Midjourney's Full-Body Scanner: A 2026 Deep Dive

The AI image company that built a water-immersion ultrasound scanner, what it actually is, and what it is not.

Midjourney, a profitable AI image company with no outside investors, just unveiled a full-body ultrasound scanner and a stated goal of one billion scans a month. On June 17, 2026, founder David Holz used a two-hour livestream on X to introduce Midjourney Medical and its first piece of hardware, a machine the company brands the "Ultrasonic CT." You step into a ring of submerged transducers, sound passes through your body from every angle, and software reconstructs a three-dimensional map of what is inside you. Bloomberg picked it up the next morning under a blunt headline: an AI startup pivoting to health - Bloomberg.

The pitch is irresistible, which is exactly why it deserves scrutiny. Holz called it "the first new whole-body medical imaging modality in 50 years" and described the experience as "as powerful as an MRI, as casual as a trip to the spa" - latent.space. The company says it will eventually run 50,000 scanners worldwide, open a flagship "Midjourney Spa" in San Francisco, and make full-body imaging cheap enough for everyone. The internet responded predictably: one launch attendee said he had not walked out of a product demo thinking the world had changed since the first iPhone.

But here is the problem: almost none of those numbers describe a product that exists yet. The machine that was shown is a Gen 1 prototype that has scanned roughly a dozen people, the "60-second scan" is a target rather than a measured result, the images shown were "not yet using AI," and there is no FDA clearance, no peer-reviewed validation, and no published sensitivity or specificity data behind any of it. The most concrete, verifiable fact in the entire story is a licensing contract filed with the SEC.

This guide separates the physics from the marketing. It explains what the device almost certainly is, the genuine science of whole-body ultrasound tomography that predates Midjourney by decades, the Butterfly Network deal that actually makes it possible, the preventive-imaging market it is entering, the regulatory wall between a wellness gadget and a diagnostic device, the economics of the billion-scan dream, and the specific milestones that will tell you whether this becomes medicine or stays a beautiful demo. The aim is the insider read, written for a curious non-specialist.

Contents

1. What Midjourney actually announced
2. "Ultrasonic CT" decoded: what the machine really is
3. The Butterfly Network deal: the firmest fact in the story
4. The real science: whole-body ultrasound tomography
5. The limits sound cannot cross
6. Why an image company builds a body scanner
7. The preventive-imaging market it is entering
8. AI in medical imaging: where the intelligence actually goes
9. The regulation wall: wellness versus diagnosis
10. The economics of a billion scans a month
11. The skeptics: radiologists, researchers, and the spa critique
12. What to watch before you believe the hype
13. How to think about it: a decision framework

1. What Midjourney actually announced

The announcement is real and independently confirmed, which is worth stating plainly before everything else gets qualified. Midjourney stood up a new division, Midjourney Medical, and revealed a full-body ultrasound imaging device. Bloomberg reported the pivot the next day, and the trade publication R&D World, the most technically careful outlet to cover it, ran its own writeup the day of the event - R&D World. So the headline is not a hoax or a misattribution. A company famous for generating pictures has genuinely built a medical machine.

What is far less settled is everything past the headline. The unveiling happened on a livestream, not in a hospital or a peer-reviewed paper, and most of the specifications that flew around social media trace back to that single presentation. Midjourney described a process where you step into "a shallow pool of golden light," then descend so your body passes through a ring of underwater sensors, "each acting like a dolphin, using its echolocation." The marketing names the technology "Full Body Ultrasonic Computational Tomography" and asserts, without qualification, that "no such device has ever been built until now" - Digg. That last claim is itself contestable, as Section 4 will show.

Here is the part the headlines mostly skipped. Coverage of the actual presentation describes a machine that is unmistakably early-stage:

• A small team, reported at roughly nine people
• About a dozen people scanned in total so far
• Current scan time near 20 minutes, not the marketed minute
• Images shown "not yet using AI" for reconstruction
• No diagnostic claims at launch, only body-composition maps

That list reframes the whole story. The 60-second number that anchored the headlines is the engineering goal, contingent on future gains in bandwidth, algorithms, and signal processing, while the working device reportedly takes around twenty minutes per scan - latent.space. The distinction matters because a one-minute scan and a twenty-minute scan imply completely different products, economics, and clinical use cases. A guide that repeats "scans your whole body in 60 seconds" as a present-tense fact is repeating an aspiration, and this one will not.

The launch also carried organizational signals that this is a serious, sustained bet rather than a one-off stunt. Midjourney hired Ahmad Abbas, who previously worked on Apple's Vision Pro, as its head of hardware back in 2024, and reporting around the event described the scanner as the first of several hardware and software projects in the pipeline - Crypto Briefing. That context cuts both ways. It tells you Midjourney has been quietly building physical-product capability for a while, which makes the pivot less random than it looks. It also means the company is spreading itself across multiple frontier bets at once on a roughly 100-person headcount, which raises the question of how much focus a regulated medical device will actually receive. The "first new whole-body modality in 50 years" framing, meanwhile, is best read as rhetoric: tomographic and immersion ultrasound are decades old, as Section 4 details, so the novelty is in the integration and the ambition, not in the invention of a new way to see inside the body.

The regulatory posture announced alongside the hardware is the most revealing detail of all, and it is the strongest signal of how Midjourney itself understands the limits of what it has. At launch the scanner will offer only body-composition maps, measuring the rough distribution of muscle, fat, bone, and organ tissue, and it will make no claim to detect disease. The company says it has "started discussions" with the FDA and will submit data over time to expand what it is allowed to say. That is the same wellness lane that companies like Prenuvo and Ezra already occupy, and it is a deliberate choice to avoid the medical-device approval question on day one. Understanding why that lane exists, and how high the wall is on the other side of it, is most of the real story here.

