Midjourney launches full-body ultrasound CT scanner, pivots to medical hardware

Strip away the spa imagery and the Midjourney Scanner is a bet on one specific hard problem: turning chaotic acoustic echoes into a clean three-dimensional picture of the human body. Ultrasonic tomography fires sound through tissue from every angle and records what scatters back; the raw signal is effectively noise, and reconstructing organs and muscle from it is a brutal inverse problem that has stalled previous attempts for years. Founder David Holz framed the device as the first genuinely new whole-body imaging method in roughly fifty years ^[2], and the pitch implicitly leans on Midjourney's core competence — generating a coherent image from ambiguous input.

Here is the twist almost every breathless thread missed: the shipping prototype reportedly uses no AI at all. Holz told reporters the system is 'just really cool hardware and software' so far, with the machine-learning layer still to come ^[1]. That single admission reframes the story. The 'AI image company builds AI scanner' narrative is, for now, marketing; what exists today is a dense sensor array and a conventional reconstruction stack. The genuinely interesting question is not whether Midjourney can image a body — it is whether its generative-model expertise will eventually let it reconstruct sharper anatomy from sparser, faster, cheaper scans than physics-based methods allow. That is the actual moat, and it has not been tested yet.

The hardware at the center of the Midjourney Scanner did not come from Midjourney. It is built on Butterfly Network's Ultrasound-on-Chip platform — roughly forty semiconductor modules per machine, totaling on the order of 358,000 transducer elements, paired with more than two petaflops of compute to process the torrent of data each scan produces ^[3]. Midjourney licensed that technology in a deal first disclosed in a November 2025 securities filing: about $15 million upfront, $10 million a year over five years, and up to $74 million in total ^[3].

The market reaction told the real story about where value accrued. When the scanner was unveiled, Butterfly's stock (NYSE: BFLY) spiked to a multi-year high, with reported intraday gains ranging from the high teens to north of fifty percent ^[4]. TD Cowen analyst Joshua Jennings argued the launch exposed the 'potentially underappreciated value' of Butterfly's chip platform beyond handheld ultrasound — validating it as licensable infrastructure that other companies could build novel imaging products on top of ^[4]. In other words, the most durable lesson of the announcement may be that ultrasound-on-a-chip has quietly become a platform, and Midjourney is its first marquee customer — not its inventor.

Midjourney is positioning the first deployments inside a 'Midjourney Spa' rather than a clinic, and that choice is strategic, not aesthetic. By launching with non-diagnostic 'body composition maps,' the company can operate under the FDA's general-wellness policy and begin scanning people now, while pursuing diagnostic clearance incrementally ^[6]. Even the name does quiet work: calling the device an 'Ultrasonic CT' is slightly misleading, since a CT is X-ray computed tomography and this machine uses no X-rays at all ^[6].

The claimed specifications are genuinely striking: a target full-body scan in under sixty seconds versus the sixty to ninety minutes a full-body MRI typically takes, resolution down to a fraction of a millimeter, and a long-term plan for roughly 50,000 scanners performing up to a billion scans a month ^[5]. But the gap between the keynote and the lab bench is wide. The current prototype reportedly takes closer to twenty minutes per scan, not sixty seconds, and has been used on only about a dozen people ^[7]. The sub-minute, dramatically-faster-and-cheaper-than-MRI figures are aspirations for future hardware generations, not measured results from a shipping device ^[2].

Beneath the astonishment sits a serious medical objection that has little to do with whether the hardware works. Screening healthy people's entire bodies tends to find things — and most of what it finds is harmless but alarming. University of Michigan radiologist Matthew Davenport has noted that 15 to 30 percent of diagnostic imaging in adults already turns up at least one incidental finding ^[7]. Scale that to a billion scans a month and you are potentially flagging hundreds of millions of people for follow-ups, biopsies, and anxiety over 'incidentalomas' that would never have caused harm — the overdiagnosis trap clinicians have spent two decades warning about.

The skeptics' second point is about track record. As one tech publication put it bluntly, Midjourney 'makes pictures' and has 'never built a physical product, never operated a medical device' ^[1]. The device has no FDA clearance, no published clinical validation, and no independent comparison against MRI or CT. The reaction online captured the tension well: awe at the ambition, threaded with pointed warnings. Medically literate commenters reached repeatedly for two cautionary names — Theranos, the blood-testing company that promised a diagnostic revolution without evidence, and the Therac-25, the software-controlled radiation machine whose race-condition bugs killed patients in the 1980s — a reminder that software running safety-critical medical hardware can fail in ways a demo never shows. None of this means the scanner won't work. It means the burden of proof has barely been touched.