Profile picture
Dan Kaminsky @dakami
, 19 tweets, 4 min read Read on Twitter
This is complicated, but at least it’s a technically analyzable situation. Warning everyone now, I am *absolutely* speaking in a neutral capacity here. I may not support your politics, or my own.

Summary: Video compression might have done it. It was probably edited.
First, the type of edit here claimed is not rare at all. I’ve generally heard it referred to as, well, “frame f***ing”. Happens to television shows quite a bit in syndication — scenes aren’t removed, but they’re sped up just enough to add an ad or two.
I say this because while the tech *exists* to do deep fakery, slicing frames here and there is a pretty common and well understood mechanism. It’s like an audio mixing app making the vocals louder (memories of Howard Dean).
But things get messy from here. The eye doesn’t see in 25 frames per second, because the eye doesn’t see in frames at all. It’s more a constant stream of detection events, somewhat compressed. You wouldn’t build a system like this, you just *are* the system like this.
But we approximate quite well with framing. There’s an art to it, it’s called cinematography. There’s also less artistic choices, just dealing with unifying different decisions, called telecine. An American movie filmed at 24fps looks fine on a British TV working at 50fps. How?
One way or another, some frames get duplicated, others get cut.

This has gotten infinitely more complicated in modern video compression formats. As one of my friends said recently, the Internet runs on the bandwidth left over from point to point video transmission.
Video compression formats are some of the deeper math fu humanity has ever deployed at scale, because the better we do here, the more streams we can push over existing pipes.

Sometimes the pipes are in space or under the sea. Hard to add more.
Let’s call the frame altering pattern, frame flicking. Well, there’s more than one way to flick a frame. Broadcasters and cinematographers have for years fought over how much bandwidth a frame reaaaaally needs. HD is not on or off — each frame, a system is deciding, how HD?
The metric is, how accurate does the reproduction have to be, to look similar enough to a human viewer. Nobody wants to just send *all* the pixels (which is what TV and movie theaters sort of could, if you squint) because 99% of the raw data is ignored by our eyes anyway.
(Yes, by our eyes. We’re probably not blasting tens to hundreds of megabits up our optic nerves, though I admit now I’m curious.)

So as video compression has gotten much more complicated, and the metric has become do people see things similar enough, algorithms are deciding.
They don’t know about higher layer implications, they’re “just math”, that encoded rules written by someone who wasn’t thinking about anything political. “Just psychophysics.”

We do not need machine learning to find ourselves at the debate about algorithmic effects on policy.
But back to this case. Directors hate compression artifacts, and they *really* hate frame flicking, because timing matters to humans, even at the small scale. Storytelling is a game of juggling the thoughts of others. Milliseconds matter in juggling.
So it’s important to know the claim of differential emotional reaction based on speed, is valid. Was the editing intentional.

Probably. Sorry, you want people to give you the benefit of the doubt, don’t lie all the time.

Did it have to be? Not necessarily.
It’s a reasonable assumption that frame flicking right at the moment of maximum emotional impact would be a remarkable coincidence. It would be. But it wouldn’t be coincidental. There’s a huge spike of information there, with an entire arm moving, hair moving, etc.
The same burst of data that catches our eye is the burst of data that needs to be encoded.

But wouldn’t that only change the contents, not the timing?

Oh, heavens no, video compression has worked across frames for ages. It’s all about “key frames” and “here’s what’s changed”.
If too much changes at once, maybe there’s a skip. Depends on the implementation. It’s not what I think happened, because I am a thinking adult who can see past plausible deniability claptrap.

Possible isn’t enough.

THAT BEING SAID.
In general, the absolute length of video segments should be maintained. The precise circumstances in which *this* toolchain was used to publish *this* video is inspectable. We could see pipeline recompression/telecine do this thing, on the original video feed. Or not.
Guaranteeing reproducibility of results will be tricky on this particular video, but if it naturally happens here it would happen all the time. Janky algorithms have enough jank for everybody.
The cultural matter of whether it matters exactly how many meters per second this arm moved or that frame was or wasn’t present — another debate. I will say, though.

Sorry, children of America. Please be better adults than we turned out to be. You’re not wrong, this show sucks.
Missing some Tweet in this thread?
You can try to force a refresh.

Like this thread? Get email updates or save it to PDF!

Subscribe to Dan Kaminsky
Profile picture

Get real-time email alerts when new unrolls are available from this author!

This content may be removed anytime!

Twitter may remove this content at anytime, convert it as a PDF, save and print for later use!

Try unrolling a thread yourself!

how to unroll video

1) Follow Thread Reader App on Twitter so you can easily mention us!

2) Go to a Twitter thread (series of Tweets by the same owner) and mention us with a keyword "unroll" @threadreaderapp unroll

You can practice here first or read more on our help page!

Did Thread Reader help you today?

Support us! We are indie developers!


This site is made by just three indie developers on a laptop doing marketing, support and development! Read more about the story.

Become a Premium Member and get exclusive features!

Premium member ($30.00/year)

Too expensive? Make a small donation by buying us coffee ($5) or help with server cost ($10)

Donate via Paypal Become our Patreon

Thank you for your support!