2. "Ultrasonic CT" decoded: what the machine really is

Start with the name, because the name is doing a lot of work and some of it is misleading. "CT" means computed tomography, and in standard medical usage that means X-ray computed tomography: a rotating X-ray source builds a cross-sectional image from many angles. Midjourney's machine uses no X-rays. It uses ultrasound. R&D World flagged the label as a "category error," noting that calling an ultrasound device a CT, and calling it "MRI-grade," is marketing coinage rather than a description of the underlying physics - R&D World. The honest description is a whole-body ultrasound tomography system, and that is what the rest of this guide will call it when not quoting the brand.

What the brand gets right is the word tomography, which simply means building a cross-sectional image from many measurements taken around an object. That part is accurate and it is the genuinely interesting idea. Conventional ultrasound, the kind used in pregnancy scans, presses a single handheld probe against the skin and reads the echoes that bounce straight back. Tomographic ultrasound surrounds the body with sensors and measures not just the echoes but how the sound waves change as they pass all the way through, from one side to the other. With enough angles, software can reconstruct an internal cross-section. The water is not a gimmick: sound travels efficiently through water and poorly through air, so immersing the body removes the air gaps and lets a full ring of transducers couple to the tissue from every direction at once.

The imaging chain, stripped of marketing, looks like this:

The word "computational" in the branding is the honest core of the idea, and it is worth taking seriously. Unlike a handheld probe that shows you a live echo, a tomographic system collects an enormous volume of raw acoustic measurements and then computes the image afterward, solving a mathematical problem to infer the internal structure that best explains all of those measurements taken from every angle. This is why the system needs a server cluster rather than a screen, and it is the part of the pipeline where a company with deep reconstruction expertise could, in principle, contribute something genuine. The image you would eventually see is not so much captured as calculated, which is both the source of the approach's power and the reason its quality depends so heavily on software that, by Midjourney's own admission, is not yet doing the heavy lifting.

The hardware specifics that circulated should be treated as Midjourney's own unverified claims, because that is what they are. Coverage of the livestream cited a roughly 70-centimeter ring, around 40 ultrasound-on-chip modules, on the order of 358,000 total sensing elements, 17 gigabytes per second of data capture, and reconstruction on a cluster of servers - latent.space. Some of these figures do not even appear on Midjourney's own medical page, and transducer counts differed between outlets, which is exactly the pattern you see when numbers come from social posts rather than a spec sheet or a filing. None of it has been independently measured.

One claim from the user-facing chatter deserves a direct correction, because it shows how a guide can drift from reporting into amplification. The widely repeated line that the scanner "maps more than 25 organs and anatomical structures" does not appear on Midjourney's own announcement. What Midjourney actually says is softer and more defensible: the machine produces "a 3D map of the body, down to a fraction of a millimeter," and uses AI segmentation to identify internal structures, with an initial focus on muscle, fat, bone, and organs - Midjourney. There is no published count of organs it reliably resolves, and given the physics in Section 5, any such number would need clinical validation before it meant anything. The safe reading is that the device generates a volumetric body-composition map and labels broad tissue types, not that it delivers a verified atlas of two dozen named organs.

3. The Butterfly Network deal: the firmest fact in the story

If you want the one thing in this entire announcement that is documented rather than asserted, it is the supply chain. Midjourney did not invent its ultrasound hardware. It licensed it from Butterfly Network, a publicly traded company (NYSE: BFLY) that pioneered putting ultrasound transducers on a semiconductor chip. Because Butterfly is public, the deal had to be disclosed, and it was, in an SEC filing that gives us hard numbers instead of livestream slides - SEC EDGAR.

The terms are specific and they predate the public launch by seven months. On November 17, 2025, Butterfly granted Midjourney an exclusive, non-transferable license to certain ultrasound-on-chip technology within a specified field of use. Midjourney pays a one-time fee of $15 million, plus a $10 million annual license fee billed quarterly over a five-year term, plus up to $9 million in milestone payments, plus revenue sharing on any hardware Midjourney sells and separate payments for the chips themselves - Investing.com. Analysts put the guaranteed minimum at roughly $65 million across five years before milestones and royalties. The market took it seriously: Butterfly's stock jumped about 16% on the announcement - Yahoo Finance.

Why does an ultrasound chip matter so much that a contract for it is the spine of the story? Because the chip is the reason the whole concept is even arguable. Traditional ultrasound probes are built from hand-tuned piezoelectric crystals, diced and bonded and wired by the thousand, which is slow, low-yield, and expensive. Butterfly's approach replaces the crystals with capacitive micromachined ultrasonic transducers fabricated directly onto a standard semiconductor wafer, the same kind of process that prints computer chips. Founder Jonathan Rothberg, who previously commercialized DNA sequencing on a chip, described the economic logic bluntly: "We put all the elements onto a semiconductor wafer, then we can dice up the wafer to make 48 ultra low-cost ultrasound machines" - IEEE Spectrum.

Butterfly is not a startup either, which matters for judging whether the supply chain is dependable. Founded in 2011 by Jonathan Rothberg and Nevada Sanchez, it shipped the first handheld whole-body ultrasound probe that plugs into a phone, went public through a roughly $1.5 billion merger in 2021, and has iterated through several generations of its iQ probe, the latest cleared by the FDA in early 2024 - Mintz. For Midjourney, licensing a mature, regulated, public-company technology is far less risky than building transducers from scratch, and it is a large part of why the hardware claim is credible at all. The tension worth holding onto is that the very filing proving the partnership also frames the license as a field of use outside Butterfly's core diagnostic business, which is consistent with a wellness launch and quietly inconsistent with the grander medical framing the brand name implies.

