Hacker News new | ask | show | jobs
by dmitrybrant 800 days ago
My biggest gripe with these AI film enhancements is that they are adding information that was never there. You're no longer watching the original film. You no longer have a sense of how contemporary film equipment worked, what its limitations were, how the director dealt with those limitations, etc.
10 comments

I don't think that's universally true of all AI enhancement though. Information that is "missing" in one frame might be pulled in from a nearby frame. As others have pointed out, we are in the infancy of video enhancement and the future is not fundamentally limited.

If that takes away from the artistic nature of the film I understand the complaint, but I look forward to seeing this technology applied where the original reel has been damaged. In those cases we are already missing out on what the director intended.

In part, we need more vocabulary to distinguish different techniques. Everyone is just "AI" right now, which could mean many different things.

Standard terminology would help us discuss what methods are acceptable for what purposes and what goes too far. And it has to be terminology that the public can understand, so they can make informed decisions.

I also think we already have a bunch of old words for the techniques, like "upscaling" or "statistics". It's "AI" everything now but the old words for the old techniques are waiting to be used again.
> Information that is "missing" in one frame might be pulled in from a nearby frame

Yeah - does anyone know if anyone is actually doing this? Like some sort of DLSS for video? I'd love to read about it.

If there is a movie which is only shot in 1080p and I have a 4k TV, it seems like there’s three options. One, watch it in the original 1080p with 3/4 of the screen as black border. Two, stretch the image, making it blurry. Three, upscale the image. If you give me the choice, I’m choosing 3 every time.

Sorry if it sounds crass, but I feel the process of shooting the movie is less important than the story it is trying to tell.

Upscaling exactly 2x is also an option.
That’s what I meant by option 3?
How do you differentiate between "stretch" and "upscale"? Or how would "stretching" by exactly 2x would make the image blurry?
Upscaling algorithms vary from the extremely basic to the ML models we see today that straight up replaces or adds new details. Some of the more naive algorithms do indeed just look blurry.

https://en.wikipedia.org/wiki/Comparison_gallery_of_image_sc...

Most people don't care. Photographers had a real great time pointing out that Samsung literally AI replaced the moon, but some Samsung S21 Ultra users were busy bragging how great “their” moon pictures turned out. Let's judge AI enhancements like sound design: Noticeably good if done well, unnoticeable if done satisfactory and noticeably distracting if done poorly. The article shows a case of noticeably distracting, so they're better off with the original version.
It's a fundamentally different concept of photography though, one that becomes more similar to a painting or collage than a captured frame of light. Regardless of the merits of one over the other are for the purposes of storytelling, it's a bit worrisome when the distinction is lost on people altogether.
I get how a film buff might care, and agree the original version should be available, but isn’t there space for people who just want to see the story but experience it with modern levels of image quality? The technical details of technology at some point of time is definitely interesting to some people, but as say the writer or others associated with the creative and less technical aspects of a film I may find the technical limitations make the story less accessible to people used to more modern technologies and quality.
What does "modern levels of image quality" mean in this context?

The article is about AI upscaling "True Lies", which was shot on 35mm film. 35mm provides a very high level of detail -- about equivalent in resolution to a 4k digital picture. We're not talking about getting an old VHS tape to look decent on your TV here.

The differences in quality between 35mm film and 4k digital are really more qualitative than quantitative, such as dynamic range and film grain. But things like lighting and dynamic range are just as much directorial choices as script, story, any other aspect of a film. It's a visual medium, after all.

Is the goal to have all old movies have the same, flatly lit streaming "content" look that's so ubiquitous today?

I think the argument against "isn’t there space for people who just want to see the story but experience it with modern levels of image quality" is that such a space is a-historical -- It's a space for someone that doesn't want to engage with the fact that things were different in the (not even very distant) past, and (at the risk of sounding a bit pretentious) it breeds an intellectually lazy and small-minded culture.

The problem with that is the content is usually shot with the certain definition in mind. If you don't film certain scenes from scratch, they can end up looking weird in higher definition, simply because certain tricks rely on low definition/poor quality, or because you get a mismatch between old VFX and new resolution, for example.

It's a widespread issue with the emulation of old games that have been made for really low resolution/different ratio screens and slow hardware, especially early 3D/2D combinations like Final Fantasy, and those that relied on janky analog video outputs to draw their effects.

