GPT-4o: AI in its current state appears to be more of a detriment than a benefit for high-quality video restoration, particularly for classic films that require careful preservation.

Question 1 of 4 for ChatGPT-4o: Please read the transcript of the video "AI Can Ruin Movies Now, Too - Aliens and True Lies on 4k" and list all the failures of AI mentioned in relation to video restoration.

ChatGPT-4o’s answer:
The transcript of the video "AI Can Ruin Movies Now, Too - Aliens and True Lies on 4k" mentions several failures of AI in relation to video restoration. Here is a list of those failures:

1. Grain Reduction Issues:

   • Excessive grain reduction turning characters into cartoons.

   • AI noise reduction removing detail instead of preserving it.

2. Color Alteration:

   • Overuse of teal, resulting in loss of original colors.

   • Odd color banding and aliasing issues.

3. Detail Preservation Failure:

   • AI algorithms failing to preserve original details, instead generating and smearing artificial details across the screen.

   • Hyper-smoothing of faces while leaving grain intact.

4. Temporal Stability Issues:

   • Inconsistent sharpness and blurriness of faces from one moment to the next.

   • Features like eyelids sharply contrasting with otherwise blurred shots.

5. Facial Distortion:

   • Faces becoming distorted with artifacts like burn scars, porcelain cracks, and unnatural sharp lines.

   • Actors' faces rendered with bizarre details, such as googly eyes and waxy textures.

   • AI-generated features making faces look like they are in the early stages of “Emperor Palpatine-ism”.

6. Motion and Lighting Problems:

   • Synthetic details remaining uncannily sharp even during motion blur.

   • Faces melting and warping under unusual lighting conditions or obstructions.

7. Artificial Enhancements:

   • Hair appearing like a video game render with digital wigs.

   • Synthetic details like eyelids being overly sharp and unnatural.

   • AI-generated facial expressions looking cold and mechanical, distorting basic human emotions.

8. Failure in Distant Faces:

   • Distant faces rendered as if possessed by a “squishy demon”.

9. Restoration Inconsistencies:

   • AI unable to maintain consistent detail across different scenes.

   • Noise reduction mistaking grain for clothing or skin detail, resulting in frozen pixels.

   • Over-sharpening causing deep lines and halos around objects, making real stunts look green-screened.

10. HDR and Color Spectrum Mismanagement:

    • Use of SDR+ instead of true HDR, resulting in underwhelming brightness and contrast.

    • Limited use of the wide color spectrum, mostly confined to the 709 color space.

11. Unrealistic Close-ups:

    • Close-up shots rendered with excessive sharpness, making actors look unnaturally aged or textured.

12. Misleading Advertising:

    • HDR listed as a feature on the box but not fully utilized.

    • Color spectrum improvements not substantial, often not surpassing blu-ray capabilities.

13. Overall Inadequate Quality:

    • AI-generated restorations deemed worse than existing blu-ray versions.

    • Use of old 2k masters instead of new 4k scans, resulting in subpar quality.

These failures highlight the challenges and drawbacks of using AI for film restoration, emphasizing the need for careful application and quality control.

Question 2 of 4 for ChatGPT-4o: Which ones could be unfixable do you think, and why?

ChatGPT-4o’s answer:
The unfixable issues related to AI in video restoration primarily stem from fundamental limitations of the technology and the irreversible nature of some alterations. Here are the issues that might be considered unfixable and the reasons why:

1. Detail Preservation Failure:

   • Reason: Once the AI smears artificial details over the original content, it can be challenging, if not impossible, to retrieve the lost original details. AI guesses at what should be there rather than preserving actual data.

2. Facial Distortion and Synthetic Details:

   • Reason: Distorted faces and synthetic details added by the AI replace the true likeness of the actors with false, generated features. Reverting these changes would require going back to the original scan, assuming it is still available and usable, and redoing the restoration without AI interference.

