more textures at a time :D.  Then it can get a little boring and thumb twiddling. 

Here is an example of one lamp and its settings. The only thing I usually change is the strength.  Not sure if this is the default set up more or less or if I made changes long ago :D. 

image.png.07a9ea2c4b5c4cf6e25afb1edbb0590d.png

 

My latest project turned out spectacularly so very happy with that. It has some subtle cast shadows but is much more "realistic" looking as well a nice and soft (the shadow parts, not the main textures). So improvements.  Just need a base and I am done. 

Thanks for the hints! 

 

 

Share this post


Link to post
Share on other sites
3 minutes ago, RipleyVonD said:

I'm sure this isn't a Blender software specific issue. I used to have the same problem when baking in other software. I fix this issue by double-ing the size of my input image.

In this case 2220x2220 x2 = 4440x4440 baked to 4440x4440.

In photoshop up-size your texture x2 using "nearest neighbor". Nearest neighbor resizes every pixel cleanly with even numbers.

In Krita up-size your texture x2 using "box"

Any blurring introduced by the baking process in on the giant image so when you shrink to your needed resolution the blurring is minimized.

You should be ok if your computer has 8 gigs of memory. If you up-size it further to 8000x8000 your computer, or Blender might crash.

I would try this without a complicated node setup first. Just a simple bake to try this.

Hey there Ripley.   

This might help the OP for sure as we don't know how hefty that computer might be. But with MY system -- while it works it takes a very long time. 

 And my graphics software isn't as classy as Photoshop so my texture will degrade.  Unfortunately for some, Cycles takes SO MUCH MORE PROCESSING POWER.    

 

CPU: Intel(R) Core(TM) i7-5820K CPU @ 3.30GHz (3298.09 MHz)
Memory: 16286 MB
OS Version: Microsoft Windows 10 64-bit (Build 16299)
Graphics Card Vendor: NVIDIA Corporation
Graphics Card: GeForce GTX 980/PCIe/SSE2

I just don't like to wait half an hour for a bake that might not be the FINAL bake. I bake at 500 for a house and I am usually baking to three or four textures at a time (or often anyway) so that is already 2048 in some sense.   

I have made some changes and I am liking what I am getting (small rustic house).  I am very happy with my current bake. Maybe I am just not a patient as I should be *wink*.

 

Share this post


Link to post
Share on other sites

Hi Chic. I understood this thread asked how to bake a texture image to a mesh with minimized blur. Which in my opinion is the least Cycles intensive task I can think of. Photoshop, Krita, or Gimp can all resize an image by 2 using a "nearest neighbor" or similar filter easily. Sounds like your using your graphics program a different way. The higher the input image the longer it takes to bake that is correct I forgot to mention that.

Actually I haven't tried this on Blender myself and I'm unsure how long it takes to bake, but resizing the input image x2 is how blurring is avoided, if its small by x4. I need to experiment with the cycles engine to see what settings need to be tweaked so baking doesn't take too long.

Edited by RipleyVonD
Grammar for clarity

Share this post


Link to post
Share on other sites
55 minutes ago, RipleyVonD said:

Hi Chic. How I understood, this thread asked how to bake a texture image to a mesh. Which in my opinion is the least Cycles intensive task I can think of. Photoshop, Krita, or Gimp can all resize an image by 2 using a "nearest neighbor" or similar filter easily. I'm sure all image editors can do that. The higher the input image the longer it takes to bake that is correct I forgot to mention that.

Actually I havent't tried this on Blender myself and I'm unsure how long it takes to bake, but resizing the input image x2 is how blurring is avoided, if its small by x4. I need to experiment with the cycles engine to see what settings need to be tweaked so baking dosen't take too long.

Gotcha --

Well "I" was assuming that the OP graphic was "part" (a close up) of a complete Cycles bake for a project -- mostly since I can't see why you would want to bake a plain texture that looks much like the texture starting out with.  :D.

 

Actually a way that DOES WORK (not advising but it does work) is to make the your object islands LARGER. This is how some designers get fantastic looking bakes.  So a 1024 coffee cup for example. (And then there was that matchstick example - LOL).

That comes at a huge texture download cost however so isn't the SMARTEST method IMHO.  Other designers soften the complete texture. It is partly a matter of taste.  You can see that there is a threshold that can't be gone under (re pixels per area) easily by making magazine covers :D.   There IS a trick there but in general you can only get SO small on the texture plain or it looks really horrible and unusable.

