I was searching for more documentation on the REDLog and PDLog LUTs* this evening, particularly so that we could get more precise input/output from our Nucoda FilmMaster for RED-to-film tests. In this case I want to convert RED Raw (R3D) footage to dpx, import that into our FilmMaster without losing any information, and then export out in a way which works for the film-out facility.
However, by the looks of this exchange, it would seem that RED is keeping that information very close to their chests (they have promised to release it to third parties in April). So for now you can only access their LUTs through REDAlert, REDCine, or SCRATCH. Even then, the conversion through the RED software is slightly nebulous, because your only choices for working with the outside world (under their workflow) is to go linear (not desirable for film) or to use the ill-defined PDLog (supposedly a R3D-to-Cineon spec). But who would know…
Not making these LUTs available as documents is like finding out your shiny new toy didn’t come with batteries on a public holiday. It’s not an insurmountable problem, but irritating. I guess we’ll have to keep rolling our own.
*Definiton: A LUT – look up table – describes how the colours and luminance of an image should actually look. Your computer uses one to show images on the screen, although maybe called something else – like ColorSync™. They are used for everything from inkjet printers, to computer monitors, to printing to 35mm film – like Ansell Adam’s zone system on steroids – and they are supposed to keep an image looking the same despite the viewing format.
UPDATE: I am quite enthusiastic about the RED, and don’t want to get down and dirty on what is, to be fair, a revolutionary camera. However, I thought this was quite interesting:
This is a crop of a REDLog output from REDAlert, material was rated at 320 ISO.

This is a crop of a PDLog 685 output from REDAlert. It maps the peak values (like the sparkle above the eye) to 685 (of the 1023 scale). Under the Cineon spec, 685 is white. The rest of the values (to 1023) are headroom. Rolling off at 685 effectively removes any notion of headroom.

This is a crop of a PDLog 985 output from REDAlert. It maps the peak values to 985 (of the 1023 scale). From what I can tell, this has a closer correlation to the original REDLog file, however, it lifts the black point significantly, so I am unsure as to the application of this except that it uses more steps of the available 10-bit scale. If this was designed to cheat in more headroom, then the black point would probably not want to lift any higher than it is in the 685 version.
Stu Maschwitz has also pondered this problem (referenced previously); he uses REC709 as a consistent (non-log) reference to explain both the theory of digital negative, as well as where someone should aim with exposure on the RED. His conclusion? To get headroom something like PanaLog (Panavision Genesis format), you need to under-expose by 2.6 stops. Having tested this, I wouldn’t do it… if you are willing to ignore the noise, however, you do get more headroom! (I will post results shortly).
As I suggested previously, Stu talks in terms of REC709 (which seems to have caused a furore), where he maybe should have talked about “linearized” images. Having seen all sorts of hijinks over correct conversion of Cineon files in the visual effects industry, it is amazing to see them all over again with this camera. I guess when people “own” the technology (camera, etc), things get a bit more emotional. When people position themselves as “experts” it gets even worse.
UPDATE to the UPDATE: And now this. I guess I should stop whining. A bit of a delay, but looks like RED are on it. I’d be interested to know what the “tiered partner” thing is about, though…
As part of expanding our workflows and our partner base, we are creating a tiered partner program that will roll out in full this year. This program will be how other companies can access R3D as is appropriate for them and for us. The first step is establishing RED Lead Development Partners. Apple and ASSIMLATE are the first and currently the only companies here. We have always said we will open up R3D to other partners, and we will. But we are going to do it right.
Read More