In the third episode of \u201cCreative Dialogues,\u201d an interview series produced by the filmmaking division of generative AI startup Runway, multimedia artist Claire Hentschker expresses a fear that AI will commoditize the artistic process to the point where art homogenizes, regressing to a sort of derivative sameness.<\/p>\n
\u201cAre you getting this increasingly narrower average of existing things?\u201d she asks. \u201cAnd then \u2014 as that keeps getting averaged \u2014 is everything is just gonna be a blob?\u201d<\/p>\n
Those are the questions I kept asking myself Wednesday at a showing of the top 10 finalists at Runway\u2019s second annual AI Film Festival, which are available on-demand<\/a> on Runway\u2019s website as of this morning.<\/p>\n Runway held two premieres this year, one in Los Angeles and a second in New York. I attended New York\u2019s, which took place at Metrograph, a theater known for its arthouse and avant-garde bookings.<\/p>\n I\u2019m pleased to report that AI isn\u2019t hastening in a blob future \u2026 not yet at least. But a skilled directorial eye \u2014 the human touch \u2014 makes a clear difference in an \u201cAI film\u2019s\u201d effectiveness.<\/p>\n All of the movies submitted to the festival incorporated AI in some form, including AI-generated backdrops and animations, synthetic voice-overs, and bullet time-style special effects. None of the elements seemed quite to the level of what state-of-the-art tools like OpenAI\u2019s Sora<\/a> can produce, but that was to be expected, given that most of the submissions were finalized early in the year.<\/p>\n Indeed, it tended to be obvious \u2014 sometimes painfully so \u2014 which parts of films were the product of an AI model, not an actor, cameraman or animator. Even otherwise strong scripts were sometimes let down by underwhelming generative AI effects.<\/p>\n Take, for example, \u201cDear Mom\u201d by Johans Saldana Guadalupe and Katie Luo, which recounts the story of a daughter\u2019s loving relationship with her mother \u2014 in the daughter\u2019s own words. It\u2019s a tearjerker. But a scene of a Los Angeles freeway with all the telltale weirdness of AI-generated video (e.g., warped cars, bizarre physics) broke the spell for me.<\/p>\n The limitations of today\u2019s AI tools seemed to box some films in.<\/p>\n As my colleague Devin Coldewey recently wrote<\/a>, control with generative models \u2014 particularly video-generating ones \u2014 is elusive. Simple matters in traditional filmmaking, like choosing a color in a character\u2019s clothing, require workarounds because each shot is created independently of the others. Sometimes not even workarounds do the trick.<\/p>\n The resulting disjointedness was on display at the festival, where several of the films were little more than tangentially related vignettes strung together by narration and a soundtrack. \u201cL\u2019\u00e9veil \u00e0 la cr\u00e9ation\u201d by Carlo De Togni and Elena Sparacino demonstrated just how dull this formula can be, with slideshow-like transitions that would make for a better interactive storybook than film.<\/p>\n L\u00e9o Cannone\u2019s \u201cWhere Do Grandmas Go When They Get Lost?\u201d falls into the vignettes category as well \u2014 but triumphs despite this thanks to a heartfelt script (a child describing what happens to grandmothers after they pass) and an exceptionally strong performance from its child star. The rest of the audience seemed to agree; the film got one of the more spirited rounds of applause of the night.<\/p>\n And for me, that really sums up the festival in a nutshell. The human \u2014 not AI \u2014 contributions often make all the difference. The emotionality in a child actor\u2019s voice? That sticks with you. AI-generated backdrops? Less so.<\/p>\n This was certainly true for festival Grand Prix winner \u201cGet Me Out,\u201d which documents one Japanese man\u2019s struggle to recover from the psychological toll of his immigration to the U.S. as a young child. Filmmaker Daniel Antebi depicts the man\u2019s panic attacks with the help of AI-generated graphics \u2014 graphics that I found to be less successful, ultimately, than the cinematography. The film ends with a shot of the man walking up a bridge just as the streetlights dotting the pedestrian lane flicker on one by one. It\u2019s haunting \u2014 and beautiful \u2014 and surely took ages to capture just so.<\/p>\n It\u2019s very possible that generative AI will one day be able to re-create scenes like this. Perhaps cinematography will eventually be replaced with prompts \u2014 a victim of the ever-growing datasets (albeit with troubling copyright status<\/a>) on which startups such as Runway and OpenAI are training their video-generating models.<\/p>\n But that day isn\u2019t today. <\/p>\n As the screening wrapped up and the award recipients marched to the front of the theater for a photo op, I couldn\u2019t help but notice the cameraman in the corner documenting the whole affair. Maybe, on the contrary, AI will never replace some things, like the humanity we humans deeply crave.<\/p>\n<\/div>\n![\"AI](\"https:\/\/techcrunch.com\/wp-content\/uploads\/2024\/05\/Screenshot-2024-05-10-at-4.30.31\u202fPM-e1715373196264.png?w=680\")
![\"AI](\"https:\/\/techcrunch.com\/wp-content\/uploads\/2024\/05\/Screenshot-2024-05-10-at-4.27.26\u202fPM.png?w=680\")
![\"AI](\"https:\/\/techcrunch.com\/wp-content\/uploads\/2024\/05\/Screenshot-2024-05-10-at-4.28.00\u202fPM.png?w=680\")