That manufacturing shift is the entire affordability thesis in one sentence, and it is genuinely sound on its own terms. Butterfly's handheld probe sells for roughly $2,000 to $4,000 against $100,000 or more for traditional cart systems, because semiconductor economics let you amortize design across enormous volume and improve with each process node - MedTech Dive. A body scanner that strings 40 of these chips into a ring inherits that cost curve. There is, however, a wrinkle that the "Midjourney Medical" branding glosses over: the Butterfly filing describes the license as being for a field of use outside traditional medical diagnostics. That is consistent with launching as a wellness body-composition product, and it quietly signals that the diagnostic ambitions are a separate, harder, later fight.

4. The real science: whole-body ultrasound tomography

Midjourney's claim that "no such device has ever been built" is the kind of sentence that should make you reach for the historical record, and the record contradicts it. Whole-body and whole-organ ultrasound tomography using a water tank is not a 2026 invention. It is one of the oldest ideas in medical ultrasound, and understanding its lineage is the single best way to calibrate your expectations for what Midjourney can and cannot deliver.

The water-immersion approach dates to the 1950s. The surgeon and physicist Douglass Howry built immersion-tank cross-sectional scanners, including a contraption nicknamed the "Somascope" that mounted a transducer on the rotating ring gear salvaged from a B-29 gun turret and spun it around a patient submerged in a water cup, producing some of the first diagnostic-quality cross-sections of the human body - History of Ultrasound. The technique fell out of favor precisely because patients had to be submerged, the scans were slow, and the apparatus was immobile. Handheld probes won because they were practical, not because tomography was a bad idea. Midjourney is, in effect, betting that modern chips and compute finally make the old immersion idea worth revisiting.

It is also worth knowing that transmission ultrasound tomography already exists as cleared, marketed medical hardware, just for one organ at a time. Delphinus Medical Technologies received FDA premarket approval in October 2021 for SoftVue, a whole-breast ultrasound tomography system that immerses the breast in water and computes a speed-of-sound map, used as an adjunct to mammography for women with dense breasts - BusinessWire. QT Imaging built a similar water-tank transmission-and-reflection breast scanner with measured in-plane resolution better than 1.5 millimeters - PMC. These systems are the direct ancestors of what Midjourney is attempting, scaled from a single organ to the torso.

It helps to understand what these machines actually measure, because it is not the grayscale echo image most people picture from a pregnancy scan. Transmission tomography records how fast sound travels through tissue and how much it is absorbed along the way, then reconstructs speed-of-sound and attenuation maps alongside ordinary reflection images. Different tissues have characteristic sound speeds, so a dense mass shows up as an "early" arrival and fatty tissue as a "late" one, which is the physical basis for distinguishing tissue types without radiation or injected contrast. SoftVue's clinical value rests exactly there: added to mammography for dense breasts, it detects on the order of 20% more cancers as an adjunct, not as a standalone replacement - FDA. That adjunct-only status, after roughly a decade in market, is the most grounded preview of where a whole-body version is likely to land: a useful additional view, not a substitute for the scans it is being compared against.

The most relevant precedent is also the most recent, and it comes from a university lab, not a startup. In April 2026, the lab of Lihong Wang at Caltech published a whole cross-section ultrasound tomography system in Nature Biomedical Engineering. The setup is strikingly similar to Midjourney's: a patient sits in a water tank with the head above water, and a ring of 512 transducers scans up and down the body to image cross-sections - Caltech. This is the credible, peer-reviewed version of the same concept, and its published numbers are the realistic yardstick.

What that lab actually measured is sobering next to the marketing. The Caltech system imaged the abdomen in 10-second breath-holds across five healthy volunteers, achieved an in-plane resolution of about 0.9 millimeters, and tracked a biopsy needle at seven frames per second - PMC. Those are real, validated capabilities, and they are impressive. They are also a long way from a one-minute whole-body scan that resolves dozens of organs. The same paper is candid about what it cannot do, which is the subject of the next section, and that candor from a peer-reviewed source is the most valuable input we have for judging Midjourney's much louder claims.

5. The limits sound cannot cross

This is the section that the marketing cannot wish away, because it is governed by acoustics, not engineering effort. Ultrasound is mechanical sound energy, and sound behaves very differently at the boundaries between materials of different density. When the acoustic impedance of two tissues is similar, sound passes through and you get an image. When it is wildly different, almost all of the energy reflects at the boundary and nothing useful gets to the other side. This is not a limitation that better chips or smarter AI can remove. It is physics.

The numbers make the problem concrete. The acoustic impedance of air is roughly 0.0004, soft tissue is around 1.6, and bone is about 12 - StatPearls. Those mismatches are enormous, which is why sound cannot meaningfully penetrate the air-filled lungs, the gas in the bowel, or dense bone like the skull and ribs. It also attenuates fast: bone absorbs sound at roughly 20 decibels per centimeter per megahertz, against well under one for fat and liver. The practical consequence is that an ultrasound system, no matter how sophisticated, is largely blind to the lungs, the brain behind the skull, anything hidden behind bone, and anything obscured by intestinal gas.

This is precisely why the "replaces MRI and CT" framing is so misleading, and why it is worth being concrete about the division of labor between modalities. CT is the workhorse for lungs, bone, and trauma exactly because X-rays pass through air and image calcified structures well. MRI is unmatched for the brain and detailed soft-tissue contrast. Ultrasound is superb for soft tissue you can reach through a clear acoustic window, the liver, kidneys, blood vessels, muscle, and fat, and useless for the things behind air or bone. A whole-body ultrasound therefore cannot be a whole-body replacement for the modalities people fear most, because the cancers and conditions that drive screening anxiety frequently hide in exactly the tissues sound cannot reach. The Caltech paper says so directly, noting it expects "negligible transmission through bones like the vertebral body" and does not even attempt to solve for them - PMC.

