xAI's Grok Imagine Doesn't Have To Be Good, Because the Psyop Is the Product

One word was published, and 6.8 million people looked. That is the entire product.

None of this would be worth writing about if the model involved were bad. It is not bad. That is what makes the rest of it interesting.

Here is the small thing that was actually published. On September 19, Elon Musk quote-posted a two-minute-fifty-three-second vertical clip from an account with 97,354 followers, a poster who had titled it "Victory By Any Means." Musk's addition to that clip was the word "Banger." No spec, no claim, no link, no context. Just the word.

[The clip] had already done 8.1 million views on its own. Musk's one-word endorsement added another 6.8 million, along with 22,241 likes, 2,248 reposts and 4,276 bookmarks. An account with 241.6 million followers is a distribution network, and a distribution network does not need to say anything to move product.

The Case for Calling It a Banger

The steelman is strong, and it deserves a fair hearing, because the reflex to sneer at anything Musk touches has become its own kind of slop.

xAI's Imagine Video is not a joke. It has held the top spot on image-to-video preference arenas, with xAI claiming a 52-point Elo jump when 1.5 landed — and even reviews that dislike the model concede the ranking is real. Its party trick is the one that matters most in practice: hand it a strong still image and it animates it coherently, with synchronised audio generated in the same pass, faster than the models that beat it on fidelity.

The price is the argument, and it depends on which model you name. xAI's own price list charges $0.050 per second for the unversioned grok-imagine-video and $0.080 per second for grok-imagine-video-1.5, the model this piece is about. xAI's own launch page puts a 6-second 720p clip at about 25 seconds on the Fast variant, and Curious Refuge clocked most of its renders at around 60. That combination — cheap, fast, audio included — turns it into the drafting layer of a real workflow, and the reviewers who rank it second on fidelity still put it in the loop: iterate on Grok Imagine, finish on Veo 3.1 when the brief demands the pixels.

So xAI shipped a real thing. The arena says it wins where it competes, the pricing says it wins on volume, and the speed says you can afford to be wrong five times before lunch.

There is also a resolution argument, and the honest version of it cuts against the sceptics a little. xAI's capability docs state that on grok-imagine-video-1.5, text-to-video supports native 1080p, with resolution a setting you choose. The complaints that matter are not about pixels anyway. They are about motion and physics, and those do not improve because you rendered bigger.

Why the Psyop Is the Product

Because the demo is not the thing being sold. The announcement is.

Look at what can actually be verified about the post that moved 6.8 million views. The poster frame shows a pink-haired figure in a red armoured torso rig, mouth open, with a subtitle burned into the video reading "Oh for the love of God!" That is the evidence available: a caption and a costume. Everything else about that clip's quality is inference, and the word "Banger" is the only review it received.

Now look at the model, and count what the sources agree on. All three arrive at the same one place: nobody is arguing about the top of the spec sheet any more. Curious Refuge reports the update shipping "native 1080p output," and xAI's own capability docs agree that the 1.5 model supports it for text-to-video. The fault lines are all in motion, physics and reliability, and that is where the reviewers who tested this version land: Curious Refuge found rival models consistently stronger on believable movement. Even the launch is three dates wearing one: a preview in late May, general availability on June 16, 2026, and Musk announcing "wide release" the following day. Reports from people who tested it describe the underlying model trailing on physics and motion; Curious Refuge found rival models consistently stronger on believable movement, and called out visual effects as the weakest area, with strange motion on explosions and impacts. The AIReiter review's summary line is the most useful sentence written about the product: "impressive ceiling, unreliable floor," with quality regressions it traces to server-side updates and paid users running out of generations after two or three clips. One eight-test comparison, run by Vexa Video's AI Video Lab on the earlier Grok Imagine 1.0, scored that version zero out of five on anatomy and motion accuracy against Veo 3.1's three, and 30/40 overall against 36/40. No independently scored benchmark of that kind exists yet for 1.5, so read those numbers as the direction of travel rather than a verdict on the model this piece is about.

None of this is fraud. It is a bush-league sleight of hand performed at planetary scale, and the robots have run the same play for years. In October 2024, at the "We, Robot" event, Optimus units poured drinks and handed out snacks; Tesla's Optimus engineer Milan Kovac subsequently said the robots were "human assisted to some extent" (as carried by Benzinga). By January 28, 2026, on Tesla's fourth-quarter earnings call, Musk told investors the robot was not being used in any "material way" and was in the factory "to learn" (Business Insider). heise noted he had promised 10,000 units at the start of 2025 and built far fewer, and that "We had Optimus perform a few simple tasks in the factory" was the honest summary of a robot that had been sold as a labour force.

Then, six months later, the promise came back, wearing the same suit. On the July 22, 2026 call, Musk said there is no humanoid robot that can do generalised tasks, and that Optimus would be the first — "where it's not just a demo, it's genuinely useful in day-to-day life."

The tell is in that sentence. Nobody selling a working product needs to specify that it is not a demo. The disclaimer is the strategy, restated as a boast.

And in the middle of that cycle sits a single word under someone else's clip, collecting 6.8 million looks and costing absolutely nothing to produce. It does not need to be true. It does not need to be false. It only needs to be louder than the earnings call.

xAI will keep shipping demos, and Musk will keep calling them bangers, and the two facts will keep protecting each other. A banger costs nothing to produce, which makes it the one line item in the company's history that has never once missed a deadline.