So like for many things, it is about finding some middle ground -- IMHO of course :D.

Will look forward to hearing your findings.   

Share this post


Link to post
Share on other sites

"Well "I" was assuming that the OP graphic was "part" (a close up) of a complete Cycles bake for a project -- mostly since I can't see why you would want to bake a plain texture that looks much like the texture starting out with"

Had feeling we understood differently. Baking a texture to your mesh is useful when you paint on your mesh, with brush strokes or projected textures, or both, then later decide to change the uv islands around. Instead of recreating all your work you can bake it to you new uv islands.

"Actually a way that DOES WORK (not advising but it does work) is to make the your object islands LARGER. This is how some designers get fantastic looking bakes.  So a 1024 coffee cup for example. (And then there was that matchstick example - LOL).

That comes at a huge texture download cost however so isn't the SMARTEST method IMHO.  Other designers soften the complete texture. It is partly a matter of taste.  You can see that there is a threshold that can't be gone under (re pixels per area) easily by making magazine covers :D.   There IS a trick there but in general you can only get SO small on the texture plain or it looks really horrible and unusable."

It's true optimizing your uv islands for maximum pixel density is a good workflow plan.

"Will look forward to hearing your findings"

Yeah I'll definitely share it here. Ok hopefully I don't need to edit this later. Lol

Share this post


Link to post
Share on other sites
3 hours ago, Chic Aeon said:

Volume Sampling was set at 0.1 by default but I typed in 0 and got 0.000001  :D  

I don't get why a round zero is not accepted, but it's ok xD

3 hours ago, Chic Aeon said:

Here is what the Lightpath area looks like (defaults)

image.png.4a90e8aac732c4b1d00495aa3620a2ee.png

So anything in there that I would normally turn off?  I use diffuse and glossy for sure. Volume is at 0 as it.  I don't use transparency but don't bake to a alpha texture anyway

Checkboxes: reflective is what goes with Gloss shader, Refractive is what would go with IOR. This latter is really needed only if you use IBL (Image Based Lighting), which means hooking up an environemt HDRI image (thing that i STRONGLY recommend to do). Unless you have thick lens shaped glass, Refraction is not really much needed, but you should test it. Metals may benefit from that too.

Bounces side: Max is set to the same value of the property that has the highest value among those on the lower side.

Transmission actually means the translucency (example: foliage or wax) or simple non refractive glass. If you don't have any of these two material types, dump it to zero. 

Now for the core properties you want to use for baking: Diffuse is set to 4, but the higher you go, the better lit (and less noisy) the diffuse colors should be. When you have pretty dark and noisy corners, setting this higher improves both luminosity and noise. Same goes for Glossy, which is the important bit for metals as they don't have a diffuse component, so if a metal reflection looks noisy in its color, this is the setting to touch, not the diffuse. So i would suggest to recycle those spare 12 samples from the unused Transmission over to these two properties, setting them to maybe 6 each one. Once you've found a good balance, the highest value among those should be set as max bounces. Min bounces should stay at 3 (really the bare minimum, but you can up this to maybe 4).

I'd like to know what Filter G:0.00 is in its extended version xD

3 hours ago, Chic Aeon said:

Here is an example of one lamp and its settings. The only thing I usually change is the strength.  Not sure if this is the default set up more or less or if I made changes long ago :D. 

image.png.07a9ea2c4b5c4cf6e25afb1edbb0590d.png

Now here: samples from this light look like default for an open world type of environment, and to me it looks way overkill if set to 1024 max bounces. Samples from this light would just flood the scene and after a few bounces, they will travel in the void and won't stop being accounted for. So, something more realistic for this may be around 10 or 12, and i would go higher only if set to Portal mode. However i don't really know how this property was actually coded within Blender, you'd better do testing on this. I'm talking from a Maya-Arnold perspective and Blender devs might have had a different measurement unit in mind.

Portal mode: this is something that goes along with an environmental HDRI, so that the light set as portal doesn't actually add up to the total in the scene, it just funnels the lighting from the HDRI into an INDOOR area: great example for this is to put them on the windows pointing indoors. This mode is especially useful when you do render of an interior and the environment lighting is not needed outside of your mesh, so that the exterior doesn't get calculated. You can mix normal mode area lights for the exterior and portal mode ones for the interior, saving a lot of resources and increasing quality brought from the HDRI environment

1 hour ago, RipleyVonD said:

Baking a texture to your mesh is useful when you paint on your mesh, with brush strokes or projected textures, or both, then later decide to change the uv islands around. Instead of recreating all your work you can bake it to you new uv islands.

