Alexander Kucera

Well, guess what? As I already thought it would happen, some people don’t seem to like the conclusions of my test. Zealots are so easy to predict and so hard to please.

####“You cannot do it like this. Testing the renderers with a simple teapot in the scene is totally unfair!”

Well, huh? How can a teapot be unfair? It is a simple geometry for crying out loud! If I had used a sphere or a torus (something not obviously produced by Pixar) it wouldn’t have made a difference.

####“You didn’t use the right MotionBlur Settings, that is unfair!”

Well, true, MentalRay also has something called Rapid Motion Blur. But that sucks too. I didn’t include it for two reasons. (1) It sucks. Or in a longer version: Every time I tried it, I got major render artifacts, so I just didn’t consider it for this test. (2) It sucks. Longer version: It is still way slower then Renderman for Maya. Well, my bad, I should have included it for sake of comparison.

####“You didn’t tweak MentalRay right, there is way more to tweak to make it faster!”

Well, it wasn’t part of this test to find out which renderer I can waste the most time with. In my business it is important to get results fast. Every minute spent at tweaking render times is nearly the same as rendering a slow renderer. Because if I spent two hours at tweaking or loose two hours due to bad render settings doesn’t matter. In the commercial business I have to deliver. Fast.

In my tests there was nearly no tweaking involved for Renderman (which I knew nothing about before these tests), but quite a bit to make MentalRay to look even that good (which I have worked with for quite some time now). In my book that means MentalRay is slower (and yes slower means also that it has more knobs to tweak it and loose time in the process).

####“Sure Renderman is great, but Renderman for Maya is a completely different product. You can’t compare that!”

Well, yes and no. It is about RfM, but the only difference to PRman is the interface and what you can do with it. The rendering engine is 100% the same. So what goes for one, goes for the other, too. With the addition that PRman is more configurable. Same goes for MentalRay for Maya. It is an interface for the MentalRay renderer. As such you have a small subset of what is possible in a point and click way. I tested this point and click way. Not any nifty shaders and special moves (which you can do with both RfM and MentalRay for Maya by the way).

Anyway, the results are the same, no matter how you turn it. MentalRay has it’s niche in the raytracing hyper-realism section, where there is no need for motion blur or fast displacements. But as far as good production value goes for day to day projects in a commercial environment (and for movies too as far as the general knowledge goes) Renderman is still superior. It is fast and easy to tweak (especially the RfM plugin) and gives fast and superior results.

Anyone interested can read the [Renderman vs MentalRay discussion on cgTalk] 1.

###Following are the comments from the old blog

virgil Says:

man, how can you render in RfM without errors and hours of tweaking? this is one thing I’m strugling with right now, that I keep having all kinds of weird issues, errors, and such, and I manage to debug it in the end, but at the cost of losing many-a-neuron one other question, as I actually started a quest of trying to figure out the mistery behind the allmighty render-passes in RfM. how do you… do it? that tab in the render settings is alien technology for me right now, any tutorials, advices, anything? I’d surely most apreciate that. thanks in advance.

AlexK Says:

I actually don’t know. I just use it. Works fine.

Concerning the Render Passes…well…same thing here. I just use em. Maybe the workflow was made by persons with a brain wiring similar to mine and not yours?

But I will try to find the time to write a little about the passes and how they are set up (at least how I do it when I need them).

virgil Says:

thanks alex, could you give me a simple example of how you’d set up a shadow pass? I just need to render my shadows separately, that’s all. I haven’t touched passes until now… and they seem very simple, but amazingly ilogical, meaning that whatever I do, more-or-less-nothing works now don’t tell me I’m stupid, please, because you have to admit there are weird things in rfm, like tweaking that alpha gain when rendering displacement… could they make it any more obscure? so what’s the secret behind rendering these passes anyway? it looks to me like I can only set up one pass… and it renders that pass, like the shadow pass… plus a beauty pass. I haven’t tried anything else yet. I presume it can render a diffuse pass plus the beauty pass, hehehe. because each time I try adding a different pass, it replaces my previous work… I can’t see that it’s adding that pass to the sandwich, it’s replacing the earlier pass, that’s all. so I’m always left with only one pass. but how do you render 10 passes, and what if you don’t want that beauty pass?

and last thing would be that I worked really hard to get the shadow pass to work, and I don’t know how I did it. trying to emulate the process with a different file didn’t work. I really did the same thing… I think… like it works just when it feels like, hehehe. yeah, I know, I know, it’s simply my fault - I don’t understand it well but then another fact is that with RAT, and RfM as well, an insignificant little detail that you miss can result in complete disaster at rendertime. very sensitive. and I get nothing for free… well, with RfM I do RfM is much friendlier, so instead of tweaking things for days… I only need hours but errors seem to be inevitable. all the time. I’m not a rendering specialist and I have no one here to set it up for me, so I do all this stuff by myself.

