Thanks for the input, Blake!

On Wed, Jun 5, 2013 at 4:55 PM, Blake Sloan <[email protected]> wrote:
> Also working toward full OCIO color management. But a OCIO colorspace menu
> in Read/Write nodes is not a high priority. We typically set 'colorspace' to
> 'raw data' on Nuke Read/Write nodes for non-linear formats and use explicit
> color transformations in the script body.

This is a pretty good option 4.  It just requires more of a training
component, which can sometimes be hard to stay on top of in our
commercials context -- for freelancers in particular.  Definitely a
viable option, though.

> Seems bundling color transforms other than the standard 1D display encoding
> into Read and Write nodes might create more problems than it solves.

A big problem I have is that the standard encodings are not what I
want, and actually are doing more harm than good.  The rec.709 curve,
in particular, is pretty irrelevant and causes a lot of confusion, as
almost nothing is encoded as scene-linear Rec.709.  Rather, footage
from video tends to arrive as display-referred gamma 2.4.

If I decide to force people to use the OCIO colorspace nodes, I might
actually consider just doing away with the default luts entirely.  I
just hate deviating too far from vanilla.

> For Jpeg's to be correctly interpreted by browsers, consumer apps, they must
> be sRGB encoded, which is the default for Read and Write nodes. We often
> bake grades and viewer LUTs into jpeg dailies, but in our experience,
> there's more danger overloading the Write node than just creating an
> upstream color transformation.

Applying luts/looks explicitly with upstream nodes makes sense to me,
since looks aren't really encodings, and read/write operations should
be all about encodings, be they scene-referred or display-referred.

> Lately, DPX plates from vendors are rarely Cineon log encoded (the default
> DPX encoding by Nuke Read nodes). They tend to use the camera vendor's own
> log encoding (Arri, SLog, etc). So the default is usually wrong. But giving
> the artist control over which input transform to use is probably fraught.

This is where I feel OCIO would be really handy, as it would let us
define defaults based on roles, so as to have a fighting chance of
getting the right value without a cumbersome amount of communication.
For example, I could define a colorspace as "Main Camera (SLog2)" for
a show and compositors would know I had set the main camera to be
SLog2.  That kind of streamlining tends to be a non-issue for a
feature, but can be handy for short projects.

> Internally, VFX trafficks in scene-linear EXR with no transform applied on
> read or write. So the major concerns are that that the CG uses the same
> color primaries as the camera-captured media it is intended to integrate
> with, and that the color rendering predicts the theater experience.
>

Yes, one thought I had was just to avoid the issue by insisting that
all media coming into the pipeline be pre-transcoded as linear exr.
Not the worst option, but it's more of a workaround than a solution.
It's a sad day when Nuke needs a workaround for dealing with color.
_______________________________________________
Nuke-dev mailing list
[email protected], http://forums.thefoundry.co.uk/
http://support.thefoundry.co.uk/cgi-bin/mailman/listinfo/nuke-dev

Reply via email to