Midjourney reports Medical targets an end-2027 SF spa and 10x-lower hardware costs

• Midjourney used its AMA to pin down access: David Holz said the first public venue is a San Francisco spa targeted for the end of 2027, while Holz's trial timeline reply said research trials come first.
• The current scanner is still a physics story, not an AI one. In Holz's AI reply he said Midjourney is "not actually using any AI yet," and his capability summary described a conventional ultrasound system with known soft-tissue strengths and bone and lung limitations.
• Cost is one of the sharper reveals: Holz's hardware-cost reply put the machine at more than 10 times cheaper than MRI hardware, even as his bandwidth note said the current build is still choking on USB 3 throughput.
• Midjourney says the machine already works on real people, but only at tiny scale so far. Holz's scan-count reply said the team has scanned about 12 people and pointed to a scan gallery on Midjourney's website.
• The project is still small inside the company. Holz's team-size reply said a new team of nine built it, while more than 80 percent of Midjourney still works on image models.

The AMA filled in the parts the first announcement left fuzzy. You can browse the scan gallery, read Holz's note on the nine-person team, and compare that with Aakash Gupta's reconstruction thesis, which frames the whole thing as an imaging inverse-problem bet rather than a random hardware detour.

Midjourney's public plan is narrower than the launch-day reactions implied.

Holz said the scanners are headed to a Midjourney spa in San Francisco and that the company hopes it will be available by the end of 2027. In a separate reply, he said research trials will run before then, and that the spa is the main way the open public will be able to use it at first.

That makes the near-term product less like a general medical device rollout and more like a controlled venue with trial access first, then a branded physical location.

The most surprising answer in the AMA was how little AI is in the system right now.

Holz said Midjourney is not using AI yet, even though he thinks it could get much better once the pieces are combined correctly. He also said the current system already works, uses normal ultrasound, avoids radiation side effects, and can resolve features around half a millimeter.

The caveat sits in the same reply: it does not work as well through bone or lungs as it does in soft tissue. Gupta's commentary thread argues the real prize is reconstruction, but the company's own AMA answers describe a scanner that is still operating on ultrasound physics first.

The technical constraints are unusually concrete for an AMA.

Holz pegged present resolution at about half a millimeter, then said going past that would require sub-wavelength diffraction work and a bigger team. He also said the current system throws away data because USB 3 cannot stream everything from all sensors, and that the final version is moving to PCI.

In another reply, Holz's imaging-process explanation said the device emits large volumes of sound-wave data, terabytes by his description, and then reconstructs those streams into a stack of body images. The scan outputs are small. The raw capture is the firehose.

Midjourney is making a sharper economic claim than a medical one.

Holz said the hardware is more than 10 times cheaper than an MRI machine. He also said the scanner is not meant to replace MRI, X-ray, or other imaging systems, but to make much more body data available in a casual setting.

That pairing matters. Midjourney is not claiming diagnostic supremacy in the AMA. It is claiming cheaper hardware, faster access, and a setting that feels more like consumer wellness than hospital imaging.

For now, the project is closer to a lab build than a deployed platform.

Holz said Midjourney has scanned about 12 people so far and linked to the company's scan gallery. He also said the work was done by a new team of nine, while more than 80 percent of the company remains focused on image models.

That small footprint undercuts the idea that Midjourney quietly turned into a medical-device company overnight. The evidence points to a side program with real hardware, limited testing, and a long runway.

One other useful reveal is what Midjourney is not building itself.

Holz said Midjourney is collaborating with existing ultrasound companies such as Butterfly for the transducers. In a separate reply, he said the office machine shown in the teaser is physically real, even if the imaging process is not yet as push-button or productized as the video suggested.

So the novelty here is not a clean-sheet hardware stack. It is Midjourney assembling an ultrasound capture system, buying time with partners for key components, and trying to turn messy acoustic data into a usable imaging experience.