Finally I found the time to do my little Renderman - MentalRay comparison. I used Maya 6.5 on a Mac OS X - which comes with MentalRay 3.4.1.5 and the Pixar Renderman for Maya plugin version 1.2 evaluation.

As test scenes, I used the tutorial scenes that come with the Renderman for Maya plugin. I am aware, that this might be a bit unfair to MentalRay, but my reasoning is that the features I tested should work in both renderers and the test scenes where not specifially pro Renderman anyway. I also tweaked the scenes to get the best out of Renderman and MentalRay, so as to not penalize either one.

The tested features where:

• Displacement
• Motion Blur
• Instances (as a side effect of the Motion Blur test)
• Depth of Field
• Fur and soft shadows
• Reflections
• Global Illumination

For a side by side comparison of rendered output and rendertimes please download the following PDF → (link: rmanvsmr_forWeb.pdf text: Renderman vs MentalRay comparison)

Update: MasterZap over at cgTalk pointed me to a few shortcomings in my tests. First he is of the opinion, that I should have used MentalRay’s “Rapid Motion Blur”. Every time I tested this feature, I either got artefacts or it was not significantly faster, so I just didn’t use it here. Second, he points out that MentalRay has something similar to Renderman’s DeepShadow Maps. I forgot about them completely. My bad.

###Displacement

Both renderers can handle really fine displacement really well. However, Renderman wins this contest hands down, as it does displacements with virtually no extra overhead. The surfaces get subdivided into Micropolygons anyway, so the renderer already has the neccessary geometry for displacment. MentalRay however doesn’t handle the large masses of geometry as well and it is much more hassle to tweak it (one has 4 different knobs per object to fiddle around with in MentalRay).

###Motion Blur…

Now Motion Blur is MentalRays (and most of the other renderers out there) Achilles heel. They just don’t cut it. Slow, grainy, memory intensive. That is the common knowledge about 3D Motion Blur.

This is definitely not true for Renderman. Man this renderer flies through it with a quality that is amazing! And the really nice thing about it is, that depending on the scene, Renderman gets even faster with Motion Blur. That is due to the fact, that one is able to give Renderman a second quality parameter for Motion Blur. So you just lower the quality where you won’t see it anyway (meaning in the blurry areas).

MentalRay on the other hand…just sucks…big time. No wonder people have such an expression of panic in their eyes, whenever rendered Motion Blur is the topic. If I didn’t know it better, I would not touch 3D Motion Blur either.

###…and Instances Rendering the Motion Blur scene without blur, a severe bug in the Pixar plugin is revealed. It renders instances way slower then it should. That is a known bug and the big brother Pixar Renderman doesn’t suffer from this.

Nevertheless, at the current time MentalRay wins this category easily and superfast.

###Depth of Field

Both engines work nearly the same here. Nothing special to report. Fast and nice.

###Fur and soft shadows

Well, this one I would say is a matter of personal taste. Renderman renders the fur best with DeepShadowMaps, which is like a shadow map, but with more bells and whistles. MentalRay doesn’t have DeepShadows and has to use Raytracing. That means, MentalRay shadows are always sharp and if you want them soft, you need more samples (=rendertime). The situation is the other way round with Renderman and DeepShadows. The default shadows look a bit blurry, but render fast and to increase the sharpness you have to increase the ShadowMap size and therefor increase rendertime.

I’d say it’s a draw. If you like sharp shadows, go with MentalRay fur. If you like softer shadows (and I do, as real world shadows are never really sharp) go with Renderman.

###Reflections

Now we are entering the domain of MentalRay. It was built from the start as a Raytracing engine. And we can see that instantly in the rendertimes. Renderman however lived happily without Raytracing for ten years (or more) and has just recently (with version 11 I think) got the ability to trace rays.

###Global Illumination

Same here. MentalRay wins hands down. It feels right at home with tracing stuff.

