That is the question.
I’m considering adding grain to the music video I’m working on for Rich Ferguson. My first step was the standard After Effects Add Grain filter. I chose 5279 as a starting place, because that was a well-known stock – fast and pretty.
Well, I don’t know what the people at Adobe were smoking, but this preset looks nothing like film grain. Compare the clean image of a comp from the video with the next image which is the preset. Look at the random colored grain. Ok, fine. But then take a look at a still frame from our soon-to-be-Ex Governor in Predator (ok, not shot on Vision 500, but you get the point). Firstly, there is very little color variation in the grain structure.
I tried to find some custom settings to match this look and it seems pretty close. Happy.
However, anybody who has looked at color film grain very, very closely will know that there are these rare and crazy, outlier grains that are fully saturated in the r, g, or b ink. There is no way to dial that kind of random color noise into this filter. Bad on them.
Now there is another interesting problem about making this digital, noiseless stuff look like film: what I’ll call random pattern resolution. Wha-at?
An image on film is captured by molecules made opaque by silver, right? Where they land on the film surface is random – organic if you will. In the case of a CMOS chip (like the Red or Alexa or 5D) it is a grid – a regularly patterned grid. So there is no variation in the placement of the pixels in the image. In the case of film, the placement of these ‘pixels’ changes every frame. This gives film one advantage and creates one side effect (that I can think of). The advantage is that the perceived resolution of film is much higher than the real resolution of film (meaning tested in a still frame), because the human mega-mind averages together many frames of random pixel placement together to form a composite frame. The side effect is that there will be no straight lines on a film frame – because the silver halides are randomly placed on the film. Playing at 24 frames per second a line will be perceived as straight, but it isn’t.
With that said, take a look at the custom image of the video. The settings approximate the natural film grain, but the lens flare created a perfectly straight line.
It seems to me that Adobe ought to add a souçon of distortion to the image. And, of course, if they were serious about the whole affair, they would add that distortion randomly to each layer of the film grain: r, g and b – and then re-average them together to create the properly dynamic image.
Now, if you are following the logic, that’s going to mean up-sampling the image (because let’s face it, nature operates on greater grid-detail than 1920×1080 or whatever – AKA infinite) to get some sassy sub-pixel interpolation, and then down-sampling it again.
Look for the iphone app soon – meantime, I suppose I could up-res and separate the rgb channels, add a different noise turbulent displacement filter to each layer, recombine and down-res. But I won’t because I would never finish the video.
My fellow digital cinema geek and frequent Reduser contributor Dan Hudgins called foul and pointed out that even film grain is an anachronism – noise is the new grain. And noise IS single pixel width so the whole question of random distortion is mute. He also points out that adding noise to an image that will be compressed for delivery is a bad idea. Perhaps, but not necessarily. If the goal is to make the image pleasing to the eye, then a little noise might be just what the director ordered. They add noise to the picture for cinema release after all. And with compression, banding is a common problem – the solution to that is, of course, noise.
So with all that in mind, I’m going to take another look at the noise/grain issue and report back with some moving samples.
