Yipe! This page's days could end on July 30. Click here to see how you can help keep this service on the web!

Base, Shade, & Highlight

Theory, Part 1

When we render a 3D surface, each pixel gets a color by summing-up the component sources of illumination as they reflect or refract through the surface. 3D renderers don't take light away, they only add. Thus the classic illumination equation (the only equation in here, honest!):

Or in English: "the final color will be the sum of the ambient, diffuse, and specular colors." Base, shade, and highlight. Fancier versions of the very same equation can add terms for opacity, reflections, and refractions, and might vary the surface colors for diffuse and specular reflection; but down there, in every renderer's secret heart of hearts, is that same simple addition.
We can start fiddling and having fun with the rendering when we realize that we can add the portions together ourselves, the way we want them, not the way some distant programmer thought they should be. We just render the different light components separately and add them up in Photoshop. Along the way, we can play all sorts of mischief on the image -- we can even "raytrace" it.

Chalk and Obsidian

Practice, Part 1

For our first experiment, let's snazz-up an opaque ball (typical idealized solid shapes for raytracing) by giving it a super-glossy highlight. The first figure shows a basic scene, unaltered. We will need to render it a couple of times to get the effect we're after -- once with no highlights, and then once with only the highlights. I call this pairing "chalk and obsidian." Since we will be manipulating the results, and I'm not texturing, even the simple "preview" rendering in MacroModel is adequate for our needs. First, we render the objects with highlights turned off, or turned to zero. Some renderers have a ready- made "matte" surface for just this effect. The result is a dull chalky look in our base color. Next, we render with highlights on and nothing else, once for each highlight. This is usually easy to do -- just turn highlights on and color the object itself black, like obsidian. For this example, you can leave the highlight fairly wide -- we don't need to make it very tight, as we might normally do to get a "glossy" look.

Unadorned simple "plastic" rendering Blue "chalk" "Obsidian" lit by key "Obsidian" lit by fill

A Curve to Sharpen Hilights

Now we read our two images into Photoshop. We display the highlight image and perform the following bit of magic: We call up the Image>Adjust>Curves menu and change the curve to something like the one in the figure (I have this curve saved for quick retreival). What did this do to our highlight? It spread-out the bright center and gave the edge transition a sharper drop-off. The effect is to accentuate gloss and to emulate the reflection of a larger-area light source -- like a photographer's studio light source. Now we select Image>Calculate>Add and add our revised highlight to the chalk image. *voila* Original "plastic" Glossy Photoshop Result

Getting Bent

Theory, Part 2

Refraction occurs when light crosses from one medium, such as air, into another, such as glass or water, and is deflected at the transition. The amount of deflection depends on the materials in question and the angle the light hits the surface at. For a raytracer, these material definitions (the indices of refraction) can be very annoying and difficult to get just right; while you can look at an object and say "that's a red glossy object," there's no easy way to just look at it and say: "That should have an index of about 6.127." Refraction indices are fairly abstract.
Simulating raytraced refraction effects is in some way easier than just raytracing them. Savvy Photoshop hacks may already suspect we will use the Photoshop "displace" filter. But how do we get a perfect displacement map for an arbitrary model? By subverting a little bit more of CG rendering theory to our own use.
Since refraction depends on the incident angle of light-to-surface, we can make a displacement map where the greatest displacement occurs in areas that are the most perpendicular to the view. How can we do this? With our old friend the "chalk" surface.

Lit Like a Musical

Practice, Part 2

Set the surface to white chalk and turn all lights off -- including any ambient light. Now set a pure-red infinite light coming directly from the right (X = infinity), and a pure-green infinite light directly from the top (Y = infinity). Render this (let's call the image "RtTop"). Be sure to leave shadowing turned off.