This is called remapping or texture transfer or texture warp and it's basically the method i was explaining in my previous post. with the difference that i suggest to re-map everything you've got on a high res texture onto smaller textures spread across multiple UV tiles. But this isn't really supported in Blender, however there is a work around. This implies the use of a second UVSet (UVMap in Blender, within the MeshData panel, the triangle icon)

1 hour ago, RipleyVonD said:

It's true optimizing your uv islands for maximum pixel density is a good workflow plan.

This also relates to my method: i do a very high res texture, free from worries of texture space, then remap it to the new UV map with the optimized UVs, organized in more UV tiles, because...

3 hours ago, Chic Aeon said:

I just don't like to wait half an hour for a bake that might not be the FINAL bake. I bake at 500 for a house and I am usually baking to three or four textures at a time (or often anyway) so that is already 2048 in some sense.   

... which means that for each texture that Blender starts to bake, the whole process of calculation RESTARTS from SCRATCH, since Blender doesn't have lightmap caching features. That is why Light Path is better on one single image: you do your tests with low samples and for the properties that you actually intend to use (dumping the useless ones) to TEST the result on a smaller res texture, so it takes less time. You can then test how it looks in regard of lit areas, whether it looks correct or not, not caring of the quality in the first stage. At this point you shouldn't care if it's noisy, you should check how the lighting hits and bounces in general. When it looks ok, you can launch the final render on the single high res which will involve calculations only once. When all is good, texture transfer to the other UVSet that is split into multiple textures. This transfer doesn't take actual render and is very fast. This way, the same high res result you would have gotten in a long time to bake, shrinks down quite noticeably but with a far greater texture quality and a directly proportional weaker headache :D

 

Edited by OptimoMaximo
  • Thanks 1

Share this post


Link to post
Share on other sites
1 hour ago, OptimoMaximo said:

I don't get why a round zero is not accepted, but it's ok xD

Checkboxes: reflective is what goes with Gloss shader, Refractive is what would go with IOR. This latter is really needed only if you use IBL (Image Based Lighting), which means hooking up an environemt HDRI image (thing that i STRONGLY recommend to do). Unless you have thick lens shaped glass, Refraction is not really much needed, but you should test it. Metals may benefit from that too.

Bounces side: Max is set to the same value of the property that has the highest value among those on the lower side.

Transmission actually means the translucency (example: foliage or wax) or simple non refractive glass. If you don't have any of these two material types, dump it to zero. 

Now for the core properties you want to use for baking: Diffuse is set to 4, but the higher you go, the better lit (and less noisy) the diffuse colors should be. When you have pretty dark and noisy corners, setting this higher improves both luminosity and noise. Same goes for Glossy, which is the important bit for metals as they don't have a diffuse component, so if a metal reflection looks noisy in its color, this is the setting to touch, not the diffuse. So i would suggest to recycle those spare 12 samples from the unused Transmission over to these two properties, setting them to maybe 6 each one. Once you've found a good balance, the highest value among those should be set as max bounces. Min bounces should stay at 3 (really the bare minimum, but you can up this to maybe 4).

I'd like to know what Filter G:0.00 is in its extended version xD

Now here: samples from this light look like default for an open world type of environment, and to me it looks way overkill if set to 1024 max bounces. Samples from this light would just flood the scene and after a few bounces, they will travel in the void and won't stop being accounted for. So, something more realistic for this may be around 10 or 12, and i would go higher only if set to Portal mode. However i don't really know how this property was actually coded within Blender, you'd better do testing on this. I'm talking from a Maya-Arnold perspective and Blender devs might have had a different measurement unit in mind.

Portal mode: this is something that goes along with an environmental HDRI, so that the light set as portal doesn't actually add up to the total in the scene, it just funnels the lighting from the HDRI into an INDOOR area: great example for this is to put them on the windows pointing indoors. This mode is especially useful when you do render of an interior and the environment lighting is not needed outside of your mesh, so that the exterior doesn't get calculated. You can mix normal mode area lights for the exterior and portal mode ones for the interior, saving a lot of resources and increasing quality brought from the HDRI environment

This is called remapping or texture transfer or texture warp and it's basically the method i was explaining in my previous post. with the difference that i suggest to re-map everything you've got on a high res texture onto smaller textures spread across multiple UV tiles. But this isn't really supported in Blender, however there is a work around. This implies the use of a second UVSet (UVMap in Blender, within the MeshData panel, the triangle icon)

This also relates to my method: i do a very high res texture, free from worries of texture space, then remap it to the new UV map with the optimized UVs, organized in more UV tiles, because...