There is a second, quieter physics tax that compounds the first: the deeper you need to see, the lower the frequency you must use, and lower frequencies carry less detail. High-frequency ultrasound produces beautiful, fine images but only near the surface, while the low frequencies needed to reach the center of a torso inevitably blur. Body habitus makes this worse, since more tissue means more attenuation, so the people most at risk for many conditions can be the hardest to image well. Motion is the third tax: a scan that takes minutes rather than a fraction of a second will smear anything that moves, and the abdomen and chest are full of motion from breathing, the heartbeat, and the gut. None of these are flaws in Midjourney's engineering. They are the standing constraints that every ultrasound system has to negotiate, and they are why the realistic near-term win is robust body composition rather than fine-grained organ diagnosis.

The honest picture of what these systems produce is also less photographic than a render suggests, and there is a published image to prove it. The Caltech team put their ultrasound tomography of an abdomen side by side with a clinical MRI of the same region, and the difference in detail is visible to a layperson.

It would be unfair to leave the impression that the modality is weak, because within its acoustic window ultrasound is genuinely excellent and still improving. It images soft tissue, blood flow, and fat and muscle distribution with no radiation and no contrast, in real time, which is why it is the most-used imaging modality in the world. The Caltech system even demonstrated tracking a biopsy needle at seven frames per second, hinting at interventional and monitoring uses that have nothing to do with the screening fantasy. The realistic, valuable product hiding inside Midjourney's announcement is precisely this: a radiation-free way to measure and track the parts of the body that sound reaches well, repeatedly and cheaply, over time. That is a real thing worth building. It is simply a smaller and more honest thing than "as powerful as an MRI."

There is one more limit that engineering can soften but not erase: resolution falls off sharply away from the imaging plane. The same Caltech work reports an in-plane resolution near a millimeter but an out-of-plane resolution of 15 to 25 millimeters, because the lower frequencies needed to penetrate the whole body trade away fine detail - PMC. A centimeter-scale blur in one dimension is fine for measuring body composition and tracking change over time, which is exactly what Midjourney is launching with. It is not fine for finding the small, early lesion that a screening enthusiast imagines they are paying to catch. Holding both of those truths at once is the key to reading this product correctly.

6. Why an image company builds a body scanner

The most common reaction to this announcement was bafflement: why is a company that makes pictures suddenly making medical hardware? The shallow answer is brand extension and ambition. The deeper, first-principles answer is more interesting, and it starts not with images but with the nature of the problem. Medical image reconstruction is an inverse problem. You have sparse, noisy measurements from sensors, in this case the way sound waves changed as they crossed the body, and you need to infer the dense, coherent three-dimensional structure that produced them. That requires a strong learned prior about what real human anatomy looks like, so the system can fill in what the sensors did not directly capture.

Now consider what generative image synthesis is. You have a sparse input, a text prompt or a latent vector, and you need to produce a dense, coherent image by drawing on a learned prior about what real-world scenes look like. Structurally, these are the same class of problem: reconstruct a high-dimensional signal from incomplete observations using a model of what is plausible. Seen this way, a company that spent years learning to turn thin inputs into rich, believable images has built exactly the muscle that volumetric reconstruction demands. The same conceptual bridge connects this story to the broader field of turning flat or partial data into 3D structure, which we covered in our guide to AI image-to-3D.

David Holz's biography makes the leap look less abrupt than the "AI art company" label suggests, and this is where the story stops being a stretch. Before Midjourney, Holz spent roughly thirteen years as co-founder and CTO of Leap Motion, building hardware that reconstructed a precise 3D model of the human hand from raw infrared sensor arrays. Before that, he did graduate work in fluid mechanics and research stints at the Max Planck Institute and NASA's Langley Research Center, and he holds well over a hundred patents and publications - The Org. A body scanner that fuses thousands of ultrasound sensors into a 3D volume is recognizably the Leap Motion problem at body scale, layered on top of the generative reconstruction Midjourney has built since. He is, at root, a sensors-and-reconstruction engineer who happened to spend a few years on pictures.

The trajectory of Midjourney's own products reinforces how natural the jump looks from the inside. The company moved from still images to a substantially new model architecture, then into video generation, then into research on world models and explorable 3D scenes, the kind of work that treats a generated environment as a navigable volume rather than a flat picture - TechCrunch. A team that has spent years getting machines to construct coherent three-dimensional worlds from sparse instructions has been rehearsing, in software, the exact representational problem a body scanner poses in hardware. Whether that rehearsal transfers to ultrasound reconstruction is unproven, but the intellectual continuity is real, and it is why people who know Holz's work treated the announcement as ambitious rather than absurd.

Two structural facts explain why Midjourney specifically could attempt this when better-funded labs did not. First, the company is fully bootstrapped with no outside venture capital, profitable from early on, and famously small, reportedly around 100 people generating estimated revenue in the hundreds of millions - The Information. Holz has described Midjourney as a "research lab" that is "not really financially motivated" and under no pressure to sell, which is precisely the freedom required to spend years and tens of millions on a speculative hardware bet. Second, its research had already left static images behind, moving into video and into world-model and 3D work of the kind that lets users generate and enter explorable environments, a lineage that connects to the broader push into generative worlds we examined through Google DeepMind's Genie world models.

Honesty requires flagging where this elegant story is partly mine and not Midjourney's. R&D World makes the sharp counterpoint that the technology Midjourney actually licensed, Butterfly's ultrasound-on-chip plus discriminative AI for guidance, has "no connection to image synthesis" and is not the company's generative model at all - R&D World. In other words, the "same inverse problem" framing is a coherent way to understand why this is not crazy, but it is an analyst's reading, not a disclosed architecture. The actual engineering overlap between Midjourney's image models and this scanner may, for now, be closer to branding than to shared code. The thesis explains the motive; it does not yet describe the machine.