Pure Red and Green Lights on a Dull White Surface

Renormalizing the intensities to 50% grey

gray plus half of "RtBot" less half of "RtTop"

myPic.dmap

Undistorted background

Background with Displacements of 5, 10, 20, 40 pixels

Prismatic Displacement

Now to add color. Call up our original colored chalk-shaded view, and composite it over a white field (If it has an alpha channel, try Select>Load_Selection, Select>Invert_Selection, and then fill with white). Now use Image>Calculate>Multiply and multiply the white-bg image with our distorted background. Too heavy? Try adjusting the levels of the white image before the multiply. Remember, you can undo and alter the intensities as much as you like.

Multiply by render-on-white to "tint"

Finally, for more solidity use Image>Calculate>Add and add our old obsidian highlight pass to create highlights on the glass surface. That's the bare-bones approach to Photoshop refraction. It might seem like a lot of steps are nessesary to "raytrace" this way, but once the look is figured out, the whole thing can be quickly set up as a QuicKey or similar macro. It's fast, interactive, and can even be an effective shortcut if you already use a "real" raytracing renderer.

...with extra highlights and "Anti-Zed" (see below)

• We didn't account for the refraction at the back of the object. If the object's symmetrical, like a ball or water droplet, try multiplying the red-green images with themselves (i.e., squaring them) before adjusting levels and building the displacement map.

• For non-symmetric objects, turn them inside out in the renderer, cull backfacing polygons, and render red-green sets, colors, and highlights for the inside back surfaces too. Do the back refraction (with negative displacement offsets) and coloring first, then do an extra refraction, color, and highlight pass for the front.

  One note -- objects with lots of internal surfaces can make this cheat reveal itself. Then again, it usually looks just fine, because it "feels" right.

• Add a blue light positioned at the camera for the red-green displacement plates (the blue channel has until now been completely black!). The blue channel will be bright in the center of the object (where the surface is flat-on to the camera view), and dark at the edges (where the surface is at right angles to the view). We can use this channel (which tells us a bit about the depth or "Z" of the object) to darken the edges of the object, to enhance contrast, and cover-up any aliasing along the edge. It often helps to tweak the levels of this image, too.
  
  The Anti-Zed
  Here is a typical manipulation of the extra (blue) channel: Image>Calculate>Difference between it and the alpha channel to get what I call the "Anti-Zed." Tweak the levels to make the brightest value about 30% gray and then Image>Calculate>Duplicate it into the Selection of your image -- now erase to black. (I like to do this just before adding the final highlight element).

• The whole displacement-map-building process can be streamlined by setting the red-green lights to 50% intensity, so that the "Levels" step can be skipped -- at the cost of some generality.

• If you have a RenderMan renderer, you can cut a lot of steps by using this custom shader that will do most of the work in one step. Just composite the result over 50% gray, and save in Photoshop format -- instant displacement map.

• If you use RayDream Designer, a similar one-step render can be made using RayDream's G-Buffer surface normal channels.

• Rendering separate light highlights can provide you with a fast interactive way to adjust the light intensities, too. Just set them all to 100% to begin with, render each set of lights, and then start adding the images together, tweaking levels, and writing down the results you like. Once you get the relative values set up right in Photoshop, just adjust them in your 3D program to match (For best results, do the same thing with the highlight and diffuse renders). Again, this is a lot faster than repeatedly rendering things just to see slight changes in the light intensities! (For a very similar technique using digital photography, see Paul Haeberli's Synthetic Lighting Page)

• Since Photoshop has a wide variety of image-processing tools and filters, try experimenting with them all, applying different techniques to different channels, and using some of our special "dummy" renderings as source and selection masks.

• Reflections can also be emulated with a technique similar to that used for refraction -- I leave it to you as a thought problem and hands-on exercise...

Ripple Glass with Highlights

1. "Difference Clouds" created the pattern on the left.
2. Lighting Effects, using the clouds as a texture channel, and with red and green lights, created the displacement map at the center.
3. The displacement was applied to our previous background to ripple the photo.
4. Finally, another layer was added, containing more ("standard") lighting effects based on the same cloud texture, to add glassy hilights over the image that align perfectly with the displacements.