The Abyss 4K UHD (1989)

[Wilfredo]

Could someone please confirm if the special edition has Spanish subtitles or only the theater version has them?

[pjl69]

Originally Posted by weezer17

Sites should not allow preorders that exceed the stock they have.

Or they should operate on FIFO basis (First In First Out) when handling preorders.

[kmhofmann]

Originally Posted by mar3o

People are down on AI because of it's destructive potential.

Traditional upscaling was just math. As long as you did the math well, it worked. The math doesn't need to understand what faces are and what they look like, what a movie is, what depth of field is, etc.

And traditional restorations required knowledgeable people to manually make decisions about how to repair damaged film, restore colors, etc.

AI uses guesswork to "intelligently" "enhance" films, without having any actual understanding of what it's doing. Unlike sharpening, traditional DNR, and other filters, which again, just used math algorithms, AI attempts to make "intelligent" decisions about how to improve an image without having any understanding of what the image is. So there's strong potential for it to wreck havoc with images and films, as it analyzes each frame and guesses at what it should do with each pixel. Is that facial hair or scars? Are those teeth? What are teeth and hair anyways? What we call "AI" today is not really AI. It's in no way intelligent or capable of understanding anything from being self-aware (that's a good thing). It's just learned algorithms. It applies enhancements without knowing what it's actually doing or why. So results are wonky.

You want to play around with that in the labs, and mess around with random Youtube clips, fine. But don't use it on major beloved Hollywood film classics when it's in it's infancy and not up to the task of being used in such a way. The mere fact that they admitted that the AI algorithms got better between each Cameron film, with True Lies and Aliens being done before The Abyss and Titanic, demonstrates that it's not a mature technology yet and should not even be considered for commercial use until it's proven itself capable.

You could have maybe said the same thing about early CGi, as many a film has been ruined by poor cgi over the years. But those were brand new films, with the filmmakers willing to use their new films as a testing ground for new effects technologies. Sometimes it worked, many times it didn't early on. No different than many older otherwise great films that featured poor practical effects due to lack of budget or expertise.

These Cameron films are established films that have been enjoyed for decades, now being harmed retroactively by applying poor technology as a crutch or a replacement for doing things properly.

^^ This.

[kmhofmann]

Originally Posted by Steedeel

Haven’t you seen the new Sora demos of human faces and humans in various complicated environments? It’s a huge jump.

So? We've all seen that and it's very impressive but that has nothing - I repeat, absolutely nothing - to do with applications to film restoration.

[sojrner]

Originally Posted by Steedeel

Haven’t you seen the new Sora demos of human faces and humans in various complicated environments? It’s a huge jump.

None of this discussion about AI in film transfers has anything to do with whether a game looks good or not, or whether random (and pointedly missing particular movements) animation is neato. To wit, the points you aren't understanding:

• So-called AI is no different than any other generating/upscaling/denoising/sharpening/repairing/re-graining algorithm in practice, it just typically has more data to play with.
• No matter how said algorithm is used in areas like gaming or animation, it is still far from transparent for live-action movies, tromping all around the uncanny valley.
• Regardless of the leaps in tech for computer-generated anything, when transferring any movie (even digital sources) to a new medium, all processing (regardless of the marketing label on it) should be minimally applied, and never on a set-it-and-forget-it basis.

[kmhofmann]

Originally Posted by sojrner

• So-called AI is no different than any other generating/upscaling/denoising/sharpening/repairing/re-graining algorithm in practice, it just typically has more data to play with.

Exactly. What is described by many here as "AI" is nothing more than a huge, ultimately stupid mathematical function that uses more context and prior knowledge than most previous algorithms.

For gaming, asset generation, any other form of image or video generation - granted, these algorithms can be very useful there and produce impressive results. Even though I think it will take a few more years before we can mostly cross the uncanny valley w.r.t. realism and fidelity.

But again, Sora and similar algorithms are completely off-topic here.

For the purposes of film restoration, it's especially these above-mentioned implicitly encoded priors that can be - and, as we can see, are - very destructive, effectively hallucinating detail that has never been there and that should not be there. It's a false look that is not only ugly but flies in the face of film restoration and preservation.

Just like smartphone cameras have all kinds of heavy "AI" processing on-board but will never be able to truly beat large-scale sensors coupled with the right optics, any kind of AI-upscaled mediocre transfer will never look as good (*) as a proper, new scan with just the right (minimal) level of processing, besides restorative measures. With 'fidelity to the true source' being the key metric in both examples.

[LordoftheRings]

Strap yourself real tight in your chair with chromed steel chains ...

The Abyss 4K UHD Review - Are The Picture Concerns Valid? - YouTube

The Abyss 4k Ultra HD Bluray Unboxing & 4k Review. - YouTube

[Lord Method Man]

Originally Posted by kmhofmann

But again, Sora and similar algorithms are completely off-topic here.

Exactly. These movies weren't created using AI. They were shot on film using real human actors on real sets, and that's how I expect them to look.

[rubystone356]

for those still looking for a copy. may try Barnes and Nobles. I recently got the Conan 2 pack from them.
https://www.barnesandnoble.com/w/dvd...=0786936900798

Jacob

[MatOfTheDead]

Originally Posted by weezer17

I preordered from Amazon on March 11 and still haven't gotten it. Sites should not allow preorders that exceed the stock they have.

same here i ordered aliens and true lies at the same time (part of the same order) and they got here saturday

[milo bloom]

Originally Posted by rubystone356

for those still looking for a copy. may try Barnes and Nobles. I recently got the Conan 2 pack from them.
https://www.barnesandnoble.com/w/dvd...=0786936900798

Jacob

I checked the app and it won't let me order it but it says it's in stock at some locations in the suburbs, but none near me and I'm going out of town this weekend.