7. The preventive-imaging market it is entering

Midjourney is not opening a frontier so much as crashing a party that is already crowded, well-funded, and controversial. A wave of companies has spent the last few years selling elective whole-body scans to healthy people who want to catch problems early, and the prices, funding, and growth tell you both why Midjourney sees an opening and why the category is contested. This is the consumer longevity and preventive-health movement, and the broader context of AI moving into clinical settings is something we mapped in our guide to applied AI in medicine.

The clearest signal of momentum is what people already pay out of pocket, with no insurance, for a scan that no major radiology body recommends for the asymptomatic. Prenuvo sells a roughly hour-long whole-body MRI for $2,499, raised a $120 million round in early 2025, and reported around $100 million in revenue and more than 110,000 members - CNBC. Neko Health, co-founded by Spotify's Daniel Ek, charges about $372 for a body scan, raised $260 million, and built a six-figure waiting list before expanding to the United States - Bloomberg. Function Health bought the MRI startup Ezra and now offers a $499 full-body MRI that runs in about 22 minutes - CNBC.

The capital flowing in is just as telling, and it explains why a bootstrapped Midjourney can fund a spa from cash flow while its rivals raise nine-figure rounds. Function Health closed a $298 million round at a $2.5 billion valuation in November 2025, while Neko's raise valued it near $1.8 billion - TechCrunch. The blood-testing and biomarker side of the same movement is large too, with Function reporting tens of millions of lab tests completed. Market researchers project the broad longevity category growing from roughly $27.6 billion in 2025 to $67 billion by 2035, though such figures use elastic definitions and should be read as directional - Longevity.Technology.

What unites these companies, and what Midjourney is betting on, is a shift in who pays and why. None of these scans are reimbursed by insurance, because no guideline body recommends them, so the entire market runs on consumers paying cash for reassurance and early detection. That is simultaneously the opportunity and the warning. The opportunity is a large, fast-growing pool of health-conscious, higher-income customers who have already shown they will pay thousands for a scan. The warning is that a market built on cash-pay reassurance is only as durable as customers' belief that the scans actually help, and that belief sits on contested clinical ground. Midjourney's spa model leans into the consumer-ritual framing harder than anyone, which maximizes both the upside and the exposure if the medical establishment pushes back.

The category also carries a cautionary tale that Midjourney should study, and it is worth correcting a common confusion about who actually failed. The startup that collapsed was not the imaging firm Q Bio, which raised $27 million in 2024 and continues to operate; it was Forward Health, which shut down in late 2024 roughly a year after a $100 million round meant to roll out autonomous "CarePods" with built-in body scanning, after over-investing in technology rather than care delivery - eMarketer. The lesson is not that consumer health hardware cannot work. It is that hardware-heavy health concepts die when the unit economics and the clinical value do not arrive before the capital runs out. Midjourney is wealthier and more patient than Forward was, but it is buying into the same fundamental tension between a magical demo and a sustainable medical business.

8. AI in medical imaging: where the intelligence actually goes

If Midjourney is an AI company, the fair question is where the AI actually lives in a product like this, because the answer is not where the branding implies. The shown images were reconstructed without AI, by the company's own admission, which means the day-one intelligence is modest. But the broader field gives a clear map of where machine learning genuinely adds value in imaging, and it is worth understanding that map to judge where Midjourney could eventually win and where it is merely catching up.

The first and most mature application is image reconstruction itself, turning faster or noisier raw data into a clean image. GE HealthCare's TrueFidelity was the first FDA-cleared deep-learning reconstruction engine for CT, and its AIR Recon DL does the equivalent for MRI, letting scanners run faster while preserving quality - Applied Radiology. This is exactly the lever Midjourney needs to pull to get from a twenty-minute scan to a one-minute scan: use a learned prior to reconstruct a good volume from less data and less time. It is also the most defensible part of the "this is an AI problem" argument, because reconstruction genuinely is one.

This is also where Midjourney's day-one modesty is most telling. By admitting the launch images were reconstructed without AI, the company is conceding that the part of the system most aligned with its expertise is the part still to be built. That is not damning, but it does invert the popular story. The headline framing is "AI company builds scanner," while the honest near-term description is closer to "hardware integrator licenses chips and has not yet applied its AI." The interesting question for the next two years is whether Midjourney's reconstruction research can compress a twenty-minute acquisition into something near a minute while keeping the images legible, because that single capability is the hinge on which both the user experience and the economics turn.

The second application is detection and triage, and here the scale of existing adoption is striking. The FDA has now authorized well over a thousand AI-enabled medical devices, and radiology accounts for the overwhelming majority, on the order of three-quarters of all such clearances - The Imaging Wire. Companies like Aidoc and Viz.ai have dozens of cleared algorithms flagging strokes, clots, and other findings across thousands of hospitals. This is the regulated, validated world that Midjourney has explicitly chosen not to enter yet, and it shows how much evidence the FDA expects before a machine is allowed to say something is wrong with you. The same trajectory of AI being trusted with real clinical judgment is visible in systems like DeepMind's diagnostic agent, which we examined in our guide to the AI co-clinician, and in focused diagnostic plays such as the heart-imaging startup Cleerly.

The frontier application is the one most aligned with Midjourney's actual strengths: foundation models for imaging, generalist systems trained on huge volumes of scans that can segment anatomy, describe findings, and adapt to new tasks. Research models like RadFM have been trained on roughly 16 million image-text pairs across X-ray, CT, MRI, and PET, pointing toward radiology systems that behave less like single-purpose detectors and more like flexible interpreters - Nature Communications. If Midjourney's scanner ever generates the volume of body scans the company dreams of, the resulting dataset could in principle train exactly this kind of model, which is the most credible long-term moat in the whole plan.

