Depth of the Field (The Cheapskate Way)
Friday, Match 16  2007
Depth of field is an expensive thing to render when its done 'Properly'. The most common (and most effective) method of attaining expensive looking depth without actually investing rendering time to it is to render off various objects in your scene according to their distance from the camera, and hence you can apply varying levels of blur in post.

The above method is fine, but is limited to certain types of scenes. So this tutorial deals with blurring using a Z-Depth, manually.

This tutorial does not cover how to do this in Combustion since its own manuals are more than sufficient. This tutorial assumes you're compositing with Shake, and also provides 'Practical Theory' for how to perform the same task in Photoshop.

Step 1 - Make a Scene
The tutorial file contains a scene that should work quite nicely with fake depth, I've just created a floor and some sphere's, nothing too difficult.

A basic scene for the purpose of this tutorial

Step 2 - Measure
In order to get the best possible Z-Depth range, we need to account for the full length of the image. Using the Tape helper, I've measured from my camera to the extent of the scene. For an animated camera, use the largest distance you've acquired after sampling the distances from a few frames.

The tape helper

Measuring your scene extent

Why are we doing this? In order to obtain depth from luminance, we need the most luminance range possible. If you're using an 8bit image, that means there's only 256 'steps' of depth, which can lead to banding and other visual artefacts that are obscenely difficult to fix.

We deal with this by getting the extent of the Z-Depth as tight as we can so no 'Step' of luminance is wasted. To also help this along, we're going to save out our Z-Depth as a 16bit Tif / Sgi / PNG file. Obviously more bit-depth means more steps, more steps means less banding and less banding means less visual artefacts in our final composite.

Using an image output with higher bit-depth

Step 3 - Render
Rendering is something you'll need to consider. For a still image, its wise to render over-size so you can scale down later. This not only improves your edges overall, but it gives you more pixels to deal with the depth.

When animating a scene, there are a variety of factors to be considered including the fact that this method of focus does not work well (at all, actually) with Motion-Blur. Also, the more blurry you want the final image, the higher quality your elements will need to be to pull it off.

I've chosen to render double res (for a 720x405 final image, hence 1440x810 render). Once again, you'll need to weigh up the pros and cons of this system. But I assure you it is faster than rendering with true ray-based depth.

Step 4 - To Shake!
Shake is extremely basic in design, and so even if it is not your compositing program of choice, you'll understand the methodology I am using in order to acquire the final results.

Import your two shots, do any precompositing necessary (screen reflections, adding specular etc.).

The elements imported into Shake

Step 5 - Utilising the Luminance
Now essentially, we're going to defocus the diffuse channel and then use the Z-Depth as a method to control where we want blurred. I personally have found the following to be the fastest and most controllable method.

Apply a Solarise to the Z-Depth, invert it. A solarise takes an arbitrary number and then makes any number above it subtract from that number.

With 0 to 100, if we set the variable to 50..

49 = 49
x = x
51 = 51-(51-x) ... Hence 49
52 = 52-(52-x) .. Hence 48

The Solarize is applied

Now when you remember that this applies to luminance, this gives us a focal point which we can invert to use as a matte for the defocused layer.

Now chuck a REORDER underneath the Solarize and make all four channels Red. This assures the Alpha is the same luminance as the RGB channels. Another method is throwing a 'LumaKey' onto it (same thing, for all intents and purposes). Now simply feed that into your Defocus.

Solarise applied, reordered so Alpha matches R channel luminance and fed into the Matte channel of the Defocus.

So you can now grab the Value of the Solarize and slide it up and down to pull focus on whichever part of the scene you desire.

If you want a tighter focus, you can put a ContrastLum node underneath the Solarize (above the reorder). If you attempt to defocus too wide, you will get edge aliasing. This is fixed either by using a large image (and then scaling it down for the final) or by simply making it less wide (and hence less blurry).

I also commonly change the gamma and up the grain slightly in areas of defocus to add some real world-ness to it.

Final Script with Basic Z-Depth DOF

Frame with subtle usage of DOF.

Click Here to Download The Max File for this Tutorial