This manic TV spot explains AI's place in video art

P.J. Accetturo's AI-generated ad helps make a notoriously abstract debate concrete

It’s as tough to avoid speculation about generative AI’s impact on creating art right now as it is to avoid it in practically every other realm of modern life. Almost no matter the industry in question, though, the debate tends to get hobbled by one of the same problems: an over-reliance on generalities.

Sure, people in and outside the art world understand that artists in 2025 can make quality images faster and cheaper than ever using genAI tools. It’s just tough to think through the effects of the shift without some kind of grounding in how much faster and cheaper the process becomes, or what kinds of images the technology does and doesn’t excel at producing.

A recent milestone in AI filmmaking has given us real, specific touchpoints for all those vaporous uncertainties. But I’m guessing most of the art world missed it, because it wasn’t made to elevate generative AI to creative respectability in a gallery, museum, or even a film festival. It was a 30-second national TV spot made to advertise an online gambling platform “prediction market” called Kalshi.¹ And it’s a wild ride, as you can tell from the preview still alone…

The spot aired for the first time during a commercial break on YouTube TV’s stream of the NBA Finals this June. (All the characters’ mentions of “Indiana” and “OKC” refer to the Indiana Pacers and the Oklahoma City Thunder, the teams that played for the championship.) Its other deliberately weird-ass scenes give examples of the mind-melting array of possibilities beyond sports that Kalshi lets you bet on “legally trade on,” like how much egg prices will increase in a given month, how many hurricanes will hit the US in 2025, and how soon the federal government might announce that extraterrestrials exist.²

Part of the reason I’m unpacking this spot is that its brass-tacks realities push through a lot of the usual walls that block off productive conversations about AI-generated art and filmmaking. For example, it’s irrelevant whether you think the imagery above is or isn’t naturalistic enough to convince a mass audience it was shot using IRL actors and locations; the company that commissioned it was satisfied enough to go ahead and air the thing already. Which moves the work past being a theoretical and intellectual flashpoint about genAI’s potential commercial applications and into being a practical test case of them.

But the other part of the reason is that the ad’s mastermind, a live-action commercial director named P.J. Accetturo, has been unusually candid about the process, the economics, and aftereffects of the project on multiple platforms. I have a decent sense that he’s telling the truth too based on my own past experience and relationships. (TGM lore nugget: I supported myself—and this project’s first phase—by working as a ghostwriter for directors of TV spots and online branded content from 2012-16, the stretch between my two stints working in LA’s gallery sector.)

Given all that, sounds like it’s time for a little self-Q&A about the specifics Accetturo has revealed and what they might mean for artists experimenting with generative AI models on the film and video front. LFG…

How much cheaper was the Kalshi spot to make than a similar ad produced traditionally?

Close to $1m at the low end, and possibly multiples of that.

Accetturo wrote in his own newsletter that he capped the production cost on the video at $2,000—a figure that Kalshi exec Jack Such confirmed separately to NPR—and charges a “five-figure fee” on top of the budget. Even if we stretch his fee to the absolute limit of that coy definition, the all-in cost to Kalshi would be no more than $102,000.

For comparison, a similar 30-second live-action spot would cost between $1m and $5m, according to my conversations with two veterans of the commercial and online branded content space.

Wtf how can the gap be that wide?!

Making the Kalshi spot IRL would mean paying for casting (including extras); location scouting, rental, and permitting (or use of a soundstage); production design; equipment (cameras, lighting, mics, etc); hair and makeup; wardrobe; stunts; pyrotechnics or visual FX; craft services (aka catering during the shoot); editing; sound design and mixing; an original soundtrack or sync music (aka song licensing); and more.

Also keep in mind that, depending on where the shoot were to happen, many of the people in charge of overseeing and executing these different facets of production would need to be union workers, which means paying a premium for their skills. It all adds up, especially over multiple days. Speaking of which…

How much faster did the genAI production move than a traditional one?

The difference from concept to completion adds up to multiple months. One commercial producer I talked to estimates that the all-in timeline for a hypothetical live-action version of Accetturo’s Kalshi spot could stretch to around three months: a few weeks of concepting, another few weeks of prep, a one-month shoot, and another month of post-production.

