I did this out of my own curiousity more than anything. I've found that some commercial vendors have somewhere between 10 to 12 different nodes setup in the material room.  I'm sure all of that can't help render times. So I set out to see if I could get similiar results using no more than 3 nodes.. not counting actual image maps. the three I used were blinn, phong , and gloss.
For lighting I used a light set from saint fox.. I think we can all agree its a good setup. 
first image is just an image map with no nodes.
second image is three nodes that I setup.
third image is a typical commercial node setup.

anyway my own conclusion is that you really dont need all the extra nodes. while there are some differences here, i'm sure someone with more knowledge than I in the mat room could come very close using minimal nodes.

I'd say that the differences between the 'commerical style' and your limited node set are palpable - the commercial style is far superior.

Word of note - since shader nodes are procedural they do not occupy much memory and don't hit the render times too much.  They probably can have an impact here, but yous plays yous pays.  That is, the better the render results you desire, the longer the render will take.  This is 3D CG 101 - same for Maya Unlimited, same for Wings3D.

BagginsBill has done some node setups with hundreds of nodes and never heard that it took days to render.  Still, if the results are fantastic it should be worth the wait.  You usually can't get them with simple illumination models and texture maps.  I'll let him chime in with his thoughts.

I have to agree with kuroyume here; especially in the latter trio of images, the differences are rather profound.

Hehee.. good one, Fixer!

Render nodes don't eat much memory for storage, but they do impact render time, and memory use.

If you really want to increase render times, remove all maps, and textures, and render the bare models without smoothing on as the lowest possible reender levels.

Course your render won't be much better then a Poser 3 release then.

In higher end applications, there is far MORE use of nodes (other then BagginsBill's Mathmatica stuff). The whole point is to use the tech available to create the best possible renders, not dumb them down for speed's sake.

That said, there are alternative node setup possibilites that might speed things up, but instead of tearing all your prodiucts materials apart, and trying to rebuild them and reinvent the wheel,  it's a much better investment of time and energy to just buy a newer faster computer.

Some of what you're seeing may be due to the render settings in general. What settings are you using for the actual render?

btw.. I'm not trying to reinvent the wheel, but I am trying to learn how it works.

raytracing is not on.  default settings in p6 firefly.

The differences may be more clearly seen closer to 'final' mode. It will also reduce the clumped look of the lashes to some extent, which is a plus. Raytracing often really helps, also.

Sorry, but the original still looks better to me. Take a look at some of BagginsBill's shaders, and then tell me what you are seeing is too complex. Sometimes he has almost 100 shader nodes hooked up to achieve a specific result, yet had outstanding rendering times. And the number of nodes has little impact on the final render, in many cases. Oversimplified it's somewhat like math:

Which looks more complex, and like it would take you longer?

1+1+1+1+1+1+2+2=10
or
3+3+3+1=10

Clearly, the first one appears like more work, but
yet they both achieve the same result, and both take a human mere seconds to compute... and a computer can handle millions of these in a second.

The shaders add specific effects to the pixels, and without the shaders, you are loosing something, even if it's hard to pinpoint exactly.

Some nodes can replace other nodes, but then you aren't saving anything in render times at all.

Yes, you can get rid of nodes, and save a little render time, but you are then also loosing rendering properties.

Just like turning down render quality, and turning it up.. turn it down, you save render time, but loose quality. Turn it up, you loose speed, but get a better quality render.

And most content creators tear each other's material settings apart, trying to learn how to achieve specific effects. I know I do all the time, and I've even bought products just tio look at the material settings, and figure out how they did what they did.

A great example, is my metalflake textures I included with my Funcat.. a number of people have achieved different material node setups for doing metal flake textures, and each has it's advantages, and disadvantages.. some sacrifice render times for a more realistic effect, and others (like mine) sacrifice some realism to decrease render time.

Since some people have been pouring over matyerial settings for a number of years now, and sharing their knowledge in the forums, I doubtr you'll come up with a "magical" simple node setup to achieve the same results without a lot of work involved.

I'd be interested in seeing what you think is a complex node setup, and having one of the experts like BagginsBill try to tackle the same result with fewer nodes, the same quality, and faster render times.

I know my metalflake textures took me a few weeks to perfect, and that was JUST playing with them over and over again, seeing what worked, and what didn't.

