Is this the expected behavior of the Whites slider?

[Mark Sirota]

I've just discovered that the Whites slider doesn't behave the way I have always thought it behaved, and as an LR user since 2006, and a frequent user of Whites since it was introduced, I'm wondering whether it has always been this way and whether this is the intended behavior.

Here's a short timelapse video illustrating the problem. Note how the sky is suddenly darker in a few frames. Note also that the darker frames are the ones with light in the foreground (which is caused primarily by people in the foreground using red flashlights).

These were shot on a Nikon D850 using the built-in intervalometer (13 second exposure every 15 seconds), in manual mode (and manual everything) with Active D-Lighting disabled. They were then all processed identically in Lightroom Classic (using manual adjustments, no auto anything). The video was then created using the Slideshow module.

I have confirmed that they're all identical (and that Active D-Lighting is disabled) by carefully comparing the EXIF data of all 380 frames. I confirmed that they were processed identically by resetting them all, then processing them together using Auto Sync.

After some investigation, it turns out that the culprit is the Whites slider. They all have Whites +60 (and exposure +1.00). In the frames with significant light in the foreground, the Whites slider doesn't brighten the sky as much. The Exposure slider doesn't seem to be affected by the brightness of the foreground.

So my question is, is this the expected behavior of the Whites slider? I would have thought it would affect equivalent pixels equally, regardless of what's happening in another part of the frame. I tried older process versions and found the same behavior (going back to PV3 when Whites was first introduced), but I'm amazed I've never noticed this before.

For those willing to dig in, here's a Google Drive folder containing the video, two adjacent raw files, and the output of exiftool for each of those raw files:
https://drive.google.com/drive/folders/18ZJFC6o2zQ1eOkP0gIBLdS_9jwHd5KXn?usp=sharing

Thanks for any insight!

[richardplondon · Accepted solution]

A number of adjustments since 2012 updates (or so) have become image content adaptive, some more powerfully than others. With early PVs only IIRC Clarity and Fill Light and Recovery were. In current PV you can strategically do some of the tonal shaping using Tone Curve instead. That acts like an overlaying filter applying the same standard relative change onto a certain pixel value wherever found in images, regardless of whether relatively lighter or darker tones may occur near to instances of that value within each photo. Tone Curve operates secondarily onto whatever range between picture-white and picture-black the other adjustments are creating. Tone Curve has no direct access to the original image data and thus cannot play any part in controlling highlight recovery and such. IOW it can dor example (nondestructively of course) further clip, but cannot itself act to reverse any clipping. So use of the Basic Panel etc cannot be eliminated.

That said, by including Exposure and Contrast in the Basic panel to do the heavier work and then reserving Shadows / Highlights / Whites / Blacks for tweaks at consequently weaker strength, a lot of the more obvious image-adaptive behaviour can IME be suppressed below the level of  noticeability.

[Mark Sirota]

Thanks Richard, that's very helpful. Further experimentation reveals that with significant enough strength, at least the Exposure, Contrast, and Whites sliders exhibit this behavior, so it appears that when consistency is important (such as for a timelapse), the tone curve is the way to go unless more moderate use of the basic sliders is sufficient.

[Conrad_C]

That conclusion is absolutely true and confirmed by the author of LRTimelapse software. He’s got a great video (linked below) explaining how this happens. He discusses which options are image-adaptive (wonderful for single frames, not so consistent for time lapses) and which options are applied the same way to every frame regardless of content (ideal for time lapses). The last 10 minutes of the video are especially enlightening (no pun intended), where he specifically tests each option adjusted in isolation in a time lapse to show you how much you might want to use or avoid that option. And, he shows how the inconsistent results are amplified if the images use an Adobe raw profile; to mimimize the problems apply a camera matching profile instead.

It’s not that you can’t use the image-adaptive options at all with a time-lapse, they are still useful for times when the linear options aren’t enough to solve a problem. But you’ll get the most consistent results among time-lapse frames if you use the image-adaptive controls as minimally as possible, and leave the heavy lifting to the more linear options like the Tone Curve.

Sometimes it’s difficult to understand what the change was between frames that caused a different result, but it can be as subtle as a transient reflection on a hill of a passing car’s headlight, or a traffic signal changing. And as he shows, it could be a changed area that you aren’t aware of because it’s outside the crop area!

[Mark Sirota]

Thanks, that's a great video. It's not 100% correct -- he concluded that some controls are ok, but it turns out they were only ok for his set of images in the amounts he tested with. It looks to me that all of the luminance-related controls in the Basic panel (exposure, contrast, highlights, shadows, whites, and blacks) are adaptive. Others may be as well, but that's the extent of my testing so far. I've also determined that applying these controls with local adjustments rather than globally doesn't change the behavior.

I agree with him that crop doesn't seem to matter. It also seems that some combinations are more problematic than others, but I haven't exhaustively tested (too many permutations). I suspect the combination problems are an issue of scale; adjusting one can move another one into the problematic (visible) range. And I haven't dug in at all to the color issue he identified with the combination of Dehaze and the Adobe profiles -- I suspect that one has a different, but related, root cause.

This has been a fun little rabbit hole. I learned something today, and I'm amazed I haven't tripped over this before.