Accetturo, meanwhile, says he finished the entire process, from writing the script to finalizing everything in post, in two days—the entirety of which he spent “in [his] underwear.”

Hold on, that thing would have taken a full month to shoot?!

Yep, a physical shoot like this would be a madman, and one of the biggest reasons is geographical. By my count, the Kalshi spot would have called for around 11 different locations to fill out its 30 seconds.³ A green screen either wouldn’t get the job done or would be prohibitively expensive to use instead. (Good luck with that final explosion-laced jet ski scene, brother!)

That means transporting the entire traveling carnival of a professional commercial shoot all over creation (as my dad would say), including to demanding, expensive locales like the Vegas Strip (or a facsimile of it) and a full-sized ice hockey arena.

In fact, shooting the ad for real may verge on the impossible. The colossal logistics are solvable in theory. But there’s a near-zero chance that the client, the production company, and the insurers would all sign off on this script given the cost, the moving parts, the risks, and the content itself.

Does the spot tell us anything useful about the current limits of generative AI imagery then?

Definitely, and it starts with the specific tools Accetturo used—mainly Veo 3, Google’s AI video model—to conjure up the Kalshi ‘footage.’ (He walks through his entire process here, including an excerpt of his script and one final video prompt.)

From my conversations, Veo 3 has been a game-changer for filmmakers because the model has been trained on millions of hours of video from YouTube (which, reminder, Google also owns). That’s made it far and away the best option for faking moving images of people, especially from the few camera angles that dominate influencer content and video interviews. It’s not a coincidence that none of the characters in the Kalshi spot are shown from behind or over the shoulder, for instance; they’re almost always shot frontally, usually directly facing the camera.

It’s also critical that every scene here unfolds in a single shot—and an extremely short one at that. The Kalshi ad is proof that Veo 3 can conjure almost anything, but like every other genAI video model, it still pretty much sucks at continuity.

This means that if a director wants to cut between different shots of the same scene, the algorithm struggles to keep the individual elements of that scene consistent from one angle to the next. Characters or their wardrobe might look noticeably different. The lighting or color grading might shift. Details of the environment itself might transform. All of which makes it tough to construct anything that lasts more than a few minutes before it devolves into a cinematic mess.

In this sense, Accetturo’s Kalshi spot is an ideal use of the tech: a bunch of lightning-quick shots of what he called “the most unhinged” scenes he could dream up, secure in the knowledge that generating them would cost only slightly more than a vastly chiller, simpler shortform narrative made with the same tools. For example, he wrote in a LinkedIn post that this minute-long satirical spot for Puppramin, a fictitious pill that attracts puppies to people experiencing clinical depression, cost him around $500.

Could you anchor all these specifics to the art world with a famous piece of video art?

Let’s try Matthew Barney’s Cremaster Cycle, the hallucinatory quintet of films he released with his then-dealer, the late Barbara Gladstone, as his co-producer between 1994 and 2002.

Although there are a few broad themes threading together the different installments of the series, each individual film operates more as a collection of esoteric yet memorable images than a cohesive narrative. The cycle is, to some extent anyway, an extremely art house take on the basic motivation Accetturo used for the Kalshi spot: grab the viewer’s attention with surreal visions that don’t let up. Barney’s just happen to center on stuff like arcane Masonic rituals in the Guggenheim, a Busby Berkeley-style chorus line mimicking the division of cells on the electric blue football field of Boise State University, and an Isle of Man-set motorcycle race representing the situation-dependent descent of the testicles.

In terms of frame-by-frame imagery, generative AI’s strengths seems to give any aspiring Barney-like artist a massive advantage over Barney himself 20+ years ago. Then there are the cost savings.

The most widely cited budget estimate for the Cremaster Cycle is $4m all in. Adjusting for inflation, that would equate to around $7.25m in 2025.⁴ If we use the most conservative conversion rate from my chats about Accetturo’s Kalshi spot—that is, $2,000 spent on Veo 3 generations gets a filmmaker around $1m worth of traditional production value—then a successor to Barney could get a Cremaster Cycle’s worth of onscreen bang for less than $15,000. That makes an extravagant mythopoetic video opus achievable for even a lone artist behind a laptop. On paper, anyway.

