Do AI Films Have A Future?

Artificial intelligence remains all the rage when it comes to the current meta discussion of art. But fairly few respectable film festivals have been willing to entertain AI generated films as serious. This changed with the Bucheon International Fantastic Film Festival (BiFan) earlier this month. BiFan is the first longstanding traditional film festival that I know of to have a competition section for AI films. This caveat is necessary because it’s difficult to frame this prompt in a way a search engine will understand.

There is, of course, a fair bit of irony to that given that prompts are what produce AI films. While we call this process artificial intelligence, on the technical level the algorithms behind AI films, or any other kind of AI art, function by reading prompts written by humans, and pulling up words and images collated by other humans in an existing database and trying to build construct something approximating those images. For all the hype about AI reducing the labor inherent in filmmaking, the work that goes into phrasing these prompts and editing the resulting film into something coherent is quite intensive. BiFan only received 114 submissions in the AI film category for just 15 slots. There were 255 films overall in the festival, most of them human-produced shorts, for which there were far more submissions.

There is an AI Film Festival circuit, for what it’s worth. AI films at BiFan also appeared in previous competitions at Runway’s AI Film Festival and the Dubai AI Film Festival. BiFan’s overall program was just more AI-centered this year, with an AI conference and also an AI workshop for filmmakers running from July 5^th to July 7^th. However, the AI Competition Jury at BiFan mostly consists of more typical film experts- Shin Chul, in his sixth year as overall head of BiFan, and Kim Tae-yong, a major South Korean director. The Turkish artist Ferdi Alici and the Estonian film festival advisor Sten-Kristian Saluveer further lend credibility to the whole enterprise.

Now, at this point you might be wondering why I’m not talking about the films themselves, and the reason for that is quite simple. They’re not especially watchable. And as it happens, if you don’t believe me, you don’t need to go to South Korea to watch them. Seven of the ten are on YouTube. These are pretty much the most sophisticated AI films that exist at the moment. And far from going viral, they languish in obscurity in the lower tiers of YouTube’s own traitorous algorithms.

Of all the films I linked, Generation from Riccardo Fusetti has done the best, with 35,218 views to its credit as of this writing, mostly because it received a spotlight on the Short of the Week YouTube channel which has, in the context of these kinds of short films, a fairly decent following. WHERE DO GRANDMAS GO WHEN THEY GET LOST? From Aideate Films only has a paltry 173 views as of this writing. Ironically, the French film won the prize for Best AI Film despite this.

Why did WHERE DO GRANDMAS GO WHEN THEY GET LOST? win? Well, my guess is because it was one of the more coherent ones. Those of you who actually tried to watch the film this a bit of an odd statement to make, given that the plot of WHERE DO GRANDMAS GO WHEN THEY GET LOST? is quite literally just about a child making idle observations about where grandmas go, concluding that they turn into mountains, or glaciers, or houses, or dust, but wherever they go, grandmas become the earth. Believe it or not, that’s actually high end AI storytelling, mainly because prompts with terms like “grandma mountain” can indeed produce images that are visibly both like grandmas and also like mountains.

Getting the AI to produce an actual story to go along with the images is, well, harder than it sounds. Most of the AI films at BiFan cheated in this regard by relying heavily on a narrator to make sense of the often bizarre images on screen, which look like storybook collages and are driving at nonsensical morals. Mic-Mac at Cirque de Freak, for example, is the story of a very judgmental snake who takes offense that the mermaid at the circus doesn’t really do any work. The moral of that story was, apparently, that we can’t find happiness by being judgmental.

Aside from simple collages the main other worthwhile images that AI art can create are scary images, which is a bit of a backhanded compliment since AI art can only do this when it looks obviously wrong in some way. I actually liked Another a decent amount, by this rather limited rubric, although using distinct real actors in AI modified form somewhat defies the spirit of the competition. It also wasn’t much of a crowd pleaser. That honor went to Snowfall, the Audience Award winner, not available on YouTube, most likely because at 14 minutes, it was one of the few features long enough to at least feel like a real movie.

That’s not entirely fair. Snowfall also won the Technical Achievement Award because it really does look quite good. What it was about, aside from a tidal wave based apocalypse and maybe aliens, I’m not totally sure. Final Scene (also not on YouTube) was one of the worst-looking films but also one of the most coherent ones. It concerns a producer who throws away so many screenplays that the evil thoughts in those screenplays become demons. They threaten to kill the producer, but fortunately, he locks them in the mansion and burns it down. While it’s all silly nonsense, that’s about all anyone can accomplish with the technology. How else to interpret something like Latex Kid, about a boy born covered in Latex who becomes a…widely imitated rock star, I think? That one got a Special Mention.

 So did One More Pumpkin, a South Korean AI film that’s sort of on YouTube. It starts at 1:28 here, and is mostly in English that combines the collage and horror aspects that AI films can do with some competency to tell a freaky story about how an elderly couple uses highly addictive pumpkin soup to cheat death. It’s all vaguely reminiscent of stories you’ve heard of before, and your brain can fill in the gaps as necessary since the whole premise is just an excuse to show off the best images created by prompts attempting to generate freaky looking pumpkins.

None of this is especially marketable, for the simple reason that without context and willing interpretation by the viewer, there’s no reason anyone should have to see any of these films. This is the problem that anyone trying to screen any short film on the festival circuit these days faces. It’s not a problem that AI can solve, because film festivals in general these days struggle for relevancy trying to spotlight film that’s actually good as opposed to just being produced for nakedly cynical sales purposes. Which is perhaps the greatest irony of all–even if anyone did manage to feed the right prompts into AI to create true art, they probably wouldn’t be able to actually get anyone to watch it.