Best TOON SHADER

[Entangle]

I always thought and used the brazil 1.0 version. I like it allot more than the max5 one.

[ThirdEye]

Well the toonshaders in XSI are probably better than Brazil toons and Cebas finalToon is better too... Are there any other toonshaders out there? :shrug:

[ZeBoxx]

This is a pretty nice one :
http://www.algorithmic.com/eva_e.html

Re,
Richard Annema

[Mauritius]

Toon shading is a different task than edge detection.
The former is a shader executed by the renderer -- usually it simply indexes into a user defineable color ramp by viewer facing ratio.

Silhouette edge and feature detection are an entirely different task that take place after rendering -- you don't need a 3d renderer anymore to perform these.

So when you talk about toon shading, do you refer to the rather trivial shading or the more complex edge detection?

.mm

[ZeBoxx]

I think he's referring to which application is best at running a Posterize (toon shading) and Gradient Edge detection (edge detection *giggle*) effect, and compositing the two together - Photoshop, or The Gimp.

:rolleyes:

When asking which the best toon shader is, I would like to believe that one doesn't generally make a highlight technical difference between Renderman shader X which does toon shading, and Renderman shader Y which does edge rendering - since 'the best' would surely at least have both forms.

More applicable questions to determine 'the best' would be :

- Do you require vector format (Adobe Illustrator, Flash, etc.) output ?

- Do you want advanced line types (i.e. dash-dot-dash stuff) without a lot of fuss in setup ?

- Do you want hashing of the shaded surface without a lot of fuss in setup ?

Any of those three questions answered will severely cut down the playing field already.

Best regards,
Richard Annema

[Mauritius]

Allow me to offer you to talk to me directly, no need to use the third person ("Am I not mercyful?"). :buttrock:

I wasn't btw. talking about RMan at all.

A shader (in an arbitrary 3d app) usually can not do the outline properly -- it is used to shade the object. It has informations about the object it is attached to and the scene it resides in at last, not the image being rendered (pixels wise).
As a "real" outline must be centered on the resp. edge and not lie inside the object's shape. Hence it will likely always be created after the [toon] shading has happend

What I was particularly aiming at with my question is that you can have a great toon shader and use a completely different app to do the edges.
Or to give you a metaphor:
Stuff like lensflares, glows and the like is for some reason usually done in a compositing app. The fact that apps can do this directly (actually compositing while/after rendering) doesn't make people trade the greater flexibility of the former method against having to press two or three buttons less in case of the latter.

Most users are not aware that toon or npr rendering (not shading) involves two steps. I thus think it was quite right to ask these questions.

For example, you can build a great toon shader, in say Lightwave, and still use a third party app to render the outlines in a style Lightwave is not capable of.

So what would benefit this thread imho was a list of apps capable of toon shading, a list of third party toon shaders (including those made up by users for apps that support some sort of shade trees) and a list of apps that do the outline. And then, maybe, how each of these work together to get nice images.

Since there are hundreds of applications for toon and npr rendering, asking what the 'best' toon shader is, is a bit like asking what the best displacement/surface/whatever shader is.
There are gazillion possibilities; so any answer will be biased by what particular use case of such a shader the respondent had in mind.

This was the reason I tried to shift this discussion a bit from 'my toon shader can beat your toon shader' (to where it was already going, if you read Entangle's reply and ThirdEye_01's 2nd post) to the mere "technical" aspects, as you call it; or rather the process, it's implications and [creative] possibilities which is probably what people might benefit from most in this particular forum.

And judging from your reply -- while actually not understanding my post's intend you nevertheless asked the right questions (imo) -- at least for some aspects of npr.

I also remember that David Gould (who, among other things, wrote the Illustrate plug-in for Max) is subscribed to cgtalk. Maybe we should ask him to join this thread?

Cheers + peace,

.mm

[ZeBoxx]

Heya,

The Renderman examples were just that - examples ;)

Of course it's excellent to steer these discussions away from "my dad's stronger than your dad" type of back-and-forths, and towards the deeper underlying techniques and how to use them together is far more intrigueing.
Actually giving some examples might help, though.

