Comping 32bit vs 16bit

[bernhard]

I recently decided to render 16bit .tif files instead of 32bit .exr files.

And I was wondering why a simple composition does not work anymore if I am using 16bit .tif files.
It doesn't matter if I try to comp inside PS, AE, Nuke or Fusion.

32bit works like a charm, while with 16bit, I can not get my final beauty RGB anymore with my render passes.
Not sure if it's a gamma issue, or a different LUT I am not aware of.

The render passes look fine, if I compare 16bit vs 32bit.. they are all the same.
But as soon as I start merging a reflection pass, it's already different within 16bit.

Please see below the super simple test.

Any thoughts what I do not see here and I am doing wrong ?


LINK FOR HIGH RES VIEW:
http://www.mediafire.com/view/?r73030vew46zi6x

[joconnell]

After effects similarly burns out with 16 bit files versus 32 bit files. Is there any chance that fusion tags the files differently if they're 16 bit and doesn't treat them as linear? What happens if you render out to 16 bit exr instead of 16 bit tif? Exr might trigger a preference in Fusion that tif doesn't?

[bernhard]

hmm.. didn't try .exr in 16bit.. I guess AE handles .tif better..... I just couldn't find any answer to this thing .... till today... so I was wondering if someone know's what's going on here...

[instinct]

If you save to standard tif settings it is probably saved as 16bit integer. That is not the same as 16bit half float. In addition 16bit int will most likely be interpreted as sRGB by default. Which is what John is referring to

Regards,
Thorsten

[joconnell]

Exactly what the more intelligent person above me said.

[bernhard]

lol; ok.. we are getting closer.. and that means ? you can not use 16bit integer for comping ? Or what is the final message & conclusion ?
just render 32bit .exr and problem fixed ? And trash all 16bit .tif renders ?

[joconnell]

Ooh - it seems that after effects won't support 16 bit float at all according to this - http://www.3dtotal.com/index_tutoria...4#.US1KLVfJd8E so maybe you're right, just use 32 bit exr. I'd have a bit of a play with the format that it saves it in too - mainly the type of compression the EXR is saved with, and also the format (scanline versus tile). As far as I know, EXRs saved in tile format are way quicker to redraw in after effects, and you can do a test saving a sequence out with the various different compression types to see which plays back quickest for you. Note that this won't change the look of your exrs at all, only how they're compressed.

I don't know anything about the fusion side unfortunately so I can't help with that.

[dans]

Since you say it is the same no matter if you try in Photoshop or a compositing package: You are probbaly trying to blend passes that are in sRGB space (or gamma corrected) instead of in linear color space. You need to output your TIF images using an Output Gamma of 1.0 or teach your post production tool to automatically compensate for that during blending.

Applying gamma correction to each individual pass and then combining the passes is not the same as first combining them and gamma correcting the composed result. Same problem with other non-linear operations such as exposure control/color mapping (except of course Linear multiply). The underlying math simply doesn't work that way (unfortunately - from a usability standpoint).

[bernhard]

yeah.. makese sense... so I guess the only way is then to stick to 32bit.exr and all these issues are gone....

[pjkw]

lol; ok.. we are getting closer.. and that means ? you can not use 16bit integer for comping ? Or what is the final message & conclusion ?
just render 32bit .exr and problem fixed ? And trash all 16bit .tif renders ?

The problem is probably that 16bit integer is either saved by max as sRGB or that the comp program interpretes them as sRGB and adds a gamma correction to them. I don't have fusion installed currently but I did a test in Nuke and it worked fine if I put colorspace to linear on all my loaders (here it automatically set it to sRGB because its an integer)

We almost exclusively use 16bit halffloat .exr, 32bit is hardly ever needed for us.