Personally, it's not really any better than mpv filters like NIS [0] (Nvidia Image Scaling, which does not use AI but is somehow more sharp than these filters, likely because Nvidia used what they learned from AI upscaling to create a better non-AI upscaler; AMD's FidelityFX Super Resolution is also available as an mpv filter on the same page at [0]) as well as Anime4k [1] (which does use AI but only in the training phase to create the initial glsl file which then does not use AI inferencing but plain old image filters).

Combine mpv with its yt-dlp [2] and SponsorBlock support [3] and you have a competent way to upscale your own YouTube (and other types of [4]) videos.

If you really want to upscale with AI (albeit not at runtime), I've found Topaz Video AI to be the best so far, much sharper than other solutions on the market.

[0] https://gist.github.com/agyild

[1] https://github.com/bloc97/Anime4K

[2] https://www.funkyspacemonkey.com/replace-youtube-dl-with-yt-...

[3] https://github.com/po5/mpv_sponsorblock

[4] https://github.com/yt-dlp/yt-dlp/blob/master/supportedsites....

AMD's FSR filter for mpv I find is better cpu-usage wise and in terms of quality:

https://gist.github.com/agyild/82219c545228d70c5604f865ce0b0...

I found it on be worse than NIS if you have an Nvidia GPU. If not, then FSR should be good enough.

I would hope with all these new technologies there is a "present the original" (or "naked version," "unmodified, unadulterated") version toggle so that while we can enjoy modern convenience we also have the option to visit actual reality from time to time.

I would hate it if one day accessing the original versions of things (whether books, videos, history, event reality, etc.) becomes an impossibility and we live in a synthetic world.

Famously, upscaled Obama: https://twitter.com/Chicken3gg/status/1274314622447820801

I'm with you, but I fear that horse is already out of the barn. I picture someone 20 years from now trying to find out what their parent really looked like when they were young. The obviously smoothed-out face filters are already giving way to AI-powered homogenization. And the filtering is moving deeper down the stack from the app to the camera itself. There will be no "original".

I really agree. All the latest rage is to upscale the older pre-digital animated tv and movies. While it should often be done, it always changes them enough to make it noticeably different than the original experience. A lot of people don't care, but it bothers the archivist in me.

I even find it largely worse than not upscaling. The smaller the image on the retina, the higher the density of detail, and the higher the perceived sharpness. I don't mind the small viewport — the brain is already adapted to concentrate on just the fovea when looking closely at things.

Downscaling is also a substitute for frame interpolation. 24-60 fps video becomes noticeably smoother, but the effect is weaker with animation ≤8 fps. Why does this even happen? That less of an absolute difference between frames is less jarring makes intuitive sense. However, I can't find any research, only the seemingly single forum thread where it's been discussed on the internet.*

Finally, I contend that even with a high-resolution source, a smaller viewport is not necessarily a worse experience, merely a different one. Because peripheral vision is so poor, by looking at a smaller image, you see the 'bigger picture' faster, without having to move your eyes. It's most noticeable in fast montages where you'd rewind to see something that you missed.

See also the opposite: "attention deconcentration"

PS: Downscaling below 1:1 increases sharpness with little detail loss when the original resolution is higher than the actual level of detail, owing to the likes of intermediate upscaling, optical limits, or compression.

* https://forums.blurbusters.com/viewtopic.php?t=380

I noticed this on a music video. Upscaling has done something odd to the faces, which just makes everything look uncanny.

I agree, most of the time it's fine. I agree, people should get the choice of the original vs the upscaled.

I just went and rewatched Star Trek TNG in the 1080p version. It is not upscaled, but scanned from the original film. It looks fantastic, but a lot of the special effects really stand out. The star fields when looking out windows have a totally different look it's almost like you can tell there is just a wall behind the glass with a bunch of lights stuck on it. And the painted landscapes similar to the original series are really given away. I bet some people hate the 1080p and some people now hate the original SD version. But I am glad both exist.

Gen Alpha won't see the difference. Media will be fungible the way drinking water is.

If they made it a DirectShow filter, then this would work for more than Edge, but that would go against their marketing desires of being able to say "only on Edge"...

Or perhaps that's already how it's implemented, but gated behind some checks that it's running inside Edge. I absolutely loathe this trend of taking a powerful and generally useful feature and severely crippling it for marketing reasons.

