Lumetri is NOT 32 bits

Question

One of the benefits of 32 bits effects are their ability to retain all details in the image, even if you overexpose it early in the chain. Try for instance to over expose a clip using ProcAmp, and then add another ProcAmp on an adjustment layer, and lower it. The details are back. This is because ProcAmp is marked as a 32 bit effect in the Effects-tab.

Lumetri Color is also marked as a 32 bits effect. Still, if I lift up the Lumetri Exposure so parts of the image is clipped, the Lumetri Curves will not bring back the details. This also applies if I crank up exposure in one Lumetri Color and then reduce it in another Lumetri Color (or using another 32 bit effect, like ProcAmp), either on the same clip, or on an adjustment layer. The details are permanently gone. Even if you overexpose with the Exposure slider and then lower it again with RGB Curves in the very same Lumetri effect. This tells me that Lumetri is NOT 32 bit but 8 bit. Or am I missing something here?

I have also tried this with LUTs. LUTs are applied first in the chain. It will definitely cause some parts in some footage to clip, and that clipping is not possible to fix later in the very same Lumetri Color. It helps if I apply a second Lumetri to all clips and apply the LUT there instead. This can't be the correct work flow, can it?

I have tried only with Windows 10, on two computers, both with GPU acceleration (Cuda) turned on. It doesn't help to turn on max bit depth in the sequence settings, or using high quality playback in the Program Monitor.

R Neil Haugen · Answer

You've got a few things that aren't connected here.

The 'bits' they refer to is the processing depth. And Lumetri is processed in computations as 32 bit float data. Period. So that is correct, the processing is 32 bit float.

(Processing bit depth math is not the same as clip bit depth ...8 bit data is only 8 bit data, whether processed in 32 bit float or not. As with 10 or 12 bit. 32 bit float processing simply gives the most detailed processing data back.)

Next ... the clipping induced by some tools. Yes, you can induce clipping with some tools, that push the data up past the math 'chain'. Others don't. So depending on the tool used, some will hard-clip. Some won't. TEST ... and you need to do that with any color correction tool in all color correction apps.

Learn what ones clip, and ... don't do that. Period.

And yes, I work in Resolve at least as much as Premiere these days. Some tools in Resolve will hard clip, also. It's what they are designed to do, to take the data as far as possible and then some. So as in Lumetri, you need to know what tools in Resolve will hard-clip.

As to LUTs ... another completley different issue.

In Resolve, when you apply a LUT in a node, then do some color correction in that same node, there isn't anything hard-clipped in that node ... because Resolve automatically processes the LUT as the last computational bit of that node.

HOWEVER ...if that LUT does cause data to go out of bounds, it will be hard-clipped for nodes past that node.

Which is why colorists all are taught, and teach!!! correct use of conversion LUTs ... they must be applied second in the node tree, so that you can trim the clip into and through the LUT in steps processed before the LUT.

So most colorists put the conversion in a second node and use a node prior to it to do the trim pass of the clip.

Premiere's Basic tab Input LUT slot is processed first ... so yea, if it clips your data, it's clipped. I've argued for years that they should change that processing order, but they haven't. Maybe Alexis will get around to that. He would know it's wrong.

Which is why I always teach applying conversion LUTs in the Creative tab dropdown list, so you can use the Basic tab to trim the clip into the LUT.

Of course, now that they have algorithmic based transforms for most media, I do not recommend using LUT based converssions unless they are required in a client's shooting process.

Because the algorithms are actual high-end mathematical conversions, and will not clip or crush, or over-saturate your data. Unlike LUTs, which aren't math at all, just a table or chart of triplicate data points.

Take this triplicate (RGB values) and change them to this triplicate.

You get several 'points' of those, and in-between are simple straight-line conversions. Not nearly as finetuned and changeable as the newer algorithms, both in Premiere and Resolve.

Sign up

To post, reply, or follow discussions, please sign in with your Adobe ID.

Sign in to Adobe Community

To post, reply, or follow discussions, please sign in with your Adobe ID.

Scanning file for viruses.

This file cannot be downloaded