The dunk tank no one asked for
You don’t expect a company whose whole thing was "AI that paints Pretty Pictures When You Ask Nicely" to pivot into medical diagnostics—unless, maybe, you’ve been tracking the quiet emergence of full-body ultrasound computational tomography (USCT). But Midjourney did it anyway. In June 2026, they announced Midjourney Medical, a new arm with a prototype scanner that looks like something out of a cheap sci-fi movie: you stand in water, bathed in golden light, then descend into a tank where hundreds of thousands of ultrasonic elements bombard your body to create MRI-like images.
The pitch is bold, the hardware is real, and the science behind it isn’t entirely fictional. But here’s the thing: Midjourney didn’t build this alone—or even mostly on its own. That detail got buried in the splashy announcement, and it matters more than you think.
Let me cut straight to what I’ve verified: the company has a working prototype, it’s borrowing tech licensed from Butterfly Network, and it hasn’t yet received FDA clearance. If you’re scanning your whole body every month for preventative health, you probably want to know who’s actually behind the machine—and whether it’s ready to be trusted with your organs.
The tank, the ring, and the 358,000 elements
Midjourney’s scanner isn’t your average MRI. There’s no big magnet humming quietly in the corner. Instead, there’s a shallow pool—golden-tinted light plays off the surface—and then you step into it. The real magic, though, happens when you’re lowered into a water-filled tank encircled by 40 modular scanner units. Each of those 40 modules packs thousands of ultrasonic transducers—358,000 in total—and they fire, listen, and adjust at up to 1,000 times per second.
That’s a lot of raw power. Each pulse travels through water, hits your body at slightly different angles, and returns with subtle distortions. The machine records those wave-shape changes and streams them to a compute cluster where reconstruction algorithms piece together a 0.5 mm resolution image. Midjourney claims that’s competitive with standard clinical MRI—though not quite up to the cutting-edge high-res devices you’ll find in top-tier hospitals today.
It’s clever engineering. Water is a great couplant for ultrasound, and using thousands of elements gives you dense angular coverage, which helps reconstruction algorithms do their job. But here’s the kicker: this isn’t a standalone invention. The underlying technique—Fullbody Ultrasound Computational Tomography, or USCT—isn’t new. Caltech researchers have been publishing on it for years, and Butterfly Network has been licensing the hardware for exactly this purpose.
The ghost partner
Midjourney’s press release didn’t mention Butterfly Network until Butterfly posted its own statement a few days after Midjourney’s announcement. The tech firm confirmed the co-development agreement, noting that it provided all 40 scanner modules—and expects to earn $74 million over five years for the hardware.
That silence is odd. Midjourney built its entire identity on being scrappy, independent, and transparent (well… until the copyright lawsuits). So why bury the lede? The Register article points out the irony: a company that has faced lawsuits for training its AI on copyrighted images now fails to credit a crucial hardware partner in a diagnostic device meant for public health.
Butterfly declined to say much beyond its official statement, and Midjourney didn’t respond to questions. That’s a red flag if you’re trying to assess the scanner’s safety or accuracy. If Midjourney owns the AI reconstruction side, and Butterfly supplies the probes, who signs off on clinical validation? Who runs the ROC curves? Where’s the peer-reviewed trial data? It’s not in the release. It’s not in the demo video. And for a device that wants to scale to 50,000 units globally by 2031, that’s a lot of unknowns.
The spa, the timeline, and the regulatory gap
Midjourney plans to open its first spa in late 2027. Concept art shows a serene, minimalist setting—think warm lighting, soft seating, and that signature golden pool. Patients bathe briefly, then step into the scanner for a minute-long exam. The company’s ambition is lofty: it believes early and frequent whole-body scans could prevent 30% of deaths and slash healthcare spending by half.
That’s… optimistic. But the bigger problem is timing. The prototype isn’t FDA-cleared yet, and Midjourney admits regulation is “the next limit.” Its strategy seems to be: start low. Declare the device as providing only “body composition maps” (not a diagnostic tool), submit incremental test results, and broaden cleared claims over time. That’s technically possible under FDA’s SaMD (Software as a Medical Device) pathway—but the devil is in the validation.
Whoever runs those validations will need access to ground-truth data. Who owns that data? The patient? Midjourney? Is it stored encrypted? Are the scans used to retrain algorithms—and if so, is that opt-in or default? There’s no disclosure on that front. That’s especially concerning given Midjourney’s history with copyrighted data and training sets. If you’re scanning your entire body every month, you probably want to know who owns the biometric data and whether it’s going into some black-box AI training loop.
Why this pivot tells us something bigger
Look, I’m not here to dunk on Midjourney. Pivoting is how startups survive—and trying to bring an expensive, clunky MRI alternative into a patient-friendly form factor is ambitious in the right way. The tech behind USCT is genuinely promising, and a water-coupled ring scanner could eventually offer faster, cheaper, and more accessible imaging.
But the execution matters. The omission of Butterfly Network isn’t just a PR miss; it’s a sign of how fast-moving teams sometimes blur lines between collaboration and attribution. The lack of clinical validation data isn’t just inconvenient—it’s a dealbreaker for any clinician worth their stethoscope. And the vague promises about preventative health? They sound good in an investor deck, but real medicine moves slowly and carefully.
Midjourney might still pull this off. But they’ll need more than a shiny demo and a well-lit pool. They’ll need transparency, partnerships spelled out in plain English, and at least a hint of humility about what their AI can—and crucially, cannot—do today.
I’ll keep watching. The spa hasn’t opened yet. Let’s see how it holds up when the lights come on.
Quick FAQ, in case you’re wondering
How long does a scan take?
The company targets one minute per full-body exam on the prototype. That’s ambitious—most clinical ultrasound exams range from 15 to 45 minutes—but they’ve built a high-speed cluster and dense transducer array to make it plausible.
Is the scanner FDA-cleared?
No. Midjourney plans to submit data incrementally, starting with non-diagnostic body composition maps before seeking clearance for clinical diagnostics.
Who provided the hardware?
Butterfly Network supplied the 40 ultrasound modules under a co-development agreement worth up to $74 million over five years, per their SEC filing.
Does Midjourney own the scans?
The press release doesn’t say. It’s unclear whether patients can opt out of data usage, or whether scans will train future models.
Is USCT actually real?
Yes. Caltech and other institutions have published peer-reviewed work on full-body ultrasound computational tomography for years. Midjourney is applying AI to it, not inventing the physics from scratch.
What happens next?
Midjourney aims to launch its first spa in late 2027. But without clearer clinical data, regulatory strategy, and transparency about partnerships and data rights, the timeline looks fragile.