> If they made it a DirectShow filter,

They won't because DirectShow is deprecated and slated for eventual removal[1][2], regardless its practical value.

[1]: https://en.wikipedia.org/wiki/DirectShow

[2]: https://en.wikipedia.org/wiki/Media_Foundation

"eventual" in Microsoft terms is a very, very long time, and I can safely say that a lot of DirectShow-based stuff has specialised hardware behind it too. Of course the core of my argument doesn't change if I replace DirectShow filter with its Media Foundation equivalent --- generic video processing functionality available to all applications on the platform, vs. being bound to one and restricted to a narrow use-case.

I can kind of see why this might be Edge-specific than Windows-specific, since Edge is platform agnostic (it's just Microsoft Chrome, after all) and having the Windows implementation be DS/MF-based would be repeated, additional work from a broader, Windows agnostic perspective.

It would make sense from a "Windows first class citizen" perspective and I prefer Windows's "operating system is the central datastore" approach to media filters, of course. Linux's "BYOF" ("Bring Your Own Filters") approach sucks arse.

You could use Nvidia Image Scaling on mpv, which I mentioned in my other comment with more details: https://news.ycombinator.com/item?id=35051351. I find it better and sharper than Edge's currently.

Will we have a new wave of conspiracy theories now?

Look what AI found in .. the Zapruder film (4k, 120fps) or What NASA and Kubrick hid from public in plain sight: Moonwalk (upscaled, HDR) shows sound stage and equipment in the crushed shadows.

I am not saying it's Aliens but ... it's Aliens!

This was a legal issue during the Kyle Rittenhouse trial. It wasn't even AI upscaling it was simply how modern desktop/video software fills in details via smoothing in between frames when you slow something down or process it.

So evidence pushed by one of the legal teams which tried to use a highly zoomed in shot on an iPad of a CCTV camera of a parking lot to prove (or disprove I can't remember) a gun was held was called into question. The lawyers had an IT expert which helped dismiss it.

So when it comes down to it when it matters there's still sanity.

Conspiracy theorists also didn't need tech to believe in nonsense before anyway. So it's mostly shades of grey.

when we see how many bad pictures and videos each of us can take just to get one good one, and the crazy expressions on our faces, I'm highly skeptical of "evidence" photos. And slow motion might be handy for seeing particular details like "knee touches ground, ball out of bounds", but otherwise slow motion is highly misleading, making fleeting facial expressions look intentional, etc. If slow motion is used for evidence, it should always be followed by several viewings at normal speed to unconvince you of false impressions.

I tried it with the SICP 240p videos.The result is not that great (probably about as good as Sony Bravia's TV's upscaling).

Sounds cool but isn't it somewhat lazy to not include at least a picture in the article of what you're talking about?

Blogspam. Direct link: https://www.microsoftedgeinsider.com/en-us/whats-new

>It's really difficult to repair reputational damage.

The people who scream "Micro$oft" and "Windoze" are a vocal minority, most respect Microsoft and Windows for what they achieved and contributed (and continue to) to personal computing.

Even the whole IE6 debacle was, in hindsight, Microsoft demonstrating hands on what happens when a monopoly is left to its own devices; a lesson the personal computing world apparently failed to appreciate given the monopoly that is Chrome.

Incidentally, the Chrome monopoly is poised to further its grasp with what looks like the opening up of iOS and the subsequent death of Safari. Netscape and Firefox are rolling in their graves.

> As for the actual feature, upscaling lower-res videos is nothing new. Even DVD players did this years ago

No, it's very new. It's not upscaling, it's machine-learning based super resolution.

It adds details and makes textures richer and sharper, etc. DVD players never did anything like that.

It's pretty awesome technology. I applaud Microsoft for releasing this.

> They did this to destroy Netscale and when Microsoft feared the Web would destroy their Windows monopoly.

To be fair, that's exactly what Netscape was saying; that the web would make Windows obsolete. In the end, they were probably right.

> Even things like IE having its own interpretation of the CSS box model that differed from everyone else.

Ah yes, but Microsoft was right about that one.

> They did this to destroy Netscale

It must have worked. You don't remember their name.

Was excited about this till the 720p limit part. Lets hope that changes to 1080

Opera already has this

All that opera does it put a CSS sharpening filter onto the video element.

Hold my beer...