Content Aware Fill doesn't seem to work with >100 IRE (super brights) HDR content

Report · Nov 06, 2021

Hi all -

I'm trying to remove some tourists wandering through my shot using Content Aware Fill. I've got the workflow down, all seemed to be working. The match lighting feature is a huge improvement for my sunrise shots with rapidly changing light (timelapse stuff).

Until I needed to do so on some HDR content. The created .exr files look fine...in SDR (CAF generated at full res, 32 bit, no proxies, etc). An area of super brights (greater than 100 IRE, greater than 1.0 on float scale, greater than 100% on SDR scale) looks fine in SDR, but when I reduced the viewed exposure, the super brights go grey - they clip at 1.0 on float/100 IRE, 100% on SDR scale. Oops. This appears to be a bug, and I could see how this would get missed in QC - CAF seemingly worked, and unless you check the exposure or view on an HDR monitor, you wouldn't catch it. I was making an SDR version by using the HDR Highlight Compression and that's when I saw it (and is also a good way to verify this result).

So:

-if have HDR content with greater than 100% brightness

-do Content Aware Fill on areas that include >100%

-the replaced areas with >100% are merely 100%, not 1000 nits or whatever the source was

-this is bad.

I've been using AE since it was CoSA in the 90s, I think I generally know what I'm doing.

So, a.) I'd love it if it were fixed, or b.) anybody know of any workarounds? I was thinking of reducing exposure on that area during CAF generation to bring all below 100% brightness, then adding exposure back in to the CAF .exr files to bring back up, but haven't tried it yet and I'm worried about amplifying any noise.

Report · Nov 07, 2021

Doubtful that this will ever work reliably with HDR content. Since it's AI-based, any such capability would be dependent on analyzing gigazillions of HDR footage, which is likely a pipe dream since no doubt it's all based on converted MP4s uploaded to the Adobe servers and such. That and of course there would still always be a level of uncertainty with any HDR content. You know, that old gag of trying to see "phantom neutrons" and interpreting what could be there rather than what actually may be there.

Mylenium

Report · Nov 08, 2021

I hear and understand your point, but I'm not sure it necessarily applies. There's no "phantom neutrons" involved, just brighter values than your monitor is currently configured to see. It's all the same math - why should it be different for a value of 0.999 than for 1.001? I'm hoping it is just an oversight during QC that someone forgot to check it. Compressor used to have the same problem (would clip all >100 IRE highlights when transcoding); hopefully fixable with an "Oh! We forgot to do that."