It is worth dwelling on why that data moat is the most strategically serious part of the plan, because it is the one place where being an AI company is a genuine advantage rather than branding. Imaging AI is bottlenecked by labeled data, and the most valuable dataset is large, consistent, and longitudinal: the same people scanned repeatedly over years on identical hardware. A spa-subscription model that scans members monthly would generate exactly that, a standardized, time-series body archive at a scale no hospital network can match. If reconstruction quality reaches the point where the maps are clinically meaningful, that corpus could train segmentation and detection models that improve with every scan, creating a flywheel. The enormous caveat is that the flywheel only spins if the underlying images are good enough to learn from, which loops straight back to the unresolved physics and the absent validation. Data scale cannot rescue an image that the laws of acoustics blurred in the first place.

The structural point worth sitting with is that AI here is migrating from a tool that produces outputs on a screen to a system embedded in machines that act on the physical world. That same migration is reshaping software, where autonomous systems increasingly run real operations rather than just answer questions, a shift we explore in our look at the autonomous agent workforce. Platforms like O-mega sit on the software side of that line, running fleets of AI agents that carry out business processes end to end rather than generating drafts for a human to execute. The founder behind that platform, Yuma Heymans ( @yumahey), has spent his career pushing AI out of the chat window and into real operational work, first in recruiting and now across whole companies, which is the same instinct, in software, that Midjourney is acting on in hardware: intelligence is most valuable when it leaves the screen and touches the world.

9. The regulation wall: wellness versus diagnosis

Everything in this story eventually runs into a single hard boundary, and Midjourney's careful launch language proves the company knows exactly where that boundary is. There are two completely different regulatory universes here. One is the general-wellness lane, where a product can describe your body composition and encourage healthy habits as long as it makes no claim about disease. The other is the medical-device lane, where any claim to detect, diagnose, or rule out a condition triggers FDA review, clinical evidence, and potentially years of work. Midjourney is launching firmly in the first lane and gesturing at the second.

The wellness lane is real and it is how the incumbents already operate. The FDA's general-wellness policy lets low-risk products avoid premarket review provided their labeling and interface never reference specific diseases or recommend clinical action - FDA. A "here is your muscle, fat, and bone distribution" map fits inside that lane, which is why Prenuvo and Ezra can sell scans today without each one being a cleared diagnostic device. Midjourney's choice to launch with body-composition maps and explicitly "started discussions" with the FDA is not timidity, it is the only legal way to ship this year.

The diagnostic lane is where the timeline gets brutal, and it is worth being concrete about why. A novel whole-body AI scanner has no existing predicate device to point to, so the fast 510(k) route, which clears products by showing they are equivalent to something already on the market, is largely unavailable. That pushes the diagnostic version toward De Novo or premarket approval, the pathways for genuinely new or higher-risk devices - FDA. Even the review clocks are long, with De Novo decisions averaging well over 300 days in practice and PMA typically requiring clinical trials on top of that, and those clocks do not include the multi-year effort to generate the evidence in the first place. The path from a body-composition map to "thousands of diagnoses" is, in the words of one analysis of the launch, "extremely uncertain."

The practical upshot is a timeline mismatch that the marketing papers over. The wellness product can ship in 2027 because it sidesteps the device-clearance question entirely, but the diagnostic capabilities that would justify the "as powerful as an MRI" rhetoric live on the far side of a multi-year evidence-and-review process that has not visibly begun. The FDA reissued its general-wellness guidance in early 2026, and the line it draws is bright: the moment a product references a specific disease or recommends clinical action, it leaves the wellness lane and becomes a regulated device - FDA. Midjourney can legally sell body-composition maps next year, but it cannot legally tell you a scan found a tumor until it has done the clinical work, and nothing in the announcement suggests that work is close.

There is a deeper objection that no amount of regulatory cleverness resolves, and it is the strongest argument against the entire concept of casual whole-body screening. The FDA itself states it knows of no scientific evidence that whole-body scanning of people without symptoms does more good than harm, and notes that essentially every major medical body, including the USPSTF and the American College of Radiology, declines to recommend it - FDA. The ACR's formal position is that there is insufficient evidence to justify total-body screening of asymptomatic people, and that such screening tends to surface "numerous non-specific findings" that drive unnecessary follow-up testing and expense - ACR. A spa that scans healthy walk-ins is, definitionally, the asymptomatic screening that the medical establishment has spent two decades cautioning against.

10. The economics of a billion scans a month

Midjourney's most quotable claims are economic: a billion scans a month, 50,000 scanners, a cost of "a few dollars" per session, and the assertion that fewer than a dozen of these rings could outscan every MRI machine on Earth. These are the lines that make the announcement feel world-changing, and they are also the lines most worth subjecting to arithmetic. The point is not to dunk on the ambition but to understand exactly which assumptions have to hold for it to be real.

Start with the denominator, because scale claims only mean something against a baseline. The world has roughly 38,000 to 50,000 MRI machines and performs somewhere north of 100 million MRI scans per year, with the United States alone doing around 40 million - Collective Minds Health. Midjourney's target of a billion scans a month works out to 12 billion a year, which is on the order of 120 times the entire planet's current annual MRI volume. That is the scale of the promise: not incrementally more imaging, but two orders of magnitude more.

The per-machine math is where the aspiration meets the prototype, and it produces a clear verdict. Twelve billion scans a year across 50,000 scanners is 240,000 scans per machine per year, or about 657 per machine per day. At the marketed 60-second scan time, a machine running around the clock could in principle do up to 1,440 a day, so 657 implies a plausible duty cycle of a little under half. At the current twenty-minute scan time, however, a 24/7 machine caps at about 72 scans a day, leaving the target roughly nine times out of reach. The billion-scan dream is therefore not arithmetically impossible, but it is arithmetically dependent on actually hitting the 60-second speed that does not yet exist, and on running machines nearly continuously, which spas full of customers do not do.