###…but…

Remember the Motion Blur advantage? Well, it is even more evident with Raytracing. MentalRay may rule with Raytracing, but it is unbelieveably slow with Motion Blur. And it gets rediculous with Raytracing/Global Illumination and Motion Blur at the same time.

Renderman however, still slow, gets a big speed boost from lowering the quality in the blurred regions. Again fantastic!

###Fazit

So as a conclusion I would say, that both renderers have their areas where they shine and it would be a wise thing to know when to use which. All sympathies aside picking the best tool for the job. And if you do not need Motion Blur (or very subtle one that can be faked in post) go MentalRay, but for most other cases I would go Renderman any time.

For a side by side comparison of rendered output and rendertimes please download the following PDF → Renderman vs MentalRay comparison

The [Doncaster Road End] 1 blog has posted about the Infernal video we re-did for the world cup (you can read about it in one of my older posts). In [their post] 2 they review several songs that were… er… well… mutilated to make a quick buck of the world cup. “From London to Berlin” actually made a reasonable standing. Thanks guys!

I am aware that this was no high quality piece, but believe me, it was at least a challenge from the technical side. And I actually think it works better then the original version. In this new one the story tells better and clearer.

Anyway, thanks for the review.

Is CG dull and boring. I asked this question in my last post. Well, since then I have seen a recent production called [“Elephants Dream“] 1, it is the first big project released, completely done with open source software, in particular [Blender] 2, the open source 3D suite. It shows that Blender is able to produce nice stuff, but it also validates my thoughts concerning dull CG.

First, I didn’t get the story. That might not be my fault, but I was never a fan of these artistic movies. Have something to say? Better make sure people understand, what you are saying. Have nothing to say? Then shut up! But that is just me.

Second, the movie is everything I talked about last time. It tries to look realistic, but fails in doing so. And because it fails, all it’s flaws become all the more obvious. Sad but true. I was distracted by the stiffness of the characters all the time. And I mean not only in motion, but also in the looks. The makers tried to be so true to nature, that everything feels like true…plastic.

###So, is CG doomed to become more boring every day?

Luckily not. While I was very disappointed by the Blender movie, I also found some marvelous examples of animation from some French students (thanks to Peter Soderbaum from Passive/Agressive). These movies saved my day and more. They gave me hope again, that somewhere out there are still creative heads, that try do lift their work out of the ordinary. Go and take a look at [Burning Safari] 3, a great funny short film. It has a great story, superb timing and a really nice look, that fits perfectly for the setting.

And while you are there, you should not miss [the other movies of this years students of Gobelins l’ecole de l’image] 4. I personally liked the Pirates movie best - even though it is not (plain) CG. But I just love the ending. Check it out -> [Pyrats] 5

I thought it is time to show a little of the work we have done so far. All videos should play fine in Quicktime player version 7, but might also run in earlier versions, I didn’t test that.

The first project or a thing that is still in development, but is already used in production, Monika and I are developing a template for magazine commercials. This short clip shows a render test I did, which was also used to show off to potential customers.

→

The next two clips are show two projects we did both last week. The first shows a 3D CD-Cover for the Soccer Worldcup’s theme song Bob Sinclair’s “Love Generation”.

→

The second is a remake of the Infernal video “From Paris to Berlin”, which we remade for the Worldcup into “From London to Berlin” with a cartoon caricature of Wayne Rooney of the English top players.
This was a real challenge, because we had to re-create the whole thing without the source materials. We only had the final product in uncompressed quality and a few Maya scene files. Quite a challenge for a one week project. What do you think about it Udo (if you ever see this)?

→

→ [Original Version “From Paris to Berlin”] 1

There are a lot of nice rendering choices out there. The range goes from free - like Aqsis, Blender or Yafray - to bundled renderers that come with the 3D package of choice - Maya software renderer or Lightwave come to mind - to seperate high-end renderers - like MentalRay (although that comes bundled with a lot of big players nowadays) or Renderman. Which one to choose? And which one is actually affordable? That is indeed a tough question. The guys at ZAON studios have [a nice overview of what makes a renderer good] 1. It is all about what the renderer is supposed to be used for. Nothing more nothing less.