For a specific simple example: multiple Star Trek TV series were shot with the assumption that SDTV resolution would hide all the rough edges of props and fake displays. Watch them in (non-remastered) HD and suddenly it's very obvious how much of the set is painted plywood and cardboard.
One somewhat funny example of this is in the first ST:TNG episode "Encounter at Farpoint". In one shot, the captain asks Data a question, and the camera turns to him to show him standing from his seat at the conn and answering. At the bottom of the screen, it's plainly visible (in the new Blu-Ray version) that a patch of extra carpet is under the edge of the seat. It was probably put there to level the seat or something. At the time, this was ignored, because on a standard SDTV screen, the edges are all rounded, so the very edge of the frame isn't normally visible.

Another thing that's plainly obvious in TNG's remastered version is all the black cardboard placed over the display screens in the back of the bridge, to block glare from lights. In SDTV, this wasn't noticeable because the quality was so bad.

Actually I would expect AI up scaling of SDTV in this case would perform better. It would assume semantically the props were real and would extrapolate them as such.
For anything that's not just "grab a camera and shoot the movie" the format that it is shot in is absolutely taken into account. I don't think you can separate the story from how the image is captured.
One perspective:

'Film buff' responses are common to every major change in technology and society. People highly invested in the old way have an understandably conservative reaction - wait! slow down! what happens to all these old values?! They look for and find flaws, confirming their fears (a confirmation bias) and supporting their argument to slow down.

They are right that some values will be lost; hopefully much more will be gained. The existance of flaws in beta / first generation applications doesn't correlate with future success.

Also, they unknowingly mislead by reasoning with what is also an old sales disinformation technique: List the positive values of Option A, compare them to Option B; B, being a different product, inevitably will differ from A's design and strengths and lose the comparison. The comparision misleads us because it omits B's concept and its strengths that are superior to A's; with a new technology, those strengths aren't even all known - in this case, we can see B's far superior resolution and cleaner image. We also don't know what creative, artistic uses people will come up with - for example, maybe it can be used to blend two very different kinds of films together.

These things happen with political and social issues too. It's just another way of saying the second step in what every innovator experiences: 'first they laugh at you, then they tell you it violates the orthodoxy, then they say they knew it all along'.

Feels like people have had over 20 years to move on from the narcissistic butchery Lucas did to Star Wars IV, but it seems we're still at Step 2.

Maybe ... just maybe ... Step 2 is where it stops sometimes because it was a bad idea and did make the films worse.

Where would you draw the line though? What is acceptable non-AI remastering?

I'm 99% confident that similar issues were raised with e.g. recolored films, HD upscales, etc.

I draw the line at edits that consider semiotic meaning. Edits are acceptable if they apply globally (e.g. color correction to compensate for faded negatives), or if they apply locally based on purely geometric considerations (e.g. sharpening based on edge detection), but not if they try to decide what some aspect of the image signifies (e.g. red eye removal, which requires guessing which pixels are supposed to represent an eye). AI makes no distinction between geometric and semiotic meaning, so AI edits are never acceptable.
Easy counterexample: dumb unsharp masking will ruin close-up scenes that are shot for softness and/or have bokeh. ML upscalers can do this too when applied mindlessly. But you can also train an upscaler on the same type of footage, or even on the parts of the same footage available in higher definition. Even if you don't, matching the upscaler with the intent behind the content is your job.

The separation you're talking about is imaginary, the line doesn't exist. Any tool will affect the original meaning if it doesn't match the execution. Remastering is an art regardless of the tool, and it's always an interpretation of the original work. It's fine to like or dislike this interpretation.

Remastering can screw up intent with something as simple as color grading.

But there is a line here. An editor that's using simple tools knows exactly what they're changing, and if they're using simple frame-global tools then they're not introducing anything that wasn't already there.

If you throw an AI at things, it will try to guess what things in the image are, and make detail adjustments based on that.

So that's three categories of edit, easily distinguished: human making frame-global changes, human deliberately changing/adding details, AI changing/adding details in a way that's basically impossible to fully supervise.

It sounds like they accept category 1 in remastering, even though it's not foolproof, and reject 2 and 3.