For example, I mentioned the Gradient Edge detect filters - pretty common in most graphics editing applications.
However, they are incapable of discerning between e.g. object edges, or texture edges - objects overlapping themselves, or object overlapping other objects, etc. when fed just the basic image without any scene information.

Thus applications that can use additional scene information to determine where to create the 'outlines' thus present a better solution, but only if that is your need.

Without knowig the exact need we can indeed only theorize, discuss algorithms, etc.
Whether or not that answers the question "Which is the best cel shader out there now ?" is a different matter altogether :)

Either way - for a fairly good resource on NPR solutions out there (but not really up-to-date) :
http://www.red3d.com/cwr/npr/

And google will get people many more.

I welcome David to the discussion, and any insights you may have to offer yourself as well :)

Best regards,
Richard Annema

[ThirdEye]

well, i actually meant NPRs, my fault :hmm:

[LucentDreams]

Mauritus you are sadly mistake in one area for sure, the best line detection methods are not post effects but 3D shaders, notice I said 3D not 2D. Final toon uses this method for example which is how they are able to control thickness and opacity accourding to depth screen angle and such. any app with a luminacne channel and diffusion control can doa cel shaded effect which your right is different from line detection. One note the term cel shade has nothing to do with the fact that its a shader, but with the fact that you are shading the object as it would be in cel animation.

Any effect done as a post effect or compositor is pretty dismal

Goe easy on a guy who doen't know all the technicalities and lingo though, he just wants to know what people think is the best NPR renderer, not app specific just which one is nicest whether it be BESM or Final toon or whatever. SO far I'd have to say for an all round NPR renderer that I think http://www.toonshade.com/ for maya is the best with final toon a close second, another that has some nice results is twoplugins for max from infrografica, their NPRreyes and their Toon reyes or whatever its called.

[Mauritius]

Err, no I'm not mistaken.
I know pretty well how line detection for npr rendering works. I've written tools for actual productions to do this twice in my life.

To defend myself, however, I have to start talking about exactly that kind of stuff which imo should not show up in this thread.


Note that I will use the term 'silhouette' to refer to silhouettes, surface boundaries, creases (or feature edges), self intersections and curves defined on the surface itself.

Basically you analyze a discontinuity in a function that is calculated from 3d data. But the function itself is 2d already then.

To use common geek terms: both image space and object space silhouette detection algorithms basically do projection to a 2d plane of some sort.
If the data on this plane is already described as pixels or still points connected by lines resp. interpolated by curves doesn't matter much -- it's 2d in either case.
Of course the object space case of using silhoutte lines extracted as polylines or even curves directly from 3d geometry gives you higher quality. Nevertless, you'll always do some sort of projection to 2d to obtain these. And rendering the final strokes in 3d space is not very useful also -- they'll all pretty much lie on a 2d plane [each] already.

In either case the code calculating this data is not a shader. A shader defines the appearance of an object. In case of outlines, this meant the shader had not only to detect them but also to render them.
A typical outline will have to be drawn approx. 50% outside the actual shape of the object is belongs to as it is centered on the resp. silhouette -- that is: On other objects or the image background. One could do this in a [3d] shader, provided this shader had access to all other geometry in the scene and the pixels in the bg.
However, using this approach surely is much more involved than the results obtained by it could probably ever justify. I also imagine it would be slower, more prone to errors and more memory intensive than either using an image map with a pixel filter or projected slhouette curves that are drawn afterwards.

In RMan there are so called 'imager shaders' and you can indeed use them to do npr stuff and take their name as a start to get picky on terminology.
Nevertheless, they are not 3d shaders but only 2d shaders at last -- a distinction btw. that nobody makes as the term 'shader' is quite misleading here anyway.

Imager "shaders" are like programs that operate on the pixel grid (that is: unlike all other shaders they're highly dependant on the image resolution), so they're more like a bitmap filter one knows from PS, expressed in a special language. They are -- to a degree suiteable to performthe image space approach of silhouette detection.
As I said, a comp package with a decent scripting language, like e.g. Shake, or a plug-in is the better choice if flexibility is what you want.
Artists always want flexibility.