Why just “on paper”? What’s with the hedging?

We need to address the runtime. Each Cremaster film lasts between 42 minutes and 182 minutes. Together, they stretch to a daunting six hours and 38 minutes. The grand scale is integral to the work’s impact.

So instead of running a genAI vs. traditional conversion for production value, let’s run one for total screentime. Accetturo’s Kalshi spot lasts 30 seconds, at a cost of $2,000. That maths out to around $67 per second of video.

With its nearly seven-hour length, The Cremaster Cycle as a whole lasts just about 24,000 seconds. At a cost of $5,000 per second of screentime, then, a genAI comp would currently cost around $1.6m. That’s 2.5X less than Barney’s alleged budget for the series—but it’s also way, way more than the $15,000 it sounded like we could produce this for based on the production value efficiencies alone.

Then there are the limitations of what you would actually be able to put onscreen given the continuity constraints of genAI video models. Even the more episodic films in The Cremaster Cycle each involve sequences that last several minutes apiece, with a consistent location and subjects captured in multiple shots from different angles. That’s a nightmare assignment for AI video tools, at least for now.

Are efficiencies and tech capabilities the only factors to consider here?

Nope! Let’s not lose sight of the chasm between how a video piece is judged in an advertising context versus a contemporary-art context.

TV spots are work-for-hire jobs. There’s no shame in that! I won’t tell you that name recognition never matters when choosing a director, but in a lot of cases it matters less than the sheer ability to execute on the brief for an attractive price. Most companies are ecstatic about the prospect of a filmmaker who can deliver 70% or 80% of the results an A-lister could deliver while costing them a fraction as much.

The art world is more discriminating. A genuinely groundbreaking work still has the capacity to vault a little-known artist to a higher level. It’s just that the professional network and the mythology behind an artist typically outweigh the merits of the work itself. In other words, a mediocre-to-bad film from a well-connected artist beats a good film by a poorly connected artist 99 times out of 100.

Accetturo’s Kalshi spot shows that generative AI has already shifted the production calculus of video works at a meaningful scale, and the technology is poised to keep improving. But no matter how much better, cheaper, and faster the models get, they won’t change the social calculus that has determined success in the art business for decades, if not centuries. Whether that consistency counts as a pro or a con probably depends on which side of the industry’s gate you’re already standing on.

1. Kalshi literally has billions of dollars riding on the argument that there is a meaningful legal difference between an online sportsbook—meaning, a platform that itself makes odds and takes bets on all kinds of outcomes in sporting events—and a “prediction market” where users can make odds and bet with each other on all kinds of outcomes in sporting events. To me, this is like claiming you didn’t punch someone in the face, their head was just directly in the way when you thrust out your fist in anger. But it’s complicated—and how the company got to this point is bizarre and twisty and fascinating. For more info, this Matt Levine walkthrough gets my highest recommendation.

2. I’m extrapolating from the scene with the beer-chugging alien and this actual prediction on Kalshi: “If the President, any member of the Cabinet, any member of the Joint Chiefs of Staff, or any US federal agency definitively states that extraterresterial [sic] life or technology exists before Jan 1, 2026, then the market resolves to Yes. Outcome verified from White House.” Users thought it had a 6% chance of coming through when I sent this newsletter.

3. They could probably have reused the same rural roadside vista for the shot with the old woman next to the truck sporting the “FRESH MANATEE” sign and the one with the shirtless old dude walking away from a flaming car wreck. I suspect the two interiors with the alien also could have been shot almost anywhere via the practical magic of some carpentry and production design.

4. I’m a little wary of the $4m estimate because, A) none of the stories that cite it actually seem to reference where it came from, and B) the films just look like they should have been more expensive to make—a possibility kept afloat by the lingering rumor that Cremaster 3, by far the longest entry in the series, cost around $8m alone, as a 2006 article for Cinema Journal notes.