3. Temporal Stability Issues:

   • Reason: The inconsistency in sharpness and blurriness from frame to frame introduced by AI is a byproduct of the algorithm's inability to maintain consistent detail. Fixing this would require reprocessing the entire film with a more advanced, consistent approach, which might not yet exist.

4. Over-sharpening and Halos:

   • Reason: Over-sharpening creates artificial edges and halos around objects that are not present in the original film. These artifacts alter the visual integrity of the scenes, and removing them without losing essential details is highly complex.

5. Misleading HDR and Color Spectrum Mismanagement:

   • Reason: If the AI-based restoration process fails to utilize true HDR and wide color spectrum correctly, the output lacks the intended visual enhancements. These are often baked into the final product and can't be simply adjusted or corrected without starting over.

6. Overall Inadequate Quality Due to Source Material:

   • Reason: Using old 2k masters instead of new 4k scans limits the quality from the outset. The AI can only work with the data it's given, and if the source material is already compromised, there's no way to enhance it adequately to true 4k standards.

7. Artificial Enhancements (e.g., Video Game-like Hair):

   • Reason: AI-generated enhancements like digital wigs and overly sharp facial features replace the natural look of actors with synthetic textures. Reverting these changes would again require starting from an unaltered source.

8. Inconsistent Grain and Noise Reduction:

   • Reason: AI often removes or misinterprets grain, which is an intrinsic part of the film's texture and detail. Once the grain is smoothed out incorrectly, restoring the original texture authentically is challenging without redoing the entire process with a better technique.

Conclusion

Most of these issues are unfixable because they involve irreversible changes or are fundamentally tied to the limitations of current AI technology. The solution often lies in avoiding AI-based restoration techniques until they are advanced enough to handle such tasks without compromising the film's integrity. The reliance on original, high-quality scans and traditional restoration methods remains critical to preserving the authenticity of classic films.


Question 3 of 4 for ChatGPT-4o: List the most remarkable and controversial statements in the transcript

ChatGPT-4o’s answer:
Here are the most remarkable and controversial statements from the transcript of "AI Can Ruin Movies Now, Too - Aliens and True Lies on 4k":

1. AI's Potential and Harm:

   • "AI has potential to do a lot of great things but also a lot of harm and it’s moving so fast that we’ve been discovering who and what it’s going to break in real time as it’s happening."

2. Critique of James Cameron's 4k Releases:

   • "James Cameron recently gave us a shot across the bow for another application of AI- Hollywood film remasters."

   • "Cameron was already known for producing one of the most disappointing 4ks in the history of the format with Terminator 2 a hyper-processed mess with enough grain reduction to turn characters into cartoons and so much teal that the original colors have been terminated from this timeline."

3. AI's Impact on True Lies:

   • "True Lies is the best demonstration of why this remastering approach was a horrendously bad idea."

   • "Faces regularly become host to a menagerie of parasites, burn scars, and porcelain cracks."

   • "Tom Arnold... looks to be in the early phases of Emperor Palpatine-ism."

4. AI's Failure in Detail Preservation:

   • "Deep learning algorithms... can’t preserve details. They make their best guess about what an object is supposed to be then pull new details out of their digital assholes and smear them across the screen."

5. Uncanny Valley Effects:

   • "Any unusual lighting conditions or obstructions confuse the results- an eyeball briefly melts like a Salvador Dali clock or warps along with the edges of a smoke line."

6. Specific Failures in Aliens:

   • "Aliens is the release I was really excited for and it doesn’t fare much better- all of the same issues are here but with even less stable image quality."

   • "Pores get unsharp-masked into mini bullet holes. Pixels that shouldn’t connect are joined together to form waxy fractal-like patterns similar to what you get when using smoothing filters on old games."

7. Comparison to Historical Restoration Failures:

   • "True Lies looks like the entirety of the film was generated frame-by-frame in Dall-e."

   • "The film was allowed to be film and the grain doesn’t hurt anything. Faces don’t mutate when more than five feet from the camera."

