Editor’s note: Following the publishing of our review, we received word from Brenden Dawes, who developed the generative system used by the filmmakers of Eno, that while the film teases the possibilities of AI and generative technology in an art practice, the film itself consists entirely of filmed new and archival materials with no AI-generated content.
A film of infinite possibilities thanks in part to a generative AI hook, Gary Hustwit’s Eno is partially a straightforward biopic featuring interviews and archival footage with composer Brian Eno, the experiential musician and artist whose credits include playing the synthesizer in Roxy Music to creating the start-up sound for Windows PCs. The film is assembled at random, with a set beginning and ending, inspired seemingly by a deck of “Oblique Strategies” cards that Eno and David Bowie used to create tension and contractions within their collaborations.
Of course, Eno is not the first film that can be randomly assembled to create a new experience each time. Apart from expanded cinema, gallery presentations, or experimental works wherein projection is performance, Erik Løchen’s 1972 Remonstrance comes to mind, assembled in 120 different combinations where each reel can be played in a different order each time. Mike Figgis’ digital video experiments (e.g. Timecode) have a narrative structure subject to change depending on which of the film’s four “real-time” frames is dialed up to call your attention to that story thread. Sam Green’s 32 Sounds was presented with a live sound mix that was also a work of presentation-as-performance, using headphones and a sound system to create an experience that was both binaural and three-dimensional but subject to change depending on who manned the mixing board.
Until this Sundance premiere, the use of AI in documentary filmmaking was most prominently discussed vis-à-vis the undisclosed replication of Anthony Bourdain’s voice in Morgan Neville’s Roadrunner. Still, in many ways, a biopic can be a seemingly random creation: how does one really slice up an entire life? Often biopics begin at the end or start at a pivotal moment in one’s career. Using AI to create six different randomly generated DCP versions of Eno for its premiere screenings at the Sundance Film Festival (each dated) is quite fascinating and, for a reviewer, a challenge.
Can its eventual home on a digital medium such as streaming allow a viewer to enter prompts, adjust the running time, skip sections they prefer not to see? The film that I had seen felt at times a little too repetitive, returning to reoccurring themes in Eno’s fifty-year body of work. Gardening, for instance, came up several times in 85 minutes. Meanwhile some sections, including his Nam June Paik-esque video installations using color TV and a camcorder purchased from a Foreigner roadie, are a minor passage.
So what we’re left with is a random series of threads, some organized quite well (by Hustiwit and editors Marley McDonald and Maya Tippett) as episodic sketches of Eno’s time working with U2, Bowie, Roxy Music, and for freelance clients. The film also unpacks his current work and previous interests wondering what makes a certain note or rhythm more emotionally appealing than another while working on new compositions using generative technology to randomly change pitch, tempo, and rhythms.
Does the film that I saw work? Not quite, which is arguably the point. A blind spot for a reviewer is knowing what material was included or excluded––likely the struggle in making any biopic, especially about an artist with a half-century’s body of work and 74 years of lived experience. Perhaps outsourcing a bespoke edit using generative AI is one way around this argument and perhaps, like ChatGPT, Eno’s final home will allow a user to customize the Eno experience by setting parameters and keywords. Perhaps one viewer might opt for more Bowie and less U2, or a film about what he’s working on now rather than one exploring his origins.
Cinematographers have an old saying: the final auteur is the film’s projectionist. They can choose to present a film with care at the proper sound and light levels as the filmmakers intended or, like many modern multiplexes, lazily leave on the RealD 3D filters, creating a dark and somewhat blurry image. The film is undoubtedly interesting in passages as Eno (in the version I saw) revisits his diaries and mini-cassette recordings, sharing his philosophy on creativity. (It involves skipping breakfast and ultimately trying to consume less to free one’s mind.)
Is there a potential version of this film that’s a masterpiece? Sure. Is there a version that’s unwatchable? Probably not. Hustwit––whose past work includes documentaries of design on a variety of subjects (the Helvetica font, urban planning, product design, the influence of Dieter Rams)––may have created a film that will present technical challenges for commercial public screenings, which is not the end goal of an ephemeral picture like this.
Editing is where meaning’s largely created, and Eno, like Steve McQueen’s recent documentary Occupied City, suggests a convergence of installation and cinema. Both documentaries are works of loose cartography that could perhaps be assembled entirely differently but would largely retain their power.
The version of Eno I saw on January 23, 3:45PM MT at the Ray Theater in Park City, UT, is far from perfect, which of course brings to mind a fundamental limit of generative AI. It creates a work product that looks viable on the surface, but probe deeper and flaws start to emerge. Justine Bateman has given the analogy of AI as a blender. Here, perhaps Hustwit has provided all the inputs but too few guardrails.
As an experiment and artifact, Eno is fascinating. Could a version be produced that has been trained on films that were structurally more coherent and also about similar artists––Amanda Kim’s Nam June Paik: Moon is the Oldest TV or Tyler Hubby’s Tony Conrad: Completely in the Present? Could it be trained on previous Sundance Film Festival documentary winners? Can a cut be optimized for New York City audiences at DOCNYC, or a more suburban audience at a regional festival?
The promise for something interesting is here, although I think when it comes to documentary film editing, there are still several years before generative AI is considered a viable approach in professional filmmaking. Eno, if anything, certainly tests its limits.
This review is based solely on the version screened on January 23 2024 at 3:45PM MT at the Ray Theater in Park City, UT.