Nver mind 'up to 100 nodes'... some of bagginsbill's loom shaders have almost 150 - and they still render at a reasonable rate. Here's a pic I did recently:

Now, it's not anything particularly special - I was just messing about with a few bits I'd recently discovered for Matmatic. But, nodewise, there's a LOT of stuff here. The kilt and shirt are both totally procedural, each with over 120 nodes. The metal buckles on the boots and belts are all procedural, but only require a few nodes to create the desired effect. And all the belt and baldric leather is also procedural, although I forgot to clock how many nodes were in each of those materials. The only map-based pieces are the axe, the hair, the skin (which I plan to change to procedural) and the boots. This was render with my own preferred settings (not any of the Poser defaults), using raytracing and IBL for the lighting. Total render time was approximately 6 minutes. Whilst that's not particularly fast, on the clunker I'm running my stuff on at the moment, that's not significantly different to simply using maps, or using less complicated node set-ups.

JonTheCelt

More nodes do make a big impact cause you are calculating stuff rather then just reading it from a map. But firefly itself is so slow and inefficient in handling maps and geometry that the difference extra nodes make isnt very noticible. So you should go for what looks best and not worry about rendertimes or node counts that much IMO. If speed is what you want, then there are a lot better renderers you can use then firefly.

Also note that the large number of nodes people use is not because the creators want it that way. More like they have no choice in the matter. A lot of the nodes are simple math nodes that are scaling stuff, isolating stuff, adding noise or reacting to scene data like light angles or intensity and so on. There is no way to replace them with simpler nodes, because they are already working at a very low level. If you replace them, you lose the detail they are adding to the tree.

to start off with I was only referring to skin shading and not procedurals for other items. Evidently I should have clarified that. If I have to create 100 to 150 nodes for a procedural texture then I'm guessing creating an image map would be less work. And for anyone thinking procedurals can't or dont slow things down is mistaken. Floating point math is cpu and memory intensive.

Everybody's making good points and I hardly need to add anything.

Nevertheless...

The first three all look terrible to me. Why the purple lighting?

Of the second three, the commercial one looks far better, IMO, but I can do better.

All this talk of "nodes do make a big impact" and "can't help render times" seems based on no actual stated measurements. Did you consider that you should actually time these renders and make an argument about how much one more node costs?

On a typical high-quality render my measurements show that most nodes are very fast indeed. 

Let's do an extreme test. Given 1000 by 1000 (a one-million-pixel) image with Min Shading Rate at 1 every node will be evaluated exactly 1 million times. Placing a one-sided square to fill the render and changing the shader I can measure the cost of any set of nodes. 

Of course this is overkill, since you rarely fill the image with any single shader, but let's measure for the sake of argument.

Some representative examples: (times are on my machine which is Pentium 4 3Ghz, using P6)

Basic Render (Diffuse_Value = 0, Specular_Value = 0 - no nodes): 19 seconds
Diffuse Node: +1 second
Glossy Node: +1 second
Diffuse+Specular (using Color_Math to add them): +2 seconds
100 Add nodes: +17 seconds
Diffuse+Specular+100 Color_Math:Add Nodes: +22 seconds
10 Fractal_Sum added together with 9 Color_Math:Add: +12 seconds
My most complex skin shader (freckles, moles, splotches, bump, sss, 24 nodes): +18 seconds
My Ultra Basic Skin Shader - UBSS (freckles, moles, splotches, bump, sss, 15 nodes): +15 seconds
Glenn Plaid  with 6 layers of fibers (283 nodes): +88 seconds

So some nodes are really cheap (math nodes) at approximately .17 to .2 seconds per megapixel. This is including two multiplies on each one as I didn't use "1" as the coefficients, but .6 and .4.

Basic lighting nodes are about 1 second per megapixel. This is impressive considering they must do cosines and powers.

Noise-based nodes are also about 1 second per megapixel. Also impressive, since an 8-octave Fractal Sum is a lot of math.

A good skin shader costs 15 seconds per megapixel.

A great skin shader costs 18 seconds per megapixel.

I don't really get what the issue is. 

I had to reduce the image quality to get it to fit in Rendo's 200K limit. The original is better.

Be sure to click to see it at full size.

onnetz,  There are clearly differences in the pictures you posted. Some people will find those differences acceptable, some will not.
A rule of thumb that I use in setting up textures is hopw visible the object is in the scene. If I'm going to do a close up portrait, I'll try and use as many tricks as I know and the computer is capable of to get the best image. Subtle differences do make a difference in the image.

If I'm doing a group scene, or something where the face is not very visible, then I will go with much simpler setup.

A little snippet from art theory, as best as I can remember it...  Human perception is very highly developed (evolved) and trained to pick up very subtle difference in facial expressions, textures, vibrancy etc.... This is what makes succesful portraits of people one of the harder things to do.

When it comes to seeing those differences in people (even other races, which is one of the causes for some sterotyping) that we are not used to seeing often, or animals, this perception for subtleties is not quite as strong. 

