Hacker News Reader

Comment by LocalH | Hacker News Reader

I don't think so. Perhaps AI-upscaled? The footage looks legit and would track with the tube cameras that would have likely been used at that time. Although it sucks that it's deinterlaced to 30fps. Video like this really needs to be preserved without immediately throwing out half of the motion

View on HN

YouTube/Google doesn't give a shit anymore. They will mangle your video and audio and serve users "enhanced" "improved" "upscaled" encodes without telling you.

---

"YouTube secretly tested AI video enhancement without notifying creators" - Aug 25, 2025

https://arstechnica.com/google/2025/08/youtube-secretly-test...

"YouTube will let you opt out of AI upscaling on low-res videos" - Oct 29, 2025

https://www.theverge.com/news/808717/youtube-automatic-ai-up...

---

Not only that, YouTube turned many old videos shot in portrait (from the 2010s and before!) into engagement-boosting "Shorts" crap which get horribly mangled in their own way.

"Notice a weird beauty filter on Shorts? YouTube says it's on purpose" - Aug 26, 2025

https://www.androidauthority.com/youtube-shorts-upscaling-ex...

Yeah, this doesn't look like AI generated. It was probably filmed on super 8 film stock. The clothing, hair cuts, manufacturing process all scream early 80s.

I could see a cheap restoration introduction artifacts as a more likely reason for the look.

It looks more like video to me, honestly. It appears to be a smooth 30fps rather than the 18fps I'd expect from Super 8. There are also telltale stair-stepped sloped lines that demonstrate the effect of the deinterlacing. There did exist luggable 3/4" U-Matic recorders which I'd have to imagine CBS would have been using in 1980.

Yup, you're probably right. Author confirmed it was done by CBS news crew.

You're not "throwing out half the motion".

Given that PC monitors these days don't have an interlaced mode, how would you display it? Line doubling, so you're throwing out half the vertical resolution?

AFAIU it is possible to get nearly full resolution and full frame rate by using motion tracking to merge the fields. I.e. if the motion is regular enough, lines from the previous and / or next frame can be inserted instead of doubling the current lines. Which is almost the same process required to achieve best image quality at half the frame rate, just without throwing away frames.

You're not throwing away frames, though. You're blending the lines from the odd and even frames so that you get the full vertical resolution, with a certain amount of motion blur on moving vertical edges.

Analogue video is 25 frames per second, 50 fields per second. You could guess at what the "missing" lines in a field are, but that doesn't magically make it 50p video.

The effective rate for 50i or 60i video, in terms of motion fluidity, is 50 or 60fps. Just because it's half the resolution per field doesn't mean that it's not a "frame" of video in abstract terms, just that it's not a "frame" in the jargon (because fields by definition are half a frame).

It is impossible to decimate a video from 50/60i to 25/30p without losing half of the motion fluidity, even if the properly interlaced source video is technically 25/30fps.

> You're not "throwing out half the motion".

Yes, you are. The odd and even lines from proper interlaced footage belong to two separate moments in time, and so when you deinterlace from 60i to 30p you are unavoidably losing half of those moments.

> Given that PC monitors these days don't have an interlaced mode, how would you display it?

You deinterlace it to 60fps. There exist several algorithms to do so without losing motion fluidity.

> Line doubling, so you're throwing out half the vertical resolution?

Losing half of the vertical resolution of a 60i video is losing half the motion.

> You deinterlace it to 60fps. There exist several algorithms to do so without losing motion fluidity.

So what does "deinterlace it to 60fps" do? How does it work?

The most basic method is to "bob" deinterlace, which is more of a form of "interlace simulation" - you take the original fields, place each one in the appropriate image lines of a separate frame of video, then perform some type of interpolation of each frame for the in-between lines. More advanced algorithms exist like yadif, that do some amount of motion detection to determine which parts of the image require deinterlacing, and which parts can be essentially done with a "weave", which is just taking part of the image data from a combined 30fps frame containing two fields.

Okay, so what you're doing there is averaging the information between the two fields.

That gives you 25fps. It can only ever give you 25fps.

There are 50 fields per second, for 25 frames per second.

If you average the odd and even fields, you get 25 frames per second.

Untrue.

Let's suppose we fill odd fields with black, and even fields with white.

The end result when played back on a CRT will be the entire screen flickering between black and white with a lot of flicker, phosphor persistence notwithstanding.

If you simply average the even and odd fields, you will be left with the completely incorrect result of 50% gray and zero change between image frames.

You seem to be under the misconception that two consecutive fields contain image from the same moment in time. This is not necessarily true with proper interlaced video.

Reply on news.ycombinator.com