Let me get this straight. The company that made it trivially easy to generate photorealistic images of cats in spacesuits is now going to scan your internal organs?
That is essentially what Midjourney announced yesterday. CEO David Holz unveiled hardware called "The Midjourney Scanner" — a full-body ultrasound device that the company claims will deliver image quality "comparable to MRI in many ways." You read that correctly. An AI art company thinks it can do what a $3 million MRI machine does.
The partnership is with Butterfly Network, a real medical imaging company that makes semiconductor-based ultrasound chips. The scanner uses 40 Butterfly Ultrasound-on-Chip imaging modules arranged in a ring. The user steps onto a platform that descends into water while thousands of transducers fire ultrasonic waves through the body.
Yes. Water. It is basically a very expensive bathtub with a robot in it.
What Midjourney Actually Built (Versus What They Claim)
Here is what is actually happening when you step into the Midjourney Scanner. The ring of transducers captures what Holz calls "vertical slices" of your body composition — muscle, fat, bone, organ structure. The AI then processes these slices into a segmented visualization, which Midjourney has been showing as phantom imaging tests (basically a calibration dummy with known internal structures) to validate accuracy.
The company's job listings describe the long-term goal as building "the world's first full-body ultrasound CT scanner" that brings "safe, fast, and high fidelity preventative scanning to billions via a magical spa experience."
A magical spa experience. For medical imaging. To billions of people.
Look — we have been down this road before. Every few months some tech company decides that medicine is a software problem. Sometimes it works. Hardware-focused AI pivots are getting more ambitious, and sometimes the engineering is genuinely good. But the gap between "our phantom tests look clean" and "this is comparable to MRI" is enormous. Let me explain why.
The MRI Comparison Is Doing Heavy Lifting
MRI machines are not just good at one thing. They excel at soft tissue contrast. They can differentiate between a healthy tendon and a partially torn one. They can identify the subtle difference between a benign cyst and early-stage malignancy. They do this using magnetic fields and radio waves — fundamentally different physics than ultrasound.
Ultrasound, even AI-enhanced ultrasound, has inherent physical limitations. Sound waves do not penetrate bone well. Air-filled organs like the lungs scatter ultrasound waves unpredictably. The resolution ceiling for ultrasound is orders of magnitude below MRI for certain tissue types.
Here is where it gets frustrating. Holz said he "aims for image quality comparable to MRI in many ways." Note the phrase "in many ways." That is not the same as "equivalent to MRI." It is a carefully worded claim that could mean the scanner matches MRI for body composition tracking (fat vs. muscle vs. bone) while falling far short for tumor detection or neurological imaging.
The FDA's medical device division does not grade press releases on enthusiasm. It grades hardware on clinical validation data. Midjourney has not published any peer-reviewed validation yet. They showed a segmented phantom scan. That is step 1 of roughly 47 in the medical device approval pipeline.
Why Midjourney? Why Now?
This is where things get more interesting, honestly. David Holz was not always an AI art guy. He founded Leap Motion (hand tracking hardware), worked at Oculus, and has a background in computer vision and hardware engineering. He is not a stranger to building physical products.
But the timing feels deliberate. AI-generated images are getting commoditized. Midjourney faces competition from Stable Diffusion, DALL-E, Flux, and increasingly from open-source models. The image generation market is crowded and getting more crowded every quarter.
The biggest AI companies are making enormous bets to stay ahead. Medical hardware — if it works — is a completely different business. Recurring scans. Subscription models. Healthcare data. Insurance reimbursement. This is not a one-time image pack. This is a long-term revenue relationship.
Holz's own framing makes this clear. He wants people doing this "once a year or every single day." He said: "I'm not the most measured man on Earth yet, you know, but maybe I want to have that daily measurable information."
Daily full-body scans. That is not a medical recommendation. That is a lifestyle product.
Ultrasound AI Is Legitimate — The Question Is Scale
To be fair to Midjourney, they are not the first company to bet that AI can make ultrasound better. Butterfly Network itself (their hardware partner) has spent years pushing ultrasound-on-chip technology. Companies like Exo Works, Caption Health (acquired by GE Healthcare), and Butterfly's own Butterfly iXi have all demonstrated that AI can meaningfully improve ultrasound image quality and interpretation.
There are real, peer-reviewed results showing AI-assisted ultrasound outperforming traditional ultrasound for specific tasks — guiding novice operators, identifying cardiac abnormalities, detecting thyroid nodules. The technology works.
