Midjourney's full-body ultrasonic CT scanner

The headline writes itself: Midjourney, famous for conjuring images from text prompts, just unveiled a full-body medical scanner. The twist is that there is no AI inside it yet. At the reveal, founder David Holz was blunt: 'We're not even using any AI in this yet, just really cool hardware and software' ^[2]. What the scanner actually does is physics. A person descends on a platform through a ring of ultrasonic transducers in a shallow pool of water, the array fires sound through the body from every angle, and software reconstructs the returning echoes into a 3D anatomical map ^[1].

So why does an image lab care? Because the hard part of ultrasonic computed tomography is not the sound, it is the reconstruction. Turning hundreds of thousands of scattered acoustic measurements into a clean picture of muscle, bone, and organs is a classic inverse problem, mathematically the same shape as turning visual noise into a coherent face. That is precisely the competency Midjourney spent years building, and the company's stated plan is to layer AI-driven reconstruction on later, once enough scans exist to train on ^[1]. In other words, the 'AI scanner' framing is a bet on the future, not a description of the present: today it is a hardware project wearing an AI brand, and the intelligence is the part that has not shipped.

Strip away the spa imagery and the scanner rests on someone else's silicon. The imaging modules come from Butterfly Network, the ultrasound-on-chip company, under a co-development agreement worth up to $74 million over five years that was filed in November 2025, months before the public reveal ^[3]. The current prototype packs 40 Butterfly modules and roughly half a million transducer elements ^[4].

Butterfly was careful in how it endorsed the project. CEO Joseph DeVivo called it 'a potentially meaningful commercial opportunity for Butterfly' and praised the mission to 'democratize access to personal imaging data,' while framing the partnership around prevention rather than vouching for any diagnostic claim ^[3]. The dependency matters because it relocates the novelty. The genuinely new piece is the reconstruction stack Midjourney is promising to build, not the off-the-shelf chips doing the sensing. The most-shared version of this story, 'Midjourney built medical hardware,' is underneath a quieter one: Midjourney integrated Butterfly's hardware and is betting its software can finish the job.

Putting these machines in a 25,000-square-foot wellness spa near Union Square, alongside saunas and cold plunges and opening at the end of 2027, is not just branding ^[5]. It is a regulatory route. A device that produces diagnostic images needs FDA clearance, a multi-year gauntlet of clinical trials and comparative studies. A service that sells 'body composition maps' as a wellness amenity does not ^[5]. By launching as a spa experience that explicitly avoids diagnostic claims, Midjourney can start scanning paying customers while it climbs the FDA ladder one approved use at a time.

That sequencing is the entire go-to-market. The stated ambition, on the order of 50,000 scanners and a billion scans a month by the early 2030s ^[4], only becomes a medical product if the regulatory path clears. Until then the spa is both the storefront and the legal shelter, and the scans it sells are, by design, not something a doctor can act on. The flip side is the prize Midjourney clearly wants: a billion monthly full-body scans would be the largest longitudinal dataset of human anatomy ever assembled, and that, not the scan fee, is the asset an image company would covet.

The marketing number is 60 seconds, a full-body scan in the time it takes to read this paragraph, pitched at 'nearly a hundred times the speed' of an MRI ^[1]. The reality today is more modest. The working prototype currently takes roughly 20 minutes per scan and has been used on about a dozen people, with the one-minute figure framed as a target for future hardware generations ^[6].

The gap cuts both ways. Even at 20 minutes the device would still be several times faster than a 60-to-90-minute full-body MRI, with no radiation and no magnets ^[5], a real advance if the image quality holds up. But 'a target for future hardware' is doing a lot of work in every breathless retelling, and the leap from a 20-minute prototype tested on twelve people to a planet-scale imaging network is exactly the part that has not been demonstrated.

The loudest counter-voices are not Luddites; many are radiologists. The recurring objection is that whole-body screening of healthy people surfaces incidentalomas, benign findings that trigger anxiety, follow-up procedures, and cost without improving outcomes ^[7]. Compounding it, there is no peer-reviewed validation of the scanner's accuracy against MRI or CT, and no independent radiologist has publicly evaluated its image quality ^[7].

Online, the technical critique sharpened into a specific fear: that Midjourney's image-generation models could be used to upscale sparse scan data, and that generative reconstruction tends to fill in the most probable tissue, which for a healthy-looking body means inventing healthy tissue and potentially papering over real disease. Comparisons to Theranos and to the Therac-25 radiation-overdose disaster circulated widely as shorthand for the discomfort of an image-generation company entering safety-critical medicine. None of this proves the scanner does not work. It explains why the people best positioned to judge medical imaging are asking to see the actual images before they believe the press release.