There is a softer constraint hiding inside that duty-cycle math, and it is human rather than technical. A spa is a scheduled, daytime, appointment-driven business, not a 24-hour factory, so real-world utilization of a flagship location with ten scanners will be a fraction of the theoretical maximum even at the target scan speed. To hit a billion scans a month, Midjourney would need not just fast machines but a vast, distributed fleet running at high occupancy across many time zones, which is an operations and real-estate problem as much as an imaging one. This is the same wall that sank Forward's CarePod ambitions: scaling physical health locations is slow, capital-intensive, and unforgiving, and it does not benefit from the near-zero marginal cost of software. The billion-scan figure is best understood as a statement of aspiration about a decade-scale build-out, not a near-term capacity that follows automatically from a clever chip.

The cost story is similarly two-sided, and the ultrasound-on-chip foundation gives it real credibility on one axis while the marketing oversells it on another. The genuine advantage is capital cost: there is no superconducting magnet, no cryogenics, no shielded suite, and the transducers ride a semiconductor cost curve, so the marginal cost of an individual scan, once a machine exists and a customer is in the water, can genuinely be small. Why ultrasound is structurally the cheap modality is clear when you line up the hardware.

The oversell is the word "affordable for everyone," because the marginal cost of a scan is not the cost of the system. Midjourney's own framing put the marginal cost at "effectively zero" while explicitly excluding facility, staffing, regulatory, physician review, liability, cleaning, and real estate, and the company estimated something like $20 billion in capital to build out the fleet - latent.space. Spread across 50,000 scanners, that is roughly $400,000 per installed unit, which is premium-MRI territory. The economics, in other words, do not come from the device being individually cheap. They come from amortizing a large capital base across enormous throughput and spa membership revenue, which loops right back to the unproven assumption that the 60-second scan and the near-continuous utilization both arrive. The affordability is real only if the volume is real.

11. The skeptics: radiologists, researchers, and the spa critique

A guide that only relayed the company's claims would be doing public relations, so it is worth giving the critics their full say, because the most informed reactions to this launch were also the most skeptical. The criticism clusters into three categories: the images are not good enough, the screening model is harmful, and the spa framing is a red flag. Each is grounded in evidence rather than reflex.

On image quality, the sharpest critique came from people who read scans for a living. On the main Hacker News discussion, a self-identified radiologist judged the released images close to useless for medicine: "Other than the shapes of the tissues in the images, there is no anatomic detail. Wouldn't be useful for diagnostics," adding that it looked "substantially worse than conventional ultrasound" - Hacker News. Another commenter pointed out that transmission ultrasound tomography "is not new," that the breast version (SoftVue) never displaced mammography, and that the "fundamental limits of sound waves (bone and gas)" plus slow acquisition mean abdominal and chest images will be motion-degraded. These are not anti-AI cranks; they are the exact failure modes Sections 4 and 5 predicted from physics.

On the screening model, the medical literature is unusually direct about harm, and this is the criticism that should weigh heaviest. Whole-body screening of healthy people generates incidental findings in a startling share of subjects while almost never catching the deadly disease the customer feared. Pooled evidence from whole-body MRI screening finds incidental findings in up to 97% of people scanned, with truly serious findings in a few percent and genuine cancers in only one to two percent, and no demonstrated survival benefit - Diagnostic Imaging.

Each of those incidental findings can trigger a cascade of follow-up imaging, biopsies, and anxiety, most of which leads nowhere good. The canonical example, cited in the radiology literature on overdiagnosis, is South Korea's thyroid-cancer episode, where screening drove a roughly 15-fold increase in detected thyroid cancer between 1993 and 2011 with essentially flat mortality, because the scans found small, indolent cancers that would never have caused harm - JACR. A monthly spa scan that finds something "interesting" in nearly everyone is a machine for manufacturing exactly this kind of cascade. The same overdiagnosis dynamic, incidentally, is part of why catching real signal early in domains like skin and overall health is harder than it sounds.

The third critique is cultural, and it is the one Midjourney invited with its own branding. Putting a medical scanner inside a luxury spa with hot tubs, saunas, and cold plunges reframes diagnosis as a wellness indulgence, which several critics found troubling. One Hacker News commenter wanted the device "tested and used by doctors in a hospital and not some spa," while another captured the broader unease about giving an image-generation startup your full interior anatomy: "Am I supposed to entrust full body scans to a startup?" There is also a genuine data-governance question, since a fleet designed to capture a billion body scans a month is simultaneously the most valuable medical-imaging training set ever assembled, owned by a private company. None of these concerns are disqualifying on their own, but together they explain why the most knowledgeable observers met a dazzling demo with raised eyebrows rather than applause.

The data question deserves more than a passing worry, because it is structural rather than reputational. A body scan is among the most intimate data a person can generate, and a fleet engineered to capture a billion of them a month would assemble the most valuable medical-imaging dataset ever held by a private company, governed by whatever terms a consumer clicks through at a spa. That concentration raises real issues about consent, secondary use, breach exposure, and what happens to a decade of someone's interior anatomy if the company is sold or pivots again. None of this is unique to Midjourney, but the scale it is targeting makes the stakes larger than for any single imaging clinic, and the wellness framing means much of it would sit outside the stricter rules that govern formal medical records. A prudent reader treats the privacy architecture, like the clinical validation, as something to be shown rather than assumed.

12. What to watch before you believe the hype

The right posture toward this announcement is neither dismissal nor credulity but patience, because the things that would prove it real are concrete, observable, and not yet present. Rather than relitigate the claims, it is more useful to define the specific milestones that will separate a genuine new modality from a beautifully marketed prototype. If these arrive, the skeptics were wrong. If they do not, the demo was the product.

