Well, Manolo Blahnik shoes aren’t cheap.
The OP needs to keep in mind that in spite of CGI, a lot of shots are still effects driven, not CGI. Explosions, smoke and other CGI particle effects have only recently started to approach “realistic”. For the most part, it’s still a lot of breaking and blowing stuff up.
I agree with those who’ve said that digital FX have stepped up the bar, and here’s a good example: backdrops.
Back “in the day” you might use a matte painting as the backdrop for a scene. You intentionally engineered the scene so that it made sense with a static backdrop, and then you just had to commission a team of artists to slap some paint together and make it work. Your only other option back then was to film on location or build the entire set. Both were often difficult.
Nowadays, producers realize that digital art can be used instead of paint. And they realize that there’s no reason to shoehorn everything into static backgrounds. They go in and add streams, traffic, bugs, smoke, clouds, etc. that all move. This costs more money than a simple matte background, but technology has made it possible to do at all.
As an example of this, just think Star Wars re-releases. A simple desert scene with a couple of buildings upon entering Mos Eisley was turned into a dinosaur zoo. Flat white panels on Bespin were turned into panoramic windows with views of moving vehicles and clouds.
Also, one more thing to consider: CGI doesn’t always work the way you think it will. Sometimes, there’s a lot of retooling required to make the models or rendering come out the way the director wants. The Incredible Hulk was said to have had a really crappy Hulk until very late in production. The Garfield movies were shipped with what is possibly the worst computer-animated cat in history.
Did you see how many Lighting TDs there were in the credits? Whew, I’ve never seen so many! They had about 20 compositors, and around 150 lighting TDs!
(Regular effects movies have maybe three lighting TDs and 60 compositors.)
I did a report on this a long time ago, and one reason CGI is so expensive is because of the labor. X seconds of CGI animation takes Y hours to render. The more complicated the effects, the bigger Y is. The ratio was (iirc) something like thousands to one, and if there’s a change, it has to be re-rendered. I interviewed one of the wranglers (a person who sits there making sure one of the renders goes correctly). So, while the technology probably gets cheaper as time goes by, the labor costs should increase according to the complexity of the graphics and rendering time.
There is an interesting point in one aspect and that is the declining cost of effects across time. Cameron has commented how the per frame cost of the Pandora virtualization process in Avatar, near the end of the film, was a tiny fraction of what it was at the beginning due mainly to the declining cost of CPU horsepower across the span of years in putting the film together.
It’s because us, CGI artists have obscene salaries and benefits that allows us to live a live of comfort and glamour… . Well, not really, but one can dream, no?
This is part of it, I remember once reading an article about Pixar, they where asked how long did each frame in Toy Story took to render with the small, slow (relatively speaking) render farm they had back then. I don’t remember the actual numbers but let’s say that it took them 1 hour per frame. Next they ask the same question about (I think Monsters Inc.) with a vastly bigger and faster render farm, the answer was that still the render time was 1 hour per frame.
The increases in computing power have been capitalized in greater complexity and realism.
People tend not to realize it, but CGI graphics are mind boggling complex affairs.
For example, I working now in a commercial with lots of flying letters (woe me, a character animator doing this), I have 20 shots to animate, just in one there may be 200 letters that I have to animate by hand because there simply isn’t an automated simulation that would give the precise results needed. So take one single letter, it has 3 degrees of freedom, position (itself with three axis XYZ), rotation (XYZ) and scale (XYZ). Now this letter moves around so at certain frames I have to define it’s position, rotation and scale values. That’s 9 values in a graph (pros animate via graphs 
) per frame, let’s say the shot has 50 frames and there are 5 frames of those that contain key frames. So for that letter I’ll have to adjust, nudge and tweak 9*5 (45) values and the way they interpolate.
Now multiply by the other 200 or so elements in the scene (boing! 900 tweaks) and then by the 20 shots in the project (18000 adjustments). On top of that there are multiple revisions of the work, in this example 3 approves where the director comes in and asks for changes, so the work gets easily doubled.
And that’s just the actual moving of the things around, not the rendering, modeling, texturing, etc, etc.
In a CG shot everything, and I mean EVERY big, little, subtle or obvious thing has to be hand made at some point. If not by actually pushing vertices in a mesh around then by some programmer burning the midnight oil.
Things that may appear simple as in “You push a button and it’s done, right?” can be the most time consuming; for example water simulations. I’ve spent up to one week or more working and refining a single 1 or 2 second long shot with small scale water simulation for TV commercials, at the higher resolutions of cinema it would take much longer, make a simple spill a gigantic tsunami and the budget gets through the roof.
Minor hijack:
When I worked for NASA doing Satellite Attitude calculations, we called that 9 degrees of freedom.
Well, we didn’t have scale, but we did have x,y,z, pitch, roll, yaw, and that is 6.
/hijack
It’s probably a similar effect as leads to software bloat. The more than can be done, the less that creators feel the need to reign themselves in and simply splurge, ending up hitting the same benchmarks as the previous generation.
Ultimately, though, you’ll note that when you factor in ticket prices and inflation, the average cost of a movie hasn’t changed all that much. Cleopatra, from 1963, is still one of the most expensive films ever made.
Ahah, I found the “law” I was looking for, but couldn’t quite find while writing my previous post:
More than that, it’s a running joke at SIGGRAPH panels. If render speeds start getting too quick, some new technique has to be invented to eat the cycles. It’s been a long time since I’ve been involved in the field, but I imagine that people are now raytracing subsurface scattering.
This post made me nostalgic for the TDI Explore system I used to work on. It got bought by Wavefront, then got gobbled by SGI and became the basis of Maya - but along the way it lost it’s knob box. 8 rotary encoders that changed function depending on what you were working on. Select the object and three knobs are now axis controls. Select one face, now two controls are scale U and V, another is transpose on normal axis, etc. So much faster.
I’ll note that Empire Strikes Back had a muppet Yoda, which was in my opinion more believable than CGI Yoda from the movies that some say don’t exist.
Parker Posey is a star! A star damnit!!
I remember reading something recently that implied SSS isn’t being used as much as they had anticipated, as there have been so many shortcuts and cheats figured out, which take a fraction of the effort and are good enough for most situations.
Sometimes I think developments in CG techniques adhere too closely to the laws of physics for not much gain. Whatever it shows them, they should then be able to extrapolate a cheat that gets them an equivalent result.
Yes and no. CG developers aren’t the only ones looking for shortcuts; many of them are already built into the “laws” of physics. Using raytracing, for instance, is itself a cheat around using true light waves, with diffractive effects.