Fix mid shot 2 frame exposure shift
I'm at wit's end on this. It's a single shot for an indie feature that's already been cut, colored, and sound designed. Their colorist has moved away so I'm correcting this single shot, but I do have his .drp so I've got access to his full grade and the source clip.
The issue is that there is an in camera aperture adjustment down 1/4 stop, in the middle of a ECU with lots of bokeh out back. The change happens over two frames, and is uneven -- i.e. it drops ~75% in the first from and ~25% in the second. I've spent hours trying everything I know, and can't get this invisible. Here's a list of some attempts:
It seems that the dynamics of the image just changed too much because of the aperture change, an no matter how close I bring them, it still looks the frame shifts very abruptly. If anyone has advice, I'd love to hear it, as this shot just won't look natural.
[Nicholas Zimmerman] "If anyone has advice, I'd love to hear it"
Seems like your colorist had a good reason to move away.
Keyframing first and last almost never works.
Strategy 2 is where most of us wind up.
Curves may still come into play in this situation. There may be different curves inside a number of area qualifications for this to work.
The rest are pretenders.
Flicker fixer induces a lot, and I mean, A LOT of blur, especially if this is more than a little blip.
Not having seen the picture, and I'm imagining a person's face with some kind of depth-of-field background, perhaps at a different exposure level...
How much relative movement is there between the frames? Are there areas in the picture that change exposure at different rates? Are you on a particularly non-linear part of the camera's internal transfer/exposure characteristic?
Something that might be a shot-in-the-dark would be to export the offending frames and a few before-and-after as dpx files, name them non-consecutively so that Resolve doesn't treat them as a clip, and try its scene-match. Never tried it and its relative success is likely for pictures that are more different than they are similar, but you never know. The machines will do whatever they want.
Its never a case of simply adjusting gain down a few points -- the entire transfer curve changes with exposure and the only point at which you know you have succeeded is when you actually play them back and you don't see anything pop. For that reason correcting the hell out of them in still mode can never tell you whether it has worked or not -- even split screens are relatively useless. This kind of takes away from the inexperienced notion of "auto-correcting" anything by the numbers. Even if the numerical values match, your eyes don't know that and while everything in a scene is dynamically affected by everything else, its well established that even identical values within a frame will look different in comparison to what separates them. Those famous two patches of grey, for example -- one surrounded by white and the other by black -- one is alway perceived to be darker than the other. So this is another thing you have to defeat -- human vision.
"I always pass on free advice -- its never of any use to me" Oscar Wilde.
Virtually impossible to achieve an undetectable fix Nick. Because there are variables such as the input by the camera operator (hand speed, combined with the actual mechanics of the aperture itself, an aperture change is neither a linear event nor one that follows a predictable curve. You'd probably do better making a cut there, cutting in an extra good from the next good frame on both the head and tail sides of the cut, and maybe trying optical flow or from blending effects to merge the frames as best as you can.
Does this make sense???
David Roth Weiss
Director/Editor/Colorist & Workflow Consultant
David Weiss Productions
David is a Creative COW contributing editor and a forum host of the Apple Final Cut Pro forum.
Thanks to both of you for the recommendations. I think the idea I'm working on right now might solve it. I'm exporting the original frame and tracking it in with Mocha Pro. This will let me fade it out over a few frames and hopefully give it a natural look. The guy and girl are matched to the point where you can't see an adjustment on them, just the background. It won't be perfect, but hopefully I can draw it out over a few more frames.
Aperture stepping with still lenses doesn't always make for even light changes. I have had a lot of those in a recent doco and a combination of frame by frame keychanges plus the OFX Flickerfree filter from Digital Anarchy got the best result. But I noticed changes in light from top to bottom of the frame in the frames where the aperture was stopped down or up.
The other thing that changes is the highlight or low light detail which is extremely hard to compensate for even with curves.
Since its just a few frames and its giving you trouble with the methods you suggested I'd try replicating the frames rather than grading to match. E.g. use optical flow in AE/Nuke using twixtor or Kronos, or even Premiere Morph dissolve, Avid Fluid Morph, or resolve's new "smooth cut", all of which are designed to help smooth over jump cuts but which may be useful in regenerating these frames. worth a shot if your other approaches are not working.
Its also worth noting that if your grading in log (or ACES) this is easier than in a gamma encoded space like rec709 since changing exposure with lift/gain/offset are much more closely aligned with actual real world exposure shifts than in a gamma encoded space. Can still require some fitness with curves but much easier in Log, or even scene referred lin than in a gamma encoded space like rec709.
And in addition to the very good advice above, my observation is that there's a point where you have to say "this is as good as it gets." Cinematographers simply can't adjust aperture in the middle of a shot. And if they do, then it's up to the editor to be smart enough to just use a different take where this didn't happen. Otherwise, just smooth it out as best as humanly possible.
Non-linear changes like this tend to affect different parts of the frame in unpredictable ways, as you've discovered. I've worked for an hour on (say) a 9- or 10-frame sequence in trying to smooth out an exposure bump, and typically all I can do is just make it better... not undetectable. If it's to the point where it's 75% or 80% better than it was originally, the filmmakers are just going to have to live with it or come up with different coverage.
Quite often on issues like this, my brain goes the other way.
Again, not seeing the issue in full motion.
Can you comp the scene out. and cause the background to escalate to that highest point over say 15 frames. (gradually brighten the background over the course of 15 frames coming in and going out again) You might not be able to mask that those frames are brighter, but you might be able to turn it into part of the shot.