Algovincian's NPR and Other Junk (2022)

algovincian

Thanks @3Diva. It's the set that deserves all the credit - not me. If you look at the Iray render, it's easy to imagine how the nooks & crannies where the dirt accumulates on the suit really help to accentuate the interior lines:

It's these interior lines (as opposed to the exterior lines that follow silhouettes) that are often hard to create and missing from NPR workflows. I know that I've spent time focusing on just this - it's one of the reasons that the analysis passes I render include an ambient occlusion pass.

- Greg

3Diva

algovincian said:

Thanks @3Diva. It's the set that deserves all the credit - not me. If you look at the Iray render, it's easy to imagine how the nooks & crannies where the dirt accumulates on the suit really help to accentuate the interior lines:

It's these interior lines (as opposed to the exterior lines that follow silhouettes) that are often hard to create and missing from NPR workflows. I know that I've spent time focusing on just this - it's one of the reasons that the analysis passes I render include an ambient occlusion pass.

- Greg

Oh I see! I wonder if darkening the ambient occlusion pass, or multiplying it, would create a similar look for images that don't have those baked-in "dirty" crevices for the algos to chew on more.

algovincian

3Diva said:

algovincian said:

Thanks @3Diva. It's the set that deserves all the credit - not me. If you look at the Iray render, it's easy to imagine how the nooks & crannies where the dirt accumulates on the suit really help to accentuate the interior lines:

It's these interior lines (as opposed to the exterior lines that follow silhouettes) that are often hard to create and missing from NPR workflows. I know that I've spent time focusing on just this - it's one of the reasons that the analysis passes I render include an ambient occlusion pass.

- Greg

Oh I see! I wonder if darkening the ambient occlusion pass, or multiplying it, would create a similar look for images that don't have those baked-in "dirty" crevices for the algos to chew on more.

Exactly! You absolutely can tweak the AO pass (by adjusting the levels, multiplying it, sharpening it with a large radius, applying any image filter you may have available, etc.) in an effort to get the algos to draw more/darker lines - and I do just that.

Let's take it a step further, though. Let's say you only want to do this in certain areas, and not in others. This can be accomplished through the use of one or more masks.

For example, let's say you look at the algo'd output and notice that the lines aren't prominent enough where an object's color is light. In this scenario, the color pass (another one of the analysis passes that get rendered out) can be used as a mask for an adjusted (darker) version of the AO pass. This would result in lines being more prominent where the object's color is light, but the lines would remain the same where the object's color is darker.

Things are a bit more complicated than that, though. One must take into account the fact that these changes to the algorithms are not only effecting the initial image where the lines in the algo'd output weren't prominent enough where an object's color is light. This will happen for *all* sets of analysis passes run through the algos - even ones where the lines were just fine as is.

This is where automation comes into play and why it's so important. Relatively large batches must be run in order to make sure that there aren't unintended consequences. It must be verfied that the changes made are good for the general case and not just in individual instances.

It's often wise to make changes like the one described in a small, incremental fashion. This is why I've always referred to the process as evolutionary. Sometimes changes that worked in an individual case don't work in the general case. When this happens, you can either revert the changes, or further refine where the lines are made more prominent through the use of additional masks/analysis passes.

And then there's the whole idea that these patterns can be recognized and decisions made by neural nets . . . but this post is long enough already lol.

I'm so glad that you're interested in this stuff and willing to have these conversations in public, @3Diva. I've gotten countless PMs through the years from lurkers and I'm sure there's even more who read without commenting.

Hopefully, this made some sort of sense. I hope that discussing these ideas in public may lead to some new ideas in peoples' minds, or at least spark some interest.

- Greg

3Diva

I find it very fascinating! While I'm not someone who can create algorithms I do learn a lot when you talk about this stuff. It totally gives me ideas of how I can use the same type of concept in my own work. Not through algos, since I'm not able to do that, but through filtering, render settings, image tweaks in Photoshop, and layering methods for creating different NPR styles. It really helps me to see where I can improve my own NPR work. Thank you for being so open to sharing your thought processes and the steps you take with your algos to get the images you're able to create! I love learning about this stuff!

algovincian

3Diva said:

I find it very fascinating! While I'm not someone who can create algorithms I do learn a lot when you talk about this stuff. It totally gives me ideas of how I can use the same type of concept in my own work. Not through algos, since I'm not able to do that, but through filtering, render settings, image tweaks in Photoshop, and layering methods for creating different NPR styles. It really helps me to see where I can improve my own NPR work. Thank you for being so open to sharing your thought processes and the steps you take with your algos to get the images you're able to create! I love learning about this stuff!

al·go·rithm

/ˈalɡəˌriT͟Həm/

noun

1. a process or set of rules to be followed in calculations or other problem-solving operations, especially by a computer.

 

Call me crazy, but it seems to me that you're creating "algorithms" with the shaders and pShop actions you've already designed, no?

Don't sell yourself short!

- Greg

 

3Diva

algovincian said:

3Diva said:

I find it very fascinating! While I'm not someone who can create algorithms I do learn a lot when you talk about this stuff. It totally gives me ideas of how I can use the same type of concept in my own work. Not through algos, since I'm not able to do that, but through filtering, render settings, image tweaks in Photoshop, and layering methods for creating different NPR styles. It really helps me to see where I can improve my own NPR work. Thank you for being so open to sharing your thought processes and the steps you take with your algos to get the images you're able to create! I love learning about this stuff!

al·go·rithm

/ˈalɡəˌriT͟Həm/

noun

1. a process or set of rules to be followed in calculations or other problem-solving operations, especially by a computer.

 

Call me crazy, but it seems to me that you're creating "algorithms" with the shaders and pShop actions you've already designed, no?

Don't sell yourself short!

- Greg

HAHA Well ...if you put it that way. Sure, I guess it's like I'm doing Algos a tiny bit. But it's likely more like a burger flipper at McDonald's (a job I actually had when I was a senior in high school) compared to a chef at a jacket required restaurant. Sure they both work to make food, but the similarities are pretty slim. hahah But thank you!

I am having fun though with my little burger flipping skills over here in the corner and watching you make such awesome things, Mr. Chef.  I AM learning and getting better with the NPR stuff. At least I hope so. lol