The single most important signal is peer-reviewed validation with real numbers. Right now there is no published sensitivity or specificity, no disease-detection benchmark, and no clinical trial, which is the precise gap between a striking image and a trustworthy test. The Caltech work shows what credible validation looks like, with measured resolution, named limits, and results on identified volunteers. Until Midjourney produces the equivalent, every performance claim should be read as a hypothesis.

A short watchlist captures the rest of what matters:

• Scan time in reality, moving from 20 minutes toward the promised minute
• An FDA submission, not just "discussions," for any diagnostic claim
• Published accuracy data on a defined population
• The San Francisco spa actually opening near its end-of-2027 target
• Independent radiologists examining real scans, not renders

Each item on that list is a place where the marketing has run ahead of the evidence, and each is falsifiable. The scan time is a physics-and-engineering problem that will either be solved publicly or quietly missed. An FDA submission is a dated, traceable event. The spa has a stated timeline and a reported lease, so its opening or slippage will be visible. And the fastest way to puncture or confirm the "as powerful as an MRI" claim is to let working radiologists read actual patient scans, which has not happened. Watching these signals is far more informative than parsing another round of announcement copy, and it is the discipline that separates analysis from amplification, a stance we apply across our coverage of AI for scientific discovery.

It is also worth tracking the company's framing discipline, because a slide from wellness language toward disease language would be the clearest tell of overreach. As long as Midjourney sells "body-composition maps," it is on solid legal and scientific ground. The moment its marketing implies the scanner finds cancer or rules out disease without the clearance to back it, the gap between what is claimed and what is proven becomes a liability rather than a rounding error. The history of consumer health hardware, including the broader pattern of AI-generated business processes outrunning their evidence base, suggests that the companies that survive are the ones that let the validation catch up to the vision.

One more signal is worth watching, and it is the cheapest to read: who Midjourney hires and partners with next. A company genuinely committed to the diagnostic path will accumulate clinical, regulatory, and radiology talent, run trials at named institutions, and publish, all of which leave a public trail. A company content to sell a wellness gadget will instead invest in retail design, membership operations, and consumer marketing. The mix of those moves over the coming year will reveal which version of the plan Midjourney is actually executing, regardless of how the announcement was framed. Intentions are cheap and arrive all at once; commitments are expensive and reveal themselves slowly, and it is the slow reveal that tells you what this really is.

13. How to think about it: a decision framework

Step back from the noise and the structural picture is clearer than either the hype or the backlash suggests. Midjourney has done something genuinely notable and something genuinely overstated at the same time, and a good mental model holds both without collapsing into a verdict that the evidence cannot yet support. The way to reason about it is to separate the four layers that the announcement deliberately blends: the technology, the prototype, the product, and the dream.

The technology layer is real and grounded. Whole-body ultrasound tomography works, it is radiation-free, the underlying chip economics are sound, and a peer-reviewed lab has demonstrated the core method. The prototype layer is early but legitimate: a small team has built a working ring that has scanned a handful of people and produces coarse volumetric maps. The product layer is a reasonable near-term bet: a wellness body-composition scanner sold through a spa, in the same legal lane and roughly the same price territory as Prenuvo and Neko, is a business that could plausibly exist by 2027. The dream layer, the billion scans a month, the diagnostic supremacy over MRI, the imaging-for-everyone-on-Earth, is unproven on every axis that matters: speed, accuracy, regulation, and economics.

Most failures of judgment about announcements like this come from collapsing those four layers into one, in either direction. Boosters treat the dream as if the prototype already delivered it, which is how you get headlines about a 60-second MRI-killer that maps two dozen organs. Cynics treat the prototype as if it discredited the technology, which is how you miss that radiation-free volumetric imaging on cheap chips is a genuinely worthwhile direction. The disciplined move is to grade each layer on its own evidence: high marks for the technology and the team, a passing-but-early grade for the prototype, a plausible-with-risk grade for the wellness product, and a not-yet-earned grade for the dream. That scorecard will shift as evidence arrives, and updating it honestly as the milestones in Section 12 resolve is far more useful than committing to a single verdict today.

The first-principles read is that the value of cheap, radiation-free imaging is genuine, but value in screening is not created by generating more images. It is created by generating actionable, validated information that changes outcomes, and that is precisely the part Midjourney has not yet built and the part the medical evidence says is hardest. The companies that win in health are not the ones with the most magical demo but the ones that close the gap between an image and a decision a doctor can stand behind. That is why the right judgment today is interested but unconvinced: the technology earns attention, the prototype earns respect, the product earns a wait-and-see, and the dream earns the same scrutiny any extraordinary claim deserves. For a sense of how concentrated and hype-prone this whole sector has become, our AI market power analysis is a useful companion read, as is the broader story of generative tools leaping into new mediums that began with products like Sora-class video generation and Nano Banana image models.

The most honest summary is the one Midjourney's own launch implied beneath the spectacle: this is the first day of a long medical-evidence project dressed up as a finished consumer marvel. The water, the ring, the chips, and the reconstruction are real engineering. The minute-long scan, the dozens of organs, and the planetary scale are intentions. Treat the engineering with curiosity and the intentions with patience, watch the milestones rather than the marketing, and you will read this story far better than either the people declaring the future arrived or the people insisting nothing happened. Something did happen. It just has not finished happening yet.

This guide reflects what was publicly known as of June 18, 2026, drawing on Midjourney's announcement, independent reporting, regulatory guidance, and peer-reviewed research. Specifications, regulatory status, and timelines for an early-stage prototype change quickly, and many figures cited here are the company's stated goals rather than verified results. Verify current details before drawing conclusions, and treat any unvalidated performance or diagnostic claim with appropriate caution.