No, "AI" is absolutely not uncontrollable magic that does something you don't want sometimes. It's not an issue really, you always have arbitrarily granular control of the end result, with ML tools or not. You can train them properly, you can control the application, you can fix the result, you can do anything with it. It's the usual VFX process, and it's not the only tool at your disposal.

The problem is that remasters don't make a lot of money, so instead of a properly controlled faithful representation (or a good rethinking) it's typically a half-assed job with a couple filters run over the entire piece. Another issue is that you now have two possibly conflicting intents - one from the original and another from the remaster. ML haven't changed anything in here, it's always been like that.

You can't tell what specific textures it's trying to replicate when it's enhancing detail. It's not magic, but the specifics are uncontrollable.
Yes, back in the mid-late 80s Turner Entertainment colorized a huge number of old films in their vaults to show on cable movie channels. It was almost universally panned. It was seen at first as a way to give mediocre old films with known stars a brief revival, but then Turner started to colorize classic, multi-award-winning films like The Asphalt Jungle and the whole idea was dismissed as a meretricious money-grab.
> how contemporary film equipment worked, what its limitations were, how the director dealt with those limitations, etc.

Non-film buffs, i.e. most viewers, don't care about this.

Any art and/or media production executed well enough to be culturally significant rests on an enormous depth of artistic and technical choices that most audiences have zero awareness of—and yet, if you took them all away, you would have nothing left. Every change takes you further from the original artist's vision, and if all you want to do is Consume Slop then that's fine I guess, but the stewards of these works should aim higher.
Well there are movies which were technically well executed with poor stories, and great stories with poor execution. And there were movies which did well in both areas.

For example Tenet. Cool story, poor audio mix. (I don’t buy the explanation that Nolan had any reason other than expediency for this.) If we use “AI” to fix the audio after the fact, that’s a win in my book.

I’m not a film buff or a purist though. I watch movies with subtitles which is certainly not what the director had in mind, but that’s ok.

I agree, most people watch the movie for the story that unfolds. Few are looking at things like framing the subject, the pull of the focus, subtle lighting differences between scenes, they are interested in the story, not the art of filmmaking. The people offended by this are the ones that are crying about the art being taken out of it.
> Non-film buffs, i.e. most viewers, don't care about this.

... consciously ...

Assuming competent cinematography, it will have an effect on the viewer whether they can analyze it or not.

The film grain will have no effect if it's not visible due to image/stream compression such as when the viewer sees the film on a video streaming service. HDR wont show up for most viewers. Details you need more than 1080p to see won't show for many (most?) viewers ... so I'd dispute your "will have an effect" here.

Good storytelling (and probably blunt spectacle) is the only thing common to all viewers that can win them over. For mainstream media everything else is gloss that may have no effect.

Most people don't even have their sound/brightness/contrast well-adjusted. Some free-to-view services regularly air content with the wrong ratio (and I've seen people happily sit watching the wrong ratio seemingly oblivious to it).

Yes, media nuances can have an effect on the unwitting, but I suspect much doesn't even have opportunity to.

> The film grain will have no effect if it's not visible due to image/stream compression such as when the viewer sees the film on a video streaming service. HDR wont show up for most viewers. Details you need more than 1080p to see won't show for many (most?) viewers ... so I'd dispute your "will have an effect" here.

You're going too low level, I'm thinking of lighting and colour and intentional blur via adjusting focus.

> Good storytelling (and probably blunt spectacle) is the only thing common to all viewers that can win them over. For mainstream media everything else is gloss that may have no effect.

You really need to reverse spectacle and storytelling in this statement. How else can the box office be dominated by superhero movies that personally I ... just ... can't ... tell ... apart?

They do but they dont know.

See new Netflix show Ripley.

All shot in B&W, beautifully shot.

The originals still exist and you’re free to watch those instead.

This just provides a new way to watch older movies should you choose to do so. Or not.

> The originals still exist and you’re free to watch those instead.

This is far from certain, unless "you" are willing to engage in piracy. It's often difficult or impossible to legitimately buy (or even rent) the original, unadulterated versions of older films.

<cough> Try buying the original versions of Star Wars... and that was before "AI".
To watch Star Wars as it was originally you need to break US law.
Its definitly an interesting point you bring up with the contemporary film equipment.

Nonetheless, i do believe that most film makers are actually want to make a film not work around contemprary limitations.

Do most people care? I just want to eat popcorn and watch a movie.
How do you feel about extended cuts?