Same principle ends up applying in facial renders, you may come relatively close to realistic with less sophisticated shaders, but the refined eyes will pick it up. 
The bottom line remains, if the quality is satisfactory for your overall scene, then there's no reason to not use it.

One thing that I often see in scenes, and it tends to cause a visual conflict for me is when there's items of varying levels of quality. One example may be a photographic background with a more cartoon like character, or a super realistic character in a plastic prop looking scene. It sort of disturbs the harmony in the image, and causes people to not like it as much as they could.

well, just my 2c :)

"third image is a typical commercial node setup."

I'm curious - is that the original Daz V4 shader? That was written by Face_off who is the undisputed king of skin shaders in Poserland. The funny thing is he dumbed it down and used techniques from one of my "simplified" skin shaders, because he was not allowed to include light-position-dependent SSS features in the shader. That's what all his "real skin" products do - they include a Python script to load your current light information into the shader, because the shader can't get that info from nodes.

So he used the Diffuse node to drive a ColorRamp instead. I don't know if he got that from me or independently came up with it, but I was the first to post that technique here at Renderosity.

I agree that the DAZ V4 shader can be improved upon, but generally not by using fewer nodes. Nor do I think its too slow.

Of course if that's not the shader you were referring to, then all I said here is pointless. :biggrin:

Thanks, Bagginbill for chiming in. You are the real node expert, so I trust your evaluations 100%.

BTW, what are your system specs? I'm just curious if you have a really fast machine for those numbers, or something a little older.

(Back to your discussion.........I just wanted to take a break.....) :biggrin:

LOL I had forgotten you work in autocad too :)
I have a terrain model I'm processing that has 1.2 milion nodes in it.  
Anyone wanna buy a node?
sowwy folks, I'm easily amused! Been staring at too many nodes!
[Enough Kitty, ENOUGH!]

Yeah, that node was a free sample.  If anyone wants more nodes, then I'll print out a 30X42 sheet of nothing but nodes for them.  And I'll only charge $1.00 per node -- cheap!

Bagginsbill, thanks for the timings.

Do you think the addition of shadows and/or raytracing would increase node calculation exponentially?

I'm guessing raytrace would but I don't think shadows would.
Hmm. What about AO?
Off to do some experimentation of my own...

I think the bottom line is what a person wants to achieve.  For portrait work you want as realistic skin tones as possible--for distance you wouldn't need that.   I for the life of me can't understand the Poser material room.  I tend to use everything out of the box and if I need to make a change I use the simple material dialogue.  I agree that it is overly complex and I'm not a rocket scientist.  I wish there could simply be more presets that could be applied on the fly and not the tangle of connections and nodes that we currently have to deal with.

BTW, I've found that working with IBLs have more of an impact on the realism of my character than any other factor.  But even there it is a matter of trial and error.

Are we to infer that more nodes = greater realism?  Are 100 nodes better than 50?  And would 500 nodes be better than 100?  Or do we eventually reach a point of diminishing returns, where adding more nodes accomplishes nothing?  Say -- to hazard a guess -- at the 150 node break-point for skin?

Having glanced briefly into the innards of some of bagginsbill's matmatic-generated shaders (and then regaining consciousness!), and reading through the txt file which generates them, I would say that most of the VERY large number of nodes used are to try and get Poser to do various mathematical calculations. Some of them can be written fairly simply in the text file, using trigonometrical functions and the like - but since there is no 'cos' function within Poser, then you have to go the long way round to calculate it - which requires more nodes (I was going to say there was no 'sin' node in Poser, but I just knew that I'd be misquoted and ridiculed for using such a phrase! ) . So in such cases, the largr number of nodes is needed, in order to create the right pattern - try to reduce the number of nodes, and the whole shader collapses.

However, I also think it's possible to have too many nodes, if all those nodes are doing is adding increasingly needless levels of detail to something. As others have pointed out, there's no point in using a complex skin shader on every single member of the audience in a darkened auditorium, if the focuse of your image is the single perofrmer onstage!

So in answer to xenophonz' last question, more nodes /= greater realism. More nodes may simply = the necessary steps to complete the calculation - or may, as the original poster suggested simply be overkill in a given situation. I don't think you can put a definitive answer on it, because as with all things we're doing here, the end results are purely subjective and depend on one major criterion: Do I, as an artist, like the finished result?

JonTheCelt

The simple answer is if it takes you 100 nodes to achive the look that you want then 99 nodes are not enough and 101 nodes are too many.

...then you get a bad node in there, and it goofs up everything, instead of a few freckles, you get a few pimples.

Good answers, all.

