We’ve already shared videos where Dr. Hojatollah Yeganeh, Principal Video Architect at SSIMWAVE Inc., Steven Robertson, software engineer at Google, and Michael Zink, Vice President, Emerging & Creative Technologies at WarnerMedia, have voiced concerns about televisions ignoring static metadata, but we ran across yet another – a presentation given by Tyler Pruitt at SMPTE in 2017 – so in the interest of completeness, we thought we might as well share that, too!
After looking at how TVs tone map using static metadata (HDR10) and dynamic metadata (Dolby Vision), Tyler Pruitt discussed how high-end televisions deal with static metadata.
“The third [method] that’s becoming very popular, especially in high-end HDR televisions, is that manufacturers realized that static HDR10 metadata wasn’t optimal to tone map based on one piece of information, so the highest end TVs are now instead of looking at the static metadata, they’re actually real-time analyzing the frames, because as TVs are getting more and more powerful and they have way more interactive functionality, they have much more powerful GPUs and CPUs and they can write shaders and stuff like that to actually analyze each frame as it comes in and adjust their tone mapping or a color volume mapping based on that; but that does require more horsepower, so the lowest end TVs that most people are buying aren’t included in that. I was recently a part of a talk a summit and a guy from NPD talked about only 8% of televisions people spend more than $1,000, so 92% of people are spending under $1,000, so most of your content is gonna be seen on these lower end TVs so we need to figure out how to make those more accurate.” [slightly edited for clarity]
It was shortly after watching that presentation that we came across a random list of a dozen or so Ultra HD Blu-ray titles that had been mastered on a 4,000-nit display, but the MaxCLL figures were much higher than that, some even reaching 10,000 nits. The person sharing that list wrote,
“Mastering using 1000/4000 nits capable monitor a lot of times can contain brighter content from 1000/4000nits, just the colorist look the waveform and RGB histogram and leave un-clipped content above 1000/4000nits.” – Ted Aspiotis
We’d wager that those figures for MaxCLL above 4,000 nits (or even below 4,000 nits, for that matter) are no more than outlier pixels that tell us little about how the colorist actually graded the highlights. If the colorist is mastering on a Dolby Pulsar, anything above 4,000 nits is going to hard clip on the display. Unlike consumer TVs, that roll off highlights to avoid clipping, reference mastering monitors hard clip anything above the display’s peak luminance. It’s highly unlikely that a colorist is going to look at unsightly clipping on a Dolby Pulsar, then say to themselves, ‘Well, the waveform’s not clipping, so I’ll just leave it be.’
An explanation of outlier pixels, along with graphics showing the same kinds of astronomical numbers for content mastered on a 4,000-nit display, can be found here.
The Daejeon Chronicles 
Above 4000 nit outliers are just data points above 920. They should be as significant as you intend for them to be and not more. If your intention is to manage them rather than have the tv do this, e.g. Ghostbusters (2016), Resolve can tone map them to the intended target and embed the adjusted MaxCLL rather highpass the outliers. It’s a bit much to detail here. I can share if you are interested. HDR10 also has the option for scene by scene or frame by frame dynamic metadata called HDR10+. Resolve analyzes the clips to generate an XML sidecar file that can be embedded with x265. It’s the same process as Dolby Vision dynamic metadata except that for the latter, the sidecar file is binary. AppleTV, Netflix and Prime Video are supporting both standards. The relevant expert is Ben Waggoner, Principal Video Specialist for Amazon Prime Video.