Artificial intelligence is increasingly being deployed to restore decades-old films, from Hollywood archives to China’s kung fu classics, enhancing image quality, repairing physical damage and reconstructing degraded audio. What was once a painstaking, manual process is becoming automated and scalable, changing how cinema history is preserved and reintroduced to modern audiences.
As reported by the Pulitzer Center, global film archives are deteriorating faster than traditional restoration methods can address, with large collections of culturally significant content at risk of becoming permanently inaccessible. Conventional restoration is labor-intensive, expensive and slow, leaving institutions unable to clear backlogs before significant degradation makes recovery impossible. AI is now entering that gap, automating image enhancement, scratch removal, resolution upscaling and audio reconstruction more efficiently than manual workflows can match.
The commercial and cultural implications reach well beyond the archives themselves. Restored content can be redistributed across streaming platforms, extending the revenue lifecycle of intellectual property that has been commercially dormant for years. For studios and rights holders sitting on large libraries of older titles, AI-assisted restoration represents a path to unlocking dormant asset value without prohibitive per-title costs that made traditional restoration economically unfeasible at scale.
China Commits to Restoring Kung Fu Classics
As reported by Radii, the China Film Foundation has announced a large-scale initiative to digitally restore 100 classic martial arts films, which was unveiled at the Shanghai International Film Festival. Titles selected for enhancement include Jackie Chan’s “Police Story,” Bruce Lee’s “Fist of Fury” and “The Big Boss,” Jet Li’s “Once Upon a Time in China” and Jackie Chan’s “Drunken Master,” all considered foundational works of Chinese martial arts cinema.
According to Radii, the project focuses on enhancing image quality, sound and production values while preserving original story and aesthetics, explicitly targeting restoration of existing footage rather than digital recreation of new performances.
Boundary Between Restoration and Reconstruction
A more controversial application of artificial intelligence to lost cinema is playing out in the United States. As reported by Futurism, startup Showrunner, now operating as Fable, announced plans to recreate missing footage from the 1942 Orson Welles film “The Magnificent Ambersons,” the director’s follow-up to “Citizen Kane.” The footage was cut by the studio against Welles’ wishes and subsequently destroyed to free up vault space.
We’d love to be your preferred source for news.
Please add us to your preferred sources list so our news, data and interviews show up in your feed. Thanks!
Advertisement: Scroll to Continue
The project, as reported by TechCrunch, involves filming live-action scenes on physically recreated sets, then using AI to overlay digital recreations of the original cast’s faces and voices onto new actors. Fable CEO Edward Saatchi, whose father co-founded the advertising firm Saatchi & Saatchi, described the project as driven by a genuine devotion to Welles rather than commercial intent, calling “Ambersons” the “holy grail of lost cinema.” The company has acknowledged it cannot commercialize the result, as it does not own the rights to the film.
According to TechCrunch, Fable’s own team has acknowledged major technical challenges in the project, including AI-generated errors such as a two-headed rendering of actor Joseph Cotten and what Saatchi described as a persistent “happiness problem,” in which the AI generates facial expressions that are inconsistent with the film’s tone.
The Authenticity Problem in AI Restoration
The Fable project surfaces a larger concern that applies across AI restoration efforts. As covered in Bloomberg Opinion, AI upgrades risk undermining the authenticity of classic films by distorting original cinematography, altering color grading, removing film grain and modifying motion in ways that change artistic intent without the knowledge or consent of original creators. These are not hypothetical risks.
TechCrunch reported that Fable’s AI reconstruction has already encountered failures where the system generated visual details inconsistent with surviving reference material, raising questions about historical integrity when AI fills gaps with plausible rather than accurate information.
The distinction between restoration and hallucination is narrow, and technically difficult to police. When AI upscales resolution or reconstructs damaged audio, it is making probabilistic inferences about what the original content contained. Those inferences may be statistically reasonable but factually incorrect, meaning the output can appear authoritative while diverging from what was filmed. For archival and historical material, that gap carries consequences beyond aesthetics.
For all PYMNTS AI coverage, subscribe to the daily AI Newsletter.
Add as Preferred Source