*BTW, what are your system specs? I'm just curious if you have a really fast machine for those numbers, or something a little older.
*My computer is 4 years old 
Pentium 4 3 Ghz processor
1 GB of RAM
80 GB hard drive

It is not really very fast at all.

*Do you think the addition of shadows and/or raytracing would increase node calculation exponentially?
*No. Raytraced shadows would slow down a Diffuse or other lighting node but not by a lot.

Depthmapped shadows don't slow the nodes down at all. They take time to pre-calculate, but that has no impact on the time your shader contributes to the render.

Ray-trace nodes (Reflect, Refract, AO) are much slower than all the others, but when you need them, removing them is simply not an option. Skin shaders don't need those, except maybe AO, and that one is pretty fast for the most part.

If you're doing skin, and you include ZERO specular type nodes, then adding one will produce greater realism. Adding another one might, if you use it right (I always use 2).

Adding a third might be necessary but you're beyond the noticeable improvement. Certainly adding four or more will not improve realism.

I made this picture the other day, explaining how a shader is an approximation. You can NEVER take into account every realistic effect. However, if the effect you're trying to produce is one that follows that BLACK curve, then the RED (one node) or GREEN (one node) straight lines are fine if all the shaders points in your scene are being asked to operate in the range that is where the RED line is close to the black curve, or the green line is close to the black curve. 

But what if BOTH conditions occur in your scene - i.e. tremendous variation in light reaching the surface. Then you need more than one node. You'd at least have to do a quadratic formula to get closer to that black curve, and that would require two nodes. But what if the realism was still not good enough - you may need 10 or 12 nodes to produce that response curve in your shader.

Clever use of the more complex nodes can reduce the node count. For example, in my Ultra Basic Skin Shader I found a way to abuse Specular and Diffuse, in conjunction with a Blender and a Subtract, to do a pretty good subsurface scattering effect. Is it as good as the 20 node version face_off uses? Nope - not as good.

But that was the point of UBSS - if 20 nodes makes your head hurt, use this 4 node version. And suffer the small loss in realism, of course.

Now when I made UBSS, I was specifically trying to get it to work in a broad range of circumstances. With or without IBL, with only white lights, or colored ones too, only on red skin or how about alien green or blue? The fewer constraints you put on the solution, the more complex it has to be.

Now you have to understand that in order to discover a 4-node solution that works "pretty good" for most scenarios, it took me over 1000 hours of experimenting in the material room. Now of course each render I do with UBSS probably saves 15 to 60 seconds (at very high render settings) per render. Let me see, did I save myself any time? 1000 hours at roughly 30 seconds saved per render means I need to do 120,000 renders before I break even. WHAT?!? So, my point is, it was utterly stupid for me to try to find the smallest number of nodes in a skin shader, IN ORDER TO SAVE TIME. It didn't save me any time at all. And, lest you think I am stupid, that was not why I did it. I studied skin shaders day after day after day (not just my own, BTW) so that I could puzzle out THE ESSENCE of the shaders job. As a result, I know way more about how skin works, and I also know way more about how to use the material room. So I'm not unhappy.

And the rest of you can enjoy the very fast and "good enough" UBSS, which in its simplest form is only 4 nodes. But I prefer the somewhat better version that includes freckles, moles, etc.

Consider: My latest shader (almost ready to share) is my Apollo Maximus skin shader. There are 5 versions - some with freckles, some without, some with moles, etc. Depending on what you ask it to do, it will make anywhere from 4 to 15 nodes. Do you choose based on node count? NO! Choose based on features. Do you choose based on 15 seconds versus 30 seconds? NO! Choose based on the effect you want.

The number one reason renders take forever is too many lights. 

Every time I load some "great" lightset, and it has 6 to 10 lights, I get ticked off at how long it takes to render. And on top of that the results are silly - the lighting looks completely arbitrary and fake.

For a portrait like this I use 3, an IBL, a key light (spot or infinite) and a rim light (spot). My portrait renders always take less than a minute.

Thank you for taking the time to write out that very detailed response, bagginsbill.  I'll need to consider it in-depth when I'm not at the office.  But it's certainly appreciated.

In my shader, this was accomplished simply by setting the "stain" color to RGB 200, 240, 150. All the subsurface scattering and other effects took this into account.

Quote - Ultra Basic Skin Shader (UBSS).

Colour me stupid, but what is the Ultra Basic Skin shader? Your skin settings are superb, and just what I'm looking for.

Attached Link: Matmatic Ultra Basic Skin Shader tutorial at RDNA

BagginsBill...those Apollo's look fantastic!

What did you do to get that displacement map to 'pop' so much? I love it!

Ahhhh, now that's an amazingly informative link ;) Thank you very much bagginsbill!