For a good overview of the current voguish methods of line detection for npr applications, I suggest this paper by Aaron Herzmann (http://www.mrl.nyu.edu/hertzmann/hertzmann-intro3d.pdf).


Cheers,

.mm

[LucentDreams]

man its evolved a bit since then, while te detection is still the same, the information is handled differently, hence why you can refract and reflect lines, cause they can now exisit n the 3d environment. thus while the detection methods are still the same the ooutput use a volume environment jsut like Hypervoxel shaders do. this is how final toon works. And while your experience is credible and would give you fair knowledge on the subject don't think that others aren't involved in development of NPR renderers as well.

[Mauritius]

Reflection and refaction is possible with all the methods I mentioned too. What you want is line width to be independent of the 3d scene -- like in a real 2d drawing. Hence these lines must be calculated according to image space. Using image maps with special information like normal or z-depth doesn't mean you can't have any geometry in front that refracts or reflects. The line detection algorithms in your compositing app will still create the right image.

"Hypervoxels" is a term used by Newtek to describe their ray marching volume renderer in Lightwave. I don't see the slightest connection with what finalToon does. Maybe you can enlighten me.

I furthermore don't see any image at the finalToon side that required a new technique.

And giving shading models a new name for marketing reasons ('temperature') doesn't help either.

Mind you that Cebas is full of such hot air ("Hyper-GI" -- lol!). Their finalRender was hyped to the olymp when initially released but it hadn't even the most basic features you need every day, like e.g. deformational motion blur.

Just an of marketing mumbo jumbo that misses the point: On the cebas front page an image is shown with 50 million polygons -- so what?
The renderer(s) I have used for the past seven years never had problems with such scenes but since finalRender doesn't even have level-of-detail support, the depictured scene will flicker like hell once you animate the camera.
And this is what interests me -- 99% of the people on this site do animation, not stills.

Cheers,

.mm

[LucentDreams]

Voxels are not just a LW thing, Their are many voxel shaders in otherapps. They are simply 3D volume shders meaning they need a environment Volumetracer to work, Final toon and a few other Line renders use this same volumetracer concept on their lines. yes their temerature and such is just hullabaloo and main reason I'm suing Finaltoon as my main example is because of NDA issues with simlar NPR renders. Heck I have even seen a volumtracer for tree that could render and entire tree from sphere or cylinder using a volumetracer. but for the lines to function totally in a 3D environment, meaning they can be affected by fog volumetrics refractions (I'd love to know how a posteffect gets refracted.) On a line calculated after the image how can you possibly reflect that line on another object. Can you show an example of one of either of these?

but now we are getting way off topic the whole thread was about what were peoples prefered NPR renderers. Least I actually offered some sugestions instead of correwcting the person and insulting their innocent misuse of a term that is commonly done within the industry anyways.

[Mauritius]

I looked at the reflect and refract examples and it seems you're right -- they're true 3d reflections/refractions. The line witdh of the refracted lines varies. However, this exactly what you do not want when npr rendering.

Neither does it render void what I said about the possibility to do this in post:

I assume you know what a normal pass is.
Imagine a normal pass being rendered of both the image with the pen and logo under the glass block and the two parallel port connectors on the mirror plane on the cebas site.

Now imagine that even though I render a normal pass, I consider any reflecting/refracting geometry (that is: reflecting/refracting geometry has its normals written out elsewhere). What happens is that the normal pass of the pen and the logo is refracted through the glass block. This way I get distorted crease edges of the pen and the logo. However, since I draw them after rendering, their thickness will be constant -- as if they had been drawn by a real person. The same goes for the reflection.
So not only is refraction and reflection possible in post, it also gives results superior to what is displayed at the cebas site and which match with what one expects from npr: Line shapes unaffected by the object space they originate from.

I also urge you to read my reply more carefully: I wrote "Hypervoxels", not "voxels" and this is indeed the special term used by Newtek marketing.

Imho it hasn't to do anything with voxels -- it's a raymarcher that accumulates density while marching the volume defined by a function like noise or turbulence -- rather than taking a limited resolution, pre-computed voxel grid and marching through that (which would justify the name). They only picked this term because "voxel" is more common to people than "raymarching". Additionally it contains the term "Hypertexture" as defined by Ken Perlin for his raymarching renderer.

