Can Artificial Intelligence help better restore and preserve heritage cinema?
That was the question asked to a high-level panel on Thursday at the Classic Film Market, which runs alongside Lyon’s Lumière Film Festival, dedicated to heritage film.
Struggling to override the clatter of the heavy downpour hitting the ceiling of the tent set up next to the Lumière Institute for the duration of the festival, participants took part in a heated debate entitled Artificial Intelligence: A Tool for Heritage, in front of a packed room of industry professionals.
Opening the discussion, Barbara Mutz, in charge of legal and regulatory matters at France’s National Audiovisual Institute (INA), said AI algorithms developed in-house hugely facilitate the archiving and location of its huge catalogue.
“We can index images and sound in a way that allows us to locate them [more easily] at a later stage, both for our own use and that of our users. We have also developed a voice recognition tool which allows us to identify male and female voices, as well as music, and segment broadcast time. This is useful for the Regulatory Authority for Audiovisual and Digital Communication to study gender parity,” she explained, adding that INA also uses AI tools to create specific formats for social media.
When it comes to film restoration, the vertiginous evolution of technology is providing new ways of restoring and revitalising old films. Using extensive databases, AI algorithms can identify the common imperfections found in damaged films and correct issues like scratches, dirt or flickering.
“AI lends itself perfectly to [restoration],” said Rodolphe Chabrier, the co-founder of French VFX powerhouse Mac Guff (“Despicable Me,” “Titane,” “Lupin”). “You can restore a damaged image – in fact, you can do more than restore it, you can interpret it, improve it, and render a much better quality than what the Lumière Brothers, for instance, had hoped for when they made [one of their first films] “Arrival of a Train at La Ciotat.”
An assertion which is at the very heart of the debate, as it begs the question of the limits of correction. As Mutz puts it, the goal of an institution like INA is “not to reconstitute films: we must be transparent about whether it’s a restored version or not.
“Above all, our job is to promote and, most importantly, to preserve our heritage, so we must always ensure traceability back to the original. Our aim is to bring back to life archives that no one would be interested in because they are of such poor quality, to make our archives accessible again via these restoration tests.”
Bringing back to life historic – and deceased – celebrities is exactly what Chabrier does with his César-winning generative AI tool, Face Engine, developed by Mac Guff’s in-house R&D department, which recently resuscitated legendary French actor Jean Gabin, among others, to be interviewed by French journalist Thierry Ardisson for his show “Hôtel du Temps.”
But how about the original films making up the datasets used to develop the algorithms – should rights owners receive royalties?When it comes to music, for instance, French song rights collecting org SACEM confirmed last week that anyone training AI models with its members’ music must first get permission, after it opted out of a data mining exception under European copyright law.
Chabrier’s view is that it’s up to film producers to sort out clearance rights. Besides, he argued, generative AI tools are based on thousands, if not millions, of sources, which makes the task extremely complex. Moreover, “AI doesn’t just copy and plagiarize. What it does is take inspiration [from the works], just like an artist who goes to a museum seeking inspiration.”
According to the panel’s third member, film journalist Julien Dupuy, what AI does is allow authors to offer a new perspective on film and images of the past.
“When Peter Jackson made “They Will Never Grow Old”, using original WWI images which he slowed down to 24 frames a second, he took responsibility for his vision as an author and the immersive experience his film offers,” said Dupuy.
This can be taken a step further, he added, saying AI can be considered as an artistic partner, citing the example AI generated short film “Thank You for Not Answering” released earlier this year by L.A.-based director and artist Paul Trillo, who said he was influenced by the work of directors such as Wong Kar-wai, David Lynch and Wim Wenders, the recipient of this year’s lifetime achievementLumièrePrize.
What the artists themselves think of this is anyone’s guess – though 2022 Lumièrelaureate Tim Burton made no secret of his opinion on the matter earlier this year, after the publication of AI generated photos of Disney characters recreated in Burtonesque style, saying: “It sucks something from you. It takes something from your soul or psyche; that is very disturbing, especially if it has to do with you. It’s like a robot taking your humanity, your soul.”
The Classic Film Market runs alongside the Lumiere Film Festival in Lyon until Oct. 20.