8. Critique of HDR and Color Spectrum Management:

   • "Aliens has a soft cap of 200 nits according to Resolve’s scopes- in other words SDR+. Which means that a thermonuclear explosion… ends up being less bright than the Ghostbusters lobby."

9. Quality of the 4k Restorations:

   • "James Cameron hyped these releases up in interviews claiming that 'every pixel of every frame' had been scrutinized. He acknowledged the importance of getting a 4k release right as this may be the last format we get."

10. Criticism of AI Usage in Film Restoration:

    • "Despite being fully aware of the stakes when preparing his filmography for posterity, Cameron still pressed the 'just fuck my shit up' button binding this early beta technology to his films for all time."

    • "These could be the versions we end up looking at for a very, very long time."

11. Concerns about Future Film Restorations:

    • "If AI is being considered for future restorations isn’t right now the most important time to judge its performance?"

    • "The point is that it has to be used carefully and the problem with these remasters is that it was not."

12. Call to Action for Consumers:

    • "If you’ve had your eye on one of these 4ks- consider avoiding them (because they’re trash)."

    • "With businesses conspiring to force physical media into an early grave we’re on the cusp of ADORABLY losing ownership and control of our history."

These statements highlight strong opinions and criticisms regarding the use of AI in film restoration, the quality of specific 4k releases, and the broader implications for film preservation and consumer choice.


Question 4 of 4 for ChatGPT-4o: Is AI in its current state a good tool for video restoration?

ChatGPT-4o’s answer:
Based on the transcript "AI Can Ruin Movies Now, Too - Aliens and True Lies on 4k," it seems that AI, in its current state, is not yet a reliable tool for video restoration, particularly for high-fidelity projects like 4k movie remasters. Here are some reasons derived from the transcript that support this conclusion:

Reasons Why AI is Not a Good Tool for Video Restoration in Its Current State:

1. Detail Preservation Issues:

   • AI struggles to preserve original details, often generating artificial details that are not accurate to the source material.

   • The deep learning algorithms used in these restorations make guesses about details, which can lead to inaccuracies and distortions.

2. Inconsistent Results:

   • Temporal stability is a significant issue, with AI producing inconsistent sharpness and blurriness from frame to frame.

   • This inconsistency leads to unnatural and distracting visual artifacts.

3. Facial Distortion and Synthetic Enhancements:

   • AI often distorts faces, creating artifacts such as burn scars, porcelain cracks, and overly sharp or smooth features that look unnatural.

   • Synthetic details like hair and facial features can appear cartoonish or video game-like, detracting from the authenticity of the restoration.

4. Lighting and Motion Issues:

   • AI struggles with unusual lighting conditions and motion, leading to melting or warping effects and inconsistent rendering of details during movement.

   • The result is often an uncanny and disjointed appearance that breaks immersion.

5. Color Management Problems:

   • AI can alter original color schemes, sometimes pushing colors into an unnatural range, like excessive teal.

   • HDR and wide color spectrum features are not effectively utilized, often resulting in subpar brightness and contrast.

6. Halo and Sharpening Artifacts:

   • AI introduces over-sharpening and halo effects, making objects look like they are green-screened and undermining the quality of practical effects and stunts.

7. Overall Quality Concerns:

   • The AI restorations discussed often result in worse visuals than previous versions or even the original prints.

   • Professional reviewers may overlook these flaws due to the perceived increase in detail, but the detail is often artificial and nonsensical.

Conclusion:

AI in its current state appears to be more of a detriment than a benefit for high-quality video restoration, particularly for classic films that require careful preservation of original details and aesthetics. The technology may improve in the future, but as of now, the inconsistencies, distortions, and artificial enhancements introduced by AI outweigh the potential benefits.

For now, traditional methods of restoration, which involve meticulous scanning, cleaning, and color grading without over-reliance on AI-generated enhancements, seem to be more effective in preserving the integrity and authenticity of classic films.