Btw.: You can also make a tree from cylinders and cones by plain using displacement (not volume) shaders: http://www.edit.ne.jp/~katsu/rms_tree.htm.

Cheers,

.mm

[Khepri]

they're true 3d reflections/refractions. The line witdh of the refracted lines varies. However, this exactly what you do not want when npr rendering.


I totally agree on that(allthough it could be a desired effect in some cases) when I have time will do some tests with this...

nice discussion though(allthough its getting more and more offtopic).

[LucentDreams]

Originally posted by Mauritius
I assume you know what a normal pass is.
Imagine a normal pass being rendered of both the image with the pen and logo under the glass block and the two parallel port connectors on the mirror plane on the cebas site.

Now imagine that even though I render a normal pass, I consider any reflecting/refracting geometry (that is: reflecting/refracting geometry has its normals written out elsewhere). What happens is that the normal pass of the pen and the logo is refracted through the glass block. This way I get distorted crease edges of the pen and the logo. However, since I draw them after rendering, their thickness will be constant -- as if they had been drawn by a real person. The same goes for the reflection.
So not only is refraction and reflection possible in post, it also gives results superior to what is displayed at the cebas site and which match with what one expects from npr: Line shapes unaffected by the object space they originate from.


Btw.: You can also make a tree from cylinders and cones by plain using displacement (not volume) shaders: http://www.edit.ne.jp/~katsu/rms_tree.htm.

Cheers,

.mm


Why would it be wrong if the line in the refraction isn't hand drawn, seems to me when I learned how to do Refracted effects of transperency in classical animation the refracted line wasn't hand drawn. Reflections are though since the reflection 90% of the time must be drawn (flipping usually doesn't work unless the composition is flat which is boring) Prime example of refraction though would be the little mermaid where the opening sequence and a few sequences under water through out have a refracted effect now done with 2D displacements, but originally done by actually shooting those section onder rippled glass. One of the rare examples of reflectections that weren't hand drawn (or paintedd rather) is Disneys the old mill where there is a reflection in a pond of the mill done using rippled glass a matte and second exposure. So saying that reflected and refracted ines aren't intended isn't true. I know what you mean, and often you wouldn't want that depending on the style.




The link to the tree thing doesn't work but I think I know what its a sample of either micropoly displacement or perpixel displacement, which are capabilities unfortunately not available to most renderers yet. But yes your right, my point on that was that voxels are quite flexible and the 3D lines used in Finaltoon work using a volumetracer just like voxels

[flingster]

sorry to interupt here...but if anybody was interested in the
illustrate! link is:
http://www.davidgould.com/


(oh and ThirdEye_01 he ain't gonna make it for cinema :thumbsdow )

[cptvideo]

/me replies a week late :)

Interesting thread.

The truth is that there is no rules about how certain things are handled in 2D toon work -- especially when it comes to handling special effects like reflections/refractions/etc. For every example of, "you want to do X this way," I can find 3 examples where it was done differently and done well. That's part of the magic of the art :) Learn the different techniques, be informed by them and know when they work best stylistically, then get creative.

As Zeboxx stated - if you need specific features, do tests and make sure the system you use meets your needs. If you're not too particular, explore. You may stumble across some invaluable and interesting feature you hadn't even considered. If you're trying to match a specific style, then study that style so that you understand the 'rules' of that particular toon and then find a toon-shader-system-thing that'll get you there.

<shameless plug>
On that note, I started putting together a Tech Focus for BrazilToon a while ago and never finished it (some day I'll find the time to finish all this). The idea was to do it up as an overview of some of the techical features in BrazilToon. It's still a WIP and I'll probably be adding new images in the future, but for those interested, you can check it out here:

http://www.splutterfish.com/sf/gallery_index.php?cat_id=6
</shameless plug>

Enjoy :) sk....