[Mark Sirota]

Just to follow up on the specifics -- you said "regardless of whether relatively lighter or darker tones may occur near to instances of that value within each photo." It doesn't appear that the location in the photo is relevant.

I created these two small PNGs (attached) for testing, and found that even though the red bar is not adjacent to the dark gray "sky" or light gray "stars", its presence affects the Exposure, Contrast, and Whites sliders (and perhaps others) for those areas. Furthermore, the two "stars" are affected equally, even though they are not the same distance from the red bar.

To test:

(1) Import them both, select them both, go into Develop and enable Auto Sync
(2) Crank up Exposure, Contrast, or Whites (one at a time).
(3) Place your mouse pointer over the dark gray "sky" or one of the light gray "stars", and use the arrow keys to flip back and forth between the two files.

You can see from the values under the histogram (or just by watching the histogram as you flip back and forth) that the existence of the red bar is affecting the way those sliders treat the grays, even though the red bar is not adjacent to them. (My earlier test of +1.00 exposure is not enough to illustrate this, but if you crank it up much further you can see it. +60 whites is more than enough to see it.)

[richardplondon]

I think 'image adaptive' can be taken to just mean that the varying content of each image can make a difference in how the adjustment applies. An especially obvious artefact here is local-contrast enhancement, which inherently deals with proximity (as when strongly positive Shadows / strongly negative Highlights give a 'HDR' style tone-mapped look).

Still, much as if one were to set an auto WB or an auto Exposure (either in the camera or in postprocessing), one expects different outcomes for the whole image when (say) prominent red brakelights were illuminated within one night exposure and not within another. 

[Mark Sirota]

Agreed, that's the expected behavior if you choose an "auto" something. In the world of photography, however, we also have the concept of "manual", but that no longer seems to exist for Lightroom. I must admit right now I'm only seeing the downsides of this implementation; I'm sure there are upsides too; I just haven't recognized them yet.

[richardplondon]

The upsides IMO are for individual images considered in isolation, as to how each can be most 'pleasingly' postprocessed in its own terms. It is probably only those doing timelapse sequences or similar operations who experience much of a practical downside from having image-adaptive rather than non-adaptive adjustments. Some workaround strategies have been discussed here which may help, and not mentioned yet has been the option of prrocessing in a much earlier PV (not very attractive as I see it), but the development of LrC's tools over time,seems to have simply moved in this direction like it or not. 

[Mark Sirota]

Ok, but what about when I'm processing a set of head shots that are supposed to look identical? Each subject's clothing or skin tone now affects Lightroom's rendering of the background. Or a subject may change their clothing, and now their skin tone registers differently.

The issue is real for any set, not just for timelapses. It is mitigated by the fact that moderate settings of the sliders do not seem to produce easily visible differences, so the issue can be avoided by shooting in a way that minimizes the need for more extreme settings. But some subjects do not lend themselves to this (e.g. astrophotography, which is what prompted this discovery for me). It may be that Lightroom just isn't the right tool anymore when this is the case. (I will go back and play with PV2 as an alternative.)

[Conrad_C]

A lot of it boils down to Lightroom Classic focusing on traditional still photography, single frames in isolation as noted. It’s not that they meant to break time lapses, but that time lapses are apparently not within the scope of what they intended to support…technically, there are no features in Lightroom Classic that specifically address time lapses.

I’m from the era where the primary way you edited photos was to edit tone curves in Photoshop. The Basic panel options in Lightroom appeared many years later, probably originally intended to be more user-friendly than Levels and Curves. Probably because of that, when you see newer photo apps being designed for desktop and mobile, their primary controls look more like the Basic panel, not the old Levels and Curves way.

Around 10-12 years ago two things happened. Camera sensors were able to record more dynamic range, but even before that, people started getting into merging bracketed exposures to create the original from of HDR, which was HDR information squeezed and tone-mapped down to SDR. With traditional tools, it can often be a challenge to make one linear adjustment look good at all points along the tonal range. Adobe may have started thinking that the Basic options could do a better job of handling these images, and found some new, advanced ways of compressing and tone-mapping long tonal ranges to SDR. I think maybe this is how the Basic panel in later Process Versions became more content-adaptive.

For example, below is a link to a 2011 research paper about advances in HDR processing, partially the work of an Adobe employee. From what I understand, this research was the basis for how image-adaptive controls improve processing in the more recent Process Versions in Lightroom and Camera Raw:

Local Laplacian Filters: Edge-aware Image Processing with a Laplacian Pyramid

All of this is a rather long-winded way to address the question “What’s the benefit of the image-adaptive options?” Compared to how things used to work, especially the Tone Curve, the current image-adaptive options can improve highlight and shadow areas far more quickly and effectively than, for example, how we used to try and get each segment of the tonal range right by adding multiple Curves adjustment layers and masking off each one. Now you just drag some Basic sliders, and it looks great and it’s done.

Just not on time-lapses or frame sequences meant to be viewed together…

But for those, that’s why they chose not to get rid of the Tone Curve and other linear options. They are still there if we need them, so we have lost nothing: We can continue to do it the linear way, or use an earlier Process Version, for the times when that works better.