STUPID/SMART
The Banger Blog · a Stupid/Smart breakdown
New · 2025

YOLO

Kalshi · PJ Accetturo, made with Google Veo 3

▶  Watch the spotMore

The insight

Kalshi is a prediction market. You buy contracts on whether a thing will happen. The category problem is that this is either extremely boring to explain or slightly alarming, depending on how you explain it, and the brand had no money for the kind of film that usually runs during the NBA Finals.

The read that makes this work is not "AI is cheap". It is that AI video, in 2025, had a very particular look: fluid, dreamlike, slightly wrong, objects not quite obeying physics, faces that hold for two seconds and then betray you. Every other advertiser was fighting that look, trying to sand it down into something that passed for real film. Kalshi asked a different question. What is this aesthetic actually good at? Answer: chaos. Unhinged, escalating, cartoon-logic chaos. Which happens to be exactly the feeling of putting money on an outcome.

The idea

Stop hiding that it is AI. Make a spot that could only be made this way.

"YOLO" is thirty seconds of escalating GTA-style lunacy, one absurd vignette after another, with the tone of a video game cutscene that has come loose from its game. Nothing in it is plausible and nothing tries to be. Director PJ Accetturo made the whole thing with Google's Veo 3, no crew, no cast, no location, no camera.

How it works

The process is the interesting part, and it is much less magical than the headline suggests. He used Gemini to turn a rough script into a shot list. Then he prompted Veo 3 in batches of five, because past five the quality drops off. Crucially, every prompt re-describes the setting, the character and the tone from scratch, because the model has no memory of the shot before it. Then the clips get cut together in CapCut and Premiere like any other edit.

Here is the number that actually matters: it took 300 to 400 generations to get 15 usable clips. That is a keep rate of around four percent. The whole thing came together in two days for about $2,000 in generation costs.

The stupid: hand your brand's national television debut to a text-to-video model. No director of photography, no talent, no shoot. Lean into the melting-face uncanny look rather than away from it, and put the result on the NBA Finals, where it will sit next to work that cost several hundred times as much.

The smart: the aesthetic and the product are genuinely matched, so the artifacts read as style instead of as poverty. The economics change what craft even means: when a take costs cents, you can throw away 96% of what you make, and the skill relocates from shooting to selecting. And the making-of story was the real media plan. "AI ad, two days, $2,000" is a headline, and it earned coverage from CNBC to NPR that a normal thirty second spot never gets.

Why we like it

The steal is not "use AI". It is: find the one job your cheap tool is genuinely better at than the expensive one, and build the idea there. Every tool has a native aesthetic. Fighting it costs money and usually loses. Kalshi won by picking a concept that only looks right when it is a bit broken.

Two honest caveats. First, the $2,000 is generation cost only. The airtime during the NBA Finals cost vastly more, and quoting the two thousand as though it were the price of the campaign is the sleight of hand in every write-up of this, including some of the admiring ones. Second, this trick has a shelf life. It worked because AI video was novel and the look was a signal. As the tools get better and everyone uses them, "we made it with AI" stops being a story and you are back to needing an idea. Kalshi got there first. First is a real advantage and it is not a repeatable one.

Results

~$2,000generation cost
2 daysconcept to finished cut
300-400generations for 15 clips

Straight talk: these are the director's own published figures, widely repeated in trade and mainstream press but not independently audited. The viral impression counts floating around (20M and up) I could not verify to a primary source, so they are not here. No sign-up or business results have been published by Kalshi.

Who made it

Client: Kalshi
Director / maker: PJ Accetturo (PJ Ace)
Generation: Google Veo 3 · shot list via Gemini
Edit: CapCut + Adobe Premiere
Ran: NBA Finals, June 2025 (Pacers vs Thunder)

Sources

The spot, "Kalshi - YOLO"
PJ Ace, how it was made in two days
Ad Age, Kalshi airs AI-generated ad during the NBA Finals
NPR, is this the future of advertising?
CNBC Television, on the Veo 3 spot

The way in

The devices behind this one, from our library of ways in:

Don't hide how it was madePush it way past sensible
← The Banger Blog