The question for Midjourney is not whether AI-enhanced ultrasound is real. It is whether a full-body ring configuration with 40 chips can actually deliver on the broad claims being made. And whether a company that has spent the last four years building image generation models has the institutional knowledge to navigate medical device regulation, clinical trials, and the kind of safety scrutiny that comes when you are scanning real human bodies.
Comparison: Midjourney Scanner vs. Current Full-Body Options
| Feature | Midjourney Scanner | Prenuvo Full-Body MRI | Ezra Full-Body MRI | Traditional Ultrasound |
|---|---|---|---|---|
| Technology | AI Ultrasound (40 chips) | MRI (1.0T+) | MRI (1.0T+) | Single probe ultrasound |
| Scan Time | Not disclosed | ~60 minutes | ~30 minutes | 15–45 min per region |
| Radiation | None | None | None | None |
| Soft Tissue Detail | Claimed "comparable to MRI" | Excellent | Excellent | Moderate |
| Bone Imaging | Limited | Very Good | Very Good | Poor |
| Price Per Scan | Not announced | ~$2,499 | ~$1,750 | $200–$500 per region |
| FDA Status | Not yet submitted | 510(k) cleared | 510(k) cleared | Various clearances |
| AI Processing | Core to product | Assisted | Assisted | Minimal |
A few things jump out here. The existing full-body MRI options are already expensive and take a long time. If Midjourney can deliver meaningful body composition data faster and cheaper using ultrasound, there is a genuine market there — just not the one they are advertising. Body composition tracking is not the same as cancer screening. Calling it "comparable to MRI" blurs that line in a way that could mislead consumers.
The Bigger Picture: AI Companies Are Getting Physical
Midjourney is not alone in this trend. We saw earlier this week Genesis AI unveil a non-humanoid robot that ditches the humanoid form factor entirely. Bezos invested $12 billion into an "artificial general engineer" at Prometheus. SpaceX bought Cursor for $60 billion to transform how engineers work.
The pattern is clear. The companies that started as pure software plays are moving into physical hardware. Software-only AI is getting commoditized. The next moat is built with atoms, not just tokens.
Regulatory pressure on AI labs is intensifying simultaneously, which makes a hardware pivot more attractive from a business perspective. Medical devices have a defined regulatory pathway. You can navigate it. You can get clearances. The rules are slow but they exist. AI model regulation — not so much.
What Should Actually Make You Skeptical
Three things:
First, the "millions of users via spa experience" framing is marketing, not engineering. Medical devices need clinical validation. Spa aesthetics do not substitute for sensitivity and specificity data.
Second, Holz's personal enthusiasm for daily scanning has no clinical basis. No medical body recommends daily full-body scans for healthy individuals. The noise-to-signal ratio would generate enormous anxiety and unnecessary follow-up procedures. More scans does not automatically mean better health outcomes. It means more data, which is different from better decisions.
Third — and this one bugs me — the phantom scan they released shows clean segmentation in controlled conditions. Controlled conditions are not real patients. Real patients move. Real patients have unusual anatomies. Real patients have metal implants or pregnancy or the thousand variations that make medicine hard.
Bottom Line
Midjourney is not wrong that AI-enhanced ultrasound has potential. The technology is real, the partner is credible, and the ambition is impressive. What makes me skeptical is the gap between the marketing claims and the clinical evidence. "Comparable to MRI" is an extraordinary claim that requires extraordinary data — data that does not exist yet.
If they can deliver a genuinely useful body composition scanner at a fraction of MRI pricing, that is a legitimate product. If they position it as a cancer-screening MRI replacement before the clinical trials prove it, that is irresponsible. Right now, based on a phantom scan and some ambitious job listings, it is too early to know which path they will take.
We will be watching the clinical data when it arrives. Until then, keep getting your regular check-ups from actual doctors with actual FDA-cleared machines.
Stay Updated on AI + Health
AI is moving fast into medicine, robotics, and infrastructure. Some of it works. Some of it is hype. We cover both. Subscribe to OnlyGFs.ai for honest, skeptical takes on what AI companies are actually building versus what they are promising.
Sources
- The Verge — "Midjourney goes from generating cat images to full-body ultrasound scans" (June 17, 2026)
- Butterfly Network — Official website (hardware partner)
- FDA — Medical devices regulatory portal
Frequently Asked Questions