In the case of Fridthjof Film, we need a renderer that is able to be as fast as possible, while doing a great job with fine detail. Flickering is one of the major issues we deal with. As we are working with Maya, MentalRay seems to be the logical choice, but it is damn slow when it comes to make fine details really smooth. MentalRay doesn’t allow to adjust the pixel sampling independent from the sampling of shaders, which makes us increase overall render time when we only needed to increase the render time of one parameter - and that in the long run increases the render times into unacceptable regions. Sad but true.

Now Renderman seems to do the trick, which I already new from the start (being pro-Renderman for quite some time now). The crux with that is, that there is not really a good bridge to Renderman for small companies that cannot afford to program their own. Sure there are free Renderman renderers, but what good are they if I only have an interface only I can use (as the other 3D guys are not as technically oriented). Even [3DeLight] 2 which has a [nice interface recently] 3 is not up to par in integration with Pixars newest coup - [Pixar’s Renderman for Maya] 4.

Renderman for Maya seems like the ideal option for a small studio like F-Film, where we are three guys right now — and only one (me) with previous Renderman experience. It might miss some nice features like RIB im- and export, but that is something I could live with, because the prime feature is usability by non-technical 3D artists.

But then again, a license for Pixars Maya plugin is not exactly cheap with $995,- per license. A major block on the road to bring this baby to the company - at least as far as my boss is concerned. I need to set up a test case (at least one) to show him the advantages and show him that they outweigh by far the disadvantages. Well, I guess there I have my next step.

Last week we spoke about naming conventions, which got inspired by a topic covered on [Lifehacker] 1 and [43 Folders] 2. A great start to a more streamlined workflow. But what good is a great filename, when you are unable to find your files in the chaotic folder structure you might have?

For exactly this reason we will talk about how to get an optimized folder structure for multi-media projects. I will obviously have a bias to look at it from a compositors/3D guys/VFX supervisors view, but I am also able to accommodate you editors and other departments.

What we will end up with is an understanding of what is important in a file and folder structure, so you are able to easily adjust the two templates, that we will discuss, to your needs.

###Template No.1 - General purpose environment

This template is geared to a more editing heavy workflow as it seems to be the case in most small TV commercial shops. It also has its room for 3D and compositing, but those are mere means to pimp the editing.

As the main structure we divide into:

• scene files
• materials
• sound
• output
• provals
• scratch
• temp

####Scene files

Obviously all the saved scenes from the involved programs — like FinalCut, Maya, shake or AfterEffects — go in here into their own subfolder. Now these may vary depending on your tools, but at them moment our scene files substructure looks like:

• FinalCut
• DVD
• Maya
• Shake
• Combustion
• Compressor
• Other
• Temp

As you notice there is a temp folder just like in the main structure. These are supposed to be the brainstorm areas of the folder structure. Everything that goes in here is just dabbling and will not be backed up and the Temp folders will be deleted at the end of the project.

Now depending on the tool you might want or need to go deeper with your folders. For example, Maya has its own project structure. You could use the default one in the above Maya directory or you could adjust it, so it takes advantage of the folder template we are just creating — which would be a wise choice, because you don’t want to have several folders with the same meaning hanging around.

####Materials

Materials is everything coming in from the outside world. Graphics from the client, video clips, sound bytes, word documents and storyboards. All that goes in here.

Again depending on your individual needs you may subdivide further into:

• Graphics
• Documents
• Fonts
• Subtitles
• Video
• Temp

Should be pretty clear what to use when, right?

####Sound

On most projects there will be sound involved. Music, sound effects, maybe speak. All that goes in here. Again within its own folder, like this:

• Final Mix
• Speak
• Music
• Sound Effects
• DVD soundfiles
• Temp

####Output

Now this is our catchall folder for renderings. We render uncompressed material for archiving. We also render back and forth from editing to VFX. The DVD department needs to receive files, etc.

A folder structure like this looks reasonable:

• Uncompressed
• ->VFX
• <-VFX
• DVD
• Sound
• Temp

The first folder catches all the uncompressed output from editing. ->VFX is all the stuff that goes into the VFX department and needs 3D or compositing effects applied. The VFX department then renders back to editing (<-VFX) the finished effects.

What the VFX department does when it has to render intermediate files you ask? That is a very good question. There are two ways to handle this in my opinion.

The correct way would be to render all the layers and pre-comps into Output/->VFX because it is something that goes again into the VFX land. I also tend to create a 3Drenders and a 2Drenders folder, so I can split up the sub-folders and files. Otherwise it might get a bit too crowded.