Not dying to own this, I've already waited 20+ years, but it's a little annoying having to hunt for it like this.

[cochon]

Originally Posted by Steedeel

And I have no idea what gaslighting is, but it sounds like one of those dreadful social media phrases that the kids say, so I won’t research it.

It's actually pretty old.

[milo bloom]

Originally Posted by Steedeel

And I have no idea what gaslighting is, but it sounds like one of those dreadful social media phrases that the kids say, so I won’t research it.

.

Originally Posted by cochon

It's actually pretty old.

Don't listen to him, there's no such thing as gaslighting. Just like there is no war in Ba Sing Se.

[deadmanjones57]

Originally Posted by Steedeel

And I have no idea what gaslighting is, but it sounds like one of those dreadful social media phrases that the kids say, so I won’t research it.

Yeah you do. Everybody knows what gaslighting is.

[DaylightsEnd]

Originally Posted by sojrner

You take a real human face and make it look like game graphics, and no matter how much you sprinkle new grain on it, it's still not right

Tru dat. No such thing as artificial fidelity.

[Steedeel]

Originally Posted by cochon

It's actually pretty old.

Originally Posted by milo bloom

Don't listen to him, there's no such thing as gaslighting. Just like there is no war in Ba Sing Se.

Originally Posted by deadmanjones57

Yeah you do. Everybody knows what gaslighting is.

Ok, I just looked it up, and it makes absolutely no sense in this context. I’m not trying to bend reality, or manipulate people, we are all adults here, film fanatics and collectors, people know there own minds, all I am doing is presenting a viewpoint, in the same way haters of these latest JC disc releases are expressing their opinion. Many members have expressed their opinion and posted screenshots of whatever grievance they have with A.I.
It’s an echo chamber if everyone feels the same way, and if it’s fine to post multiple times to prove the point, and taunt people who don’t hate it, why is it not fine to have views that are different? (I loved the look of Titanic and The Abyss for example)

Regardless, once personal attacks start, I think it’s time to bail.

This thread has been better than that, with different opinions, but still respect and humour up until now.

Will post my review of True Lies next week and will then just leave folk to it.

[Azurfel]

Originally Posted by Steedeel

Having sampled gaming that has been A.I upscaled (with other special sauce to boot)

Nvidia's DLSS was trained by refining it's results using comparison to native 16K super sampled equivalent frames with the goal of making the final result look as close to that native 16K super sampled image downsampled to the target output resolution as possible.

The Park Road Post process is, by all appearances, a LLM based successor to the Lowry process, trained to dumbly extract and combine what it registers as detail from adjacent frames while eliminating (or at least minimizing) undesired noise included in the source material (including grain).

They are doing fundamentally different things using superficially similar methodology, and there is no reason whatsoever to presume that the Park Road process will get better at making a 2K (or lesser) scan look like a 4K scan, because as far as we know that isn't remotely what it is trying to do in the first place.

[kmhofmann]

Originally Posted by Steedeel

It’s an echo chamber if everyone feels the same way, and if it’s fine to post multiple times to prove the point, and taunt people who don’t hate it, why is it not fine to have views that are different?

At the very least, you seem to be misunderstanding, or grossly overestimating the potential of the technologies that you are talking about.

It's okay to be excited about what's going on in generative AI research - hey, I am too, and have a good stake in the progress of the field.

Originally Posted by Steedeel

We live in an age where machine learning can generate any video scenario you can imagine from a text prompt, I don’t think anyone can conclude that the case is closed.

But this is a gross extrapolation.

First of all, we can't do what you claim is possible. Current algorithms can generate some videos that appear to look realistic enough at first glance. Outside of very few companies that fiercely guard their trade secrets, the field is a lot less studied. Publicly shown examples are likely to be very cherry-picked (things would probably look a lot glamorous otherwise), and there is currently no evidence of mature technology that makes these algorithms sufficiently controllable for fine-grained video generation or editing.
This is the domain of early research, not products.

Sure, you can condition on some text prompt, but the end result is still dependent on things even the creators of the algorithms cannot control in a satisfactory manner. And yes, this situation will get better over time, but I encourage you to a) not overstate what is actually achievable with current technology, b) not overextrapolate the potential, c) especially with respect to the use cases we are talking about here. They are quite distinct from the technologies you mention and, in my humble opinion, require a different approach, not just from a technology perspective.

Do not conflate these things based on your general enthusiasm for these technologies.

SOTA approaches for superresolution != SOTA approaches for video generation. And I don't even think that "run-off-the-mill" superresolution is what we need for film restoration.

Also, what Azurfel said. ^^

[kmhofmann]

Originally Posted by Azurfel

The Park Road Post process is, by all appearances, a LLM based successor to the Lowry process

It's probably not LLM-based (cause, what role would language play here? Or, more generally, foundation models that even Park Road probably hasn't trained-from-scratch or fine-tuned?) but it's likely to be some custom implementation of some task-specific design based on some of this type of research, maybe with some custom data.

[chip75]

Originally Posted by Ben_UK

They should have seen it coming frankly.

Even if they did the pressing plant is stretched to the limit, they probably couldn't get another 50,000 copies if they wanted to.

The Mexico plant seems overworked to where the quality control has dropped significantly.