[MaxAlbert]

Escuse me if I am interupting anything here, but when I started reading this thread it was too know which toon shader I could use and not to have a discusion on technicals aspects. Don't get me wrong, I respect all this knowledge you have, but I'm just an artist that want is tool too work. I used the XSI toon shader and I don't like it, I would like to know which other one I could try that might give a better result.

[Fex]

@MaxAlbert: don't know what you dislike about Xsi toon shader
in my opinion it has the cleanest and fastest edge detection
(the ink layer) for paint layer you have infinite possibilities
to built your own shading style... :)

if you want something totaly different look at this!!

http://www.algorithmic.com/index_e.html

[Khepri]

MaxAlbert:


I think "better result" would be achieved through the artist rather than the software they use.

there a couple of toonrenderers/shaders out there and each has its strength. some are post effects(finaltoon for instance), and other are handled directly through the rendering process(Mental Ray for instance).

but there is really only one thing that matters and that is how the artist uses his creativity to achieve a look by using the software as an advantage rather than using it as a limitation.

MR has very nice toonshading, and like like Fex said: its very versitaile(!?) especially when you layer stuff on top of eachother.

the beauty of toonshading is that the possibilities are endless.
that also might be a limitation because you can easely "go over the top" with it.... but still....


:D

[derelict]

.... after two freaking pages and no answer yet. The eloquence is fantastic to woo a girl or two but the point of this tread is... which is the best tooner? So which is it?

[LucentDreams]

I think the answers were quite clear, there isn't a best one, they all have very distint differences. As well there is a difference between a renderer and a shader and that outline detection is different than cel shading.

[MaxAlbert]

Thanks a lot for anwsering to my question in a straight and forward way. Maybe you r right and XSI toonshader is not that bad and that I m just too impatient so I'll go back to work and try harder.

I'll try the thing you'r talking about fex.

[derelict]

Originally posted by Kaiskai
I think the answers were quite clear, there isn't a best one, they all have very distint differences. As well there is a difference between a renderer and a shader and that outline detection is different than cel shading.

There you have it! Thanks dude:)

[MaxAlbert]

It would be nice of you if you could tell me which other one I can use whit XSI, and in the same time which one for maya & 3D max, so I could try some of them.

[cptvideo]

Best for max, IMO, is BrazilToon. It provides the most control and reliability without sacrificing any of the normal renderer functionality and workflow. It doesn't provide vector output and some styles are difficult to achieve (drawing off the edge of the object, for example). But when it comes to generating film, television, stylized and technical types of imagery, it rocks, is easy to use, fast, and genuinly fun to work with.

Blur studio just produced a stunning 2k stereo ride film for Paramount theme parks and Nickelodeon's "SpongeBob SquarePants" with Brazil Toon that is truly amazing ( http://www.kingsdominion.com/corpinfo_pressSB3D.jsp ). The images look handdrawn and the overall direction and movement is right on. Blur has done a great job on the work and I highly recommend checking it out if you can get to where it's showing.

[MaxAlbert]

Thanks, I'll try that

[Nando]

I agree Brazil Toon Rocks
Ive been doing some test and ive fallen in luv with it
a vector output would be nice to intergrate with Toonboom or US animation or just plain good ol flash ;)

That Eva page is very cool.
I luv NPR keep it coming


:buttrock:

[ZeBoxx]

Heya,

Do keep in mind that a vector renderer is something quite different from a raster renderer.
I wouldn't expect vector rendering of toon from Brazil r/s in the near to mid-term future - but one never knows.

[dixit13]

hey whats this all about , would'nt you guyz talk in simple english , all of this is going over my head, keep it simple so novic like me can learn from the experts.........:shame:

[Khepri]

a novice in english or 3d?

[tcomputerchip]

Though, I have not use Brazil Toon.

I felt the 3dsmax 5.0 toon shader did not produce enough control. When I purchased finalRender, finalToon came with it.

I love finalToon and it's very easy to use. The only thing that sucks is the guy who does the training videos. He takes forever to explain things. If you can put up with that, I would recommend finalRender if you have a small budget.

Good luck! Hope you find one you like!

[CGTalk Moderation]

This thread has been automatically closed as it remained inactive for 12 months. If you wish to continue the discussion, please create a new thread in the appropriate forum.