The Pigeon: Impossible Blog

I’m working on the rigging episodes but I’ll still be posting some additional thoughts on lighting and compositing here on the blog as they come up. One thing that I forgot to mention in the episodes is how this multi-pass approach can really help speed up rendering static shots. Consider this scenario: You’ve got a character and the environment behind him. The environment takes about 1 hour/frame to render. The main character takes about 5 minutes/frame (for all of the passes.) If the camera is static, you can just render out a single frame of the environment, then put the animated character on top. Coincidentally, about half of the shots in PI are static for this very reason.

In some of the moving shots, I’ve even gotten the animation most of the way there, but I still need to go back and polish it at some point. As long as the camera is locked off I can go ahead and render the environment if there’s nothing else in my render queue. In a couple of cases, I’ve even wanted to go back and polish the animation after the shot is completely finished. At only 5 minutes/frame it’s a no brainer to make the changes and re-render just the character. This works especially well when doing commercials or other projects with tight deadlines that require a lot of last minute changes.

I’ve gotten a few comments from people having a hard time understanding the RGB particles thing. It took me a while to figure it out the first time I saw it too. If you didn’t quite get it, here’s a different explanation that might help:

This is all one pass, and exactly what comes out of the render. What you are seeing is that each of the 3 lights is set to a unique color. Either pure red, green, or blue. We’ve also got three color channels in our image (Red, Green and Blue). So, when you look at one of those channels by itself, you only see the one light with the matching color. It’s a lot like the old anaglyph steroscopic technique (the one with the red and blue glasses.) You tint things a different color, and then it can only be seen through the opposite color.

Another way of putting it is that it’s like we’re rendering each light to its own Black and White image, but we’re storing that information in different color channels within the same file. We don’t really need those extra channels anyways because smoke is a solid color.

Let me know if that makes sense. Its really hard to put into words because its kind of like looking at the matrix encoded. Just remember that there is no spoon, and before you know it, you won’t even see pictures anymore. Just 1’s and 0’s filling up an image raster.

Podcast #007: RGB Passes is now online!

Simon Reeves has just put up an excellent breakdown on how he’s doing the compositing for his “Bank Affairs” short. It not only shows some cool tricks and methodology, but its a great example of the many different ways you can approach lighting a scene.

In actuality, Simon’s approach really isn’t that different from mine. The only visual difference is that I do a separate ambient and shadow pass because I don’t want my occlusion happening on top of the parts that are in direct sunlight. (See Podcast #3) In his case though, the lighting is much earlier in the day so the little bit of sunlight that gets through is heavily diminished by haze and cloud cover. That means that he absolutely DOES want the occlusion on top of everything because the sky light is much more prominent in his scene. Just an example of how compositing isn’t a “one trick fits all” business. Even the way in which you arrange your passes can give you a completely different look and feel. Its all about paying attention to how light behaves in the real world and figuring out the best way to represent that in the computer.

From a technical standpoint, the other main difference in our approaches is that Simon renders out everything at once which is definitely the preferable way to do it. Once you have objects broken into different passes, suddenly you have to do even more passes because of how light interacts between those different objects. (Shadows, Occlusion, Reflection, etc.)

So now you may ask, why does crazy old uncle Lucas render out 60 friekin’ passes? Well, there are definitely a few advantages to having things broken up, but the main reason is that because I’ve been working on this project for so long, I first lit the environment almost 3 years ago. Since that time, we’ve had a little technical advancement in the form of 64-bit computing. Until that point, you couldn’t have more than 2GB of RAM which is a huge bottleneck for rendering large scenes. In other words, until the last year or so, I simply COULDN’T render everything in one go. I had to break it apart into manageable chunks which is why my current setup seems a bit whacked.

If you can get away with a couple of passes and some RGB mattes (which I use as well and will talk about in tomorrow’s podcast) then save yourself a lot of trouble and remember the KISS acronym: Keep It Simple Stupid!

Podcast #006: Camera Projections is now online!

Podcast #005: Lighting the Eyes is now online!

Podcast #004: Lighting Hacks is now online!

I received an e-mail from Marcus Lee that answered my beauty pass question:

You mentioned that you don’t use or render a beauty pass. Don’t you have to render the beauty pass to get all the channels (reflection, specular, etc) though? I’ve never used the beauty pass either, but I’ve always rendered it, so that I could also save the channels and then just deleted the beauty pass. Is there a way to avoid doing this?

I’m not using the render channels, (a feature of most renders that will let you output each framebuffer separately.) While this does give you a little bit of control in compositing, I’m building my own custom passes that are designed to work hand-in-hand with how I’m compositing them. In other words, I’m approaching it in terms of “how would I composite this image?” and then figuring out the passes I need in order to do that. However, the render-channels workflow perfectly answers my beauty pass question, its just the opposite way of doing it. You’re thinking “how would I light this image” and then using compositing as a tool to put on the finishing touches. Both are valid, just different.

He also brought up a good question:

Also, why did you do the shadows with 3 passes– shadow, ambient, AO. Wouldn’t it be faster and use less memory to use only the shadow pass to matte a CC effect (ex. hue/saturation) to make the shadow? Then you don’t have to render the ambient or AO pass.

Or if you like the little dirty shadows, you could use a combo of the shadow pass (as described before) and add the AO pass on top. Then you would still be using one less pass and you could still (I think) get the same effect.

Well, as you mentioned I definitely think that the occlusion adds a lot to the image so its definitely worth it.

The short answer is that its all about control. There’s no sense in saving a couple of minutes in render time if it means that you have to combine things that you want individual control over. For example, one important thing to note is that there’s not a blanket pass of occlusion over the whole scene. I only use the occlusion on the shadows because sunlight is so bright that it basically wipes out any effects you’d get from bounced light. Whenever I see occlusion over everything, it just feels very “CG” to me but that’s a matter of personal taste. (I tried to do a mockup of how this would look but how I have my passes set up makes it impossible to do without re-rendering everything)

And finally, the reason I do a separate key pass and ambient pass is that the key pass has a sense of light direction, while the ambient is supposed to be perfectly flat-lit. Notice in these pics how there’s a highlight on the sections that are perpendicular to the sun, while the ambient is perfectly flat.

Hope that helps. Thanks for the e-mail Marcus!

I just got an excellent email asking what my beauty pass looks like before compositing. My response was that I don’t render them. However, I figured it was worth posting here because I actually need to know just what a “beauty pass” is and how other people use them?

To elaborate a little, one of the reasons that I use this multipass workflow (besides the control it gives you) is that I use some of the more expensive techniques like area lights, raytraced glossy reflections, occlusion, etc. and if I tried to render all of that stuff at once, Mental Ray would stall for a few hours, and then unceremoniously turn my computer into a flaming pile of circuit boards. I’m guessing beauty passes must be part of a completely different multipass workflow. Does anyone out there use them?

Podcast #003: Multipass Lighting is now online!

If you’re wanting to see a more in-depth image of the compositing tree, below is a screen shot of the comp in Combustion. Beware… it’s a pretty huge jpeg.