... which means that for each texture that Blender starts to bake, the whole process of calculation RESTARTS from SCRATCH, since Blender doesn't have lightmap caching features. That is why Light Path is better on one single image: you do your tests with low samples and for the properties that you actually intend to use (dumping the useless ones) to TEST the result on a smaller res texture, so it takes less time. You can then test how it looks in regard of lit areas, whether it looks correct or not, not caring of the quality in the first stage. At this point you shouldn't care if it's noisy, you should check how the lighting hits and bounces in general. When it looks ok, you can launch the final render on the single high res which will involve calculations only once. When all is good, texture transfer to the other UVSet that is split into multiple textures. This transfer doesn't take actual render and is very fast. This way, the same high res result you would have gotten in a long time to bake, shrinks down quite noticeably but with a far greater texture quality and a directly proportional weaker headache :D

 

Thanks for all that info. I will test out some things. I am extremely happy with my latest project but anything better (or FASTER) would be good. 

Happily I have the UV map input to smaller texture plane down. Spent a LONG time in Opensim with that. Don't use it all that often but it WONDERFUL when needed for sure.

Tried the HDRI (many tests) and was NOT happy with it. But I also don't like SL materials (specular and normal). So that's just me and certainly everyone should try HDRI  one time. 

I always do test bakes but sometimes along the way I need to remap and rebake. The more complex the project, the more likely this happens (to me).  So I sometimes have many "final renders". Much like "final drafts" that becomes "final draft 2, 3, 4" etc.  LOL. Simple things like furniture have minimal issues these days happily.

I will report back and hopefully this will be a thread that can be referenced for a long time. 

Share this post


Link to post
Share on other sites

And the answer is?

LIGHTING!!!!

I spent a couple of hours testing this morning and while changing the settings in the bake parameters didn't make a whole lot of difference, turning down the BOUNCE made a HUGE difference!!!!

Now to be fair the changes made in the Lightpath area may be more noticeable AFTER changing the lighting. 

I took a ton of screenshots along the way but for ease and clarity's sake I will keep it to a minimum and encourage others to make their own experiments.

First, this is what I chose to test on.  More dense of a mesh than I normally make but lots of details and both shiny and slightly bumpy surfaces.

5b214933643a9_bakingtestrender.JPG.c483310fa0d86b93b764236cae201de3.JPG

This was my lighting setup. I didn't make any changes in the original file and this was baked at 400 to a 1024

5b21498945ae6_lightingsetup.JPG.8fa89d1d6faf6d9db4c10e3e4e915429.JPG

Here is a close up of a tiny portion of my original bake.

5b2149ea5c268_bakingtestdefaults.thumb.JPG.3ea5b0a873bbefd8e61162c3564a6b71.JPG

Other tests gave similar results with the one with refractive TURNED OFF the worst.

HERE is the shot of the very same set up but with lighting turned down to 12 on each of the lamps. Note that I have also changed settings in the Lightpath here (including leaving refractive off), but the settings shown WITHOUT changing the lamps gave a texture very much like the default shown above.

5b214a5c0aa71_finaltestclose.JPG.688529bbcb22dab846c15f43ee6095dd.JPG

 

INWORLD comparison: 

You can't see TOO much difference from a distance but up close you can. AND it certainly seemed like the bake was shorter (I didn't time it :D). So with a few clicks to the lamps you obviously get a MUCH better bake.

Note below the change in color. This happened when I turned OFF refractive --- so something to think about there.

inworld.thumb.JPG.b4f0adf590e4c5845ceec03d312c2c88.JPG

Many thanks to @OptimoMaximo for all the help.

And while I have issues with change (while at the same time embracing the need) I can give HDRI another shot with the settings mentioned :D. 

 

Note that while it DID take us awhile, this seems to be the answer to the OP.  It does NOT solve the issue with pixelated diagonals -- at least not enough for my artistic sensibilities. 

Edited by Chic Aeon

Share this post


Link to post
Share on other sites

Please sign in to comment

You will be able to leave a comment after signing in



Sign In Now