The lazy way would be to stay in the <-VFX folder. That has the advantage of staying in one folder all the time. Which means human error can be minimized. It happens all too easily that you accidentally render in the wrong VFX folder. And if you have to switch between the two folder (->VFX and <-VFX) all the time, then it might happen quite easily.

DVD is obviously the folder where all the DVD files belong. All the m2v and m2a and vob’s and whatnot files only the DVD department can make use of.

->Sound catches all the files that go to the sound editing as reference files.

And Temp is the same as in all the other locations.

####Approvals

This is just a simple folder where all the files the customer needs to approve are saved. Preview clips and such.

####Scratch

This folder is similar to temp, but has a special meaning for the FinalCup folks. It is the folder where FinalCut renders all the internal caching and pre-render files.

####Temp

I said enough about the meaning of the Temp directories in the other paragraphs I think. As the name implies, the Temp folders are there for dabbling and trying things out. It will not be backed up and it also will be deleted at the end of the project.

It is a sandbox, a playground to try things out.

####That’s all folks

That was a rundown on a basic production ready file and folder structure. Combine it with the naming convention I showed you and you have a nearly bullet proof system for all the basic needs of a production.

And of course don’t be shy to adjust this template to your needs. This system lives from its ability to blend into the working environment. The idea is to make handling files in a project easy and fast, while minimizing the potential sources of human error.

This is by no means the ultimate word on folder structures and if you have improvements to the above, I am all ears.

[Keith Lango] 1 has an interesting article on his blog concerning [CG pictures/movies and how they have become boring] 2. Keith is of the opinion that the CG look has kind of worn of. We no longer have the “wow” moments when watching a nice full CG movie. And I must say I know what he means.

Nowadays every guy in the backrow knows about 3D and VFX and that it isn’t a secret art. And that kind of pisses me off. I liked to do magic with pictures. I love the expression on the faces of people watching my (or any cool stuff for that matter) and be completely blown away by the uniqueness or pure beauty of effects. Maybe I am a nerd, but it was fun.

Today all we do is making stuff realistic. Yeah! Bring it on! Let’s do stuff like we see it every day without CG!

I have only one word for that: boring

I really am fed up with someone coming to me saying: “Hey can you do a magazine or CD or whatever object floating in an undefined space doing normal stuff (like opening and closing)? …oh and by the way, can you make it as realistic as possible?”

It is kind of frustrating to work within bounds of one world, when all the candy is in the other world - the CG world - where we can do stuff that is unrealistic and exaggerated.

Or is it just me and Keith?

###Following are the comments from the old blog

stickman Says:

Well, count me in. NPR has long been neglected, with endless raytracing programs on the market yet hardly any true NPR renderers. Hard to believe, for two reasons:

1. Look. A stylistically relevant yet unique look draws an audience. Madagascar shows the potential. So do most all the coffee-table art books of “boring3D” films where the artistic development looks SO much better than the final version.
2. Efficiency. The last 10% of finalizing a Pixar-look film can comprise 90% of the effort - just think of hair groomers, particle specialists, geometry doctors. NPR looks are much more forgiving in this last phase and can allow for quicker finaling without compromising the visual impact.

In short: we need releases of software like “JOT”. For indy film-makers, students and even the big studios.

AlexK Says:

Hi David,

I have never heard of JOT. Can you give me a link for it, please? Sounds interesting.

stickman Says:

jot: jot.cs.princeton.edu

also check here for someone doing some recent research. Its just not fair that this ype of software isn’t ready for production use…

www.rendering.de/forum/vie…

and another link: visgraph.cs.ust.hk/MoXi/

AlexK Says:

Wow, JOT seems really awesome. Thanks for the link David.

I was on set last week and I felt like back in school again. People preparing lamps and set pieces, just like preparing for a school dance.

I have this feeling more often lately when talking to my colleagues. A feeling like a teenager at play. Like me and my buddy writing a script and discussing how we solve this issue and approach that scene. A feeling of adventure, of doing something like the big guys.

It is like I have no idea what I am doing, but so do all the others. The only difference between now and then being, that it is our daily job and we get paid for it. Is that good or bad?

We get to play our favorite game, sure, but on the other hand we call ourselves professionals and charge money for playing around, pretending to have a clue.

Do you know that feeling?