Midjourney pivoted from pixels to patients, and now we're getting a tour of the hardware — but not the data that matters.
The Summary
- Midjourney released a 20-minute behind-the-scenes video of its experimental ultrasound scanner, built from "ultrasound probes hacked apart and slapped on a glorified hot tub with an elevator"
- The image generation company wants to deploy these in spas for cheap, radiation-free full-body scans, but still hasn't published clinical validation or peer review
- An engineer at the company gave the tour — transparency theater without the scientific receipts
The Signal
Midjourney Medical is showing you the tank. The elevator. The Raspberry Pis. What it's not showing: peer-reviewed results, FDA pathway clarity, or comparative diagnostic accuracy against existing imaging. The video tour, hosted by tech YouTuber and Midjourney engineer Marcin Plaza, walks through a dunk-tank ultrasound setup that surrounds a human body with imaging probes. The pitch: radiation-free, low-cost whole-body scans that could democratize medical imaging.
The hardware story is straightforward. Take commodity ultrasound transducers, arrange them in a cylindrical array around a water-filled chamber, add some off-the-shelf computing power, and you've got volumetric imaging on a spa budget. Plaza describes it accurately as hacked-together components — which isn't necessarily a problem. Medical device innovation often starts in the garage.
"Midjourney is building medical hardware with the same move-fast energy it applied to image generation, but medicine has regulatory gravity that AI art never faced."
What's missing is the clinical validation trail. The video shows imaging phantoms — test objects with known structures used to calibrate scanners. Getting clean segmentation on a phantom proves your hardware can detect edges under controlled conditions. It doesn't prove you can reliably detect tumors, aneurysms, or early-stage disease in actual patients.
Traditional medical imaging companies spend years and millions on FDA approval pathways:
- Controlled clinical trials comparing new tech to gold standards
- Independent radiologist reviews for inter-rater reliability
- Safety data on false negatives and false positives
- Long-term outcome tracking to prove clinical utility
Midjourney hasn't published any of this. The company's move from generative AI to medical devices is bold, but the scientific community measures bold in peer review and regulatory approval, not YouTube production values.
The spa deployment strategy is clever as a business model — wellness imaging sits in a regulatory grey zone where diagnostic claims are carefully worded and customers pay out-of-pocket. But it creates a troubling precedent: medical-grade hardware marketed for "wellness" to sidestep clinical validation requirements.
The Implication
Watch how this plays out as a test case for AI companies entering regulated industries. If Midjourney can build working medical devices faster than medical device companies can ship software updates, it proves something important about innovation velocity. If it can't get FDA clearance or clinical adoption, it proves something equally important about regulatory moats. Either outcome shapes how the next wave of agent-driven hardware companies approaches healthcare, infrastructure, and other high-stakes domains where "move fast and break things" has real consequences.