On June 25, 2026, the World Intellectual Property Organization published a Midjourney application titled "System, devices, and methods for generating vocal data." The applicant is Midjourney Inc. The priority date is a US provisional filed December 16, 2024. The subject is not images. It is a wearable that listens to your throat without touching it.
That makes the filing more useful than a product leak. A patent application has to describe how a thing works, in claims you can read, with numbers attached. For anyone following Midjourney, it is the clearest public signal yet of what the hardware projects on its own site are for. The company's about page lists several unannounced hardware efforts next to a Medical project, all marked TBA.
Here is a ten minute reading routine that gets you from the abstract to a practical takeaway.
Step 1: Find the document, skip the coverage
Look up WO2026133050A1. Confirm four fields before you read anything else: applicant, priority date, filing date and publication date. Midjourney Inc filed the PCT on December 13, 2025, one year after the provisional, which is the normal cadence for a company that intends to pursue an idea in several countries. The legal status is pending. Nothing has been granted.
Second-hand summaries are already circulating and they compress the interesting part into one line about a collar. The claims are where the scope actually sits, so start there instead.
Step 2: read claim 1, then claim 39. Claim 1 is a method with three steps. Transmit an acoustic waveform through air to a person's skin. Measure the waveform that reflects back. Process that reflection into vocal data. That is the invention in one sentence, and everything else is narrowing detail.
Claim 39 is the device version. A transducer, a support that attaches to the body, plus a processor and memory mounted on that support. If Midjourney ever ships this, that is the shape it takes: a piece you wear, not an app.
Step 3: Check the numbers, because they are the tell
The dependent claims are unusually specific, and specifics are what separate a sketch from an engineering plan. Read them as a spec sheet for a device that does not publicly exist yet.
- Carrier frequency between about 40 kHz and about 300 kHz, well above human hearing.
- Standoff distance between about 5 cm and about 15 cm from the skin, so the transducer never makes contact.
- Field of view of about 30 to 120 degrees, with MEMS microphones named as the receiver.
- Sound pressure level between about 90 dB and about 135 dB measured at 30 cm.
- Targets: neck, jaw and chin, with the neck claims naming the larynx, thyroid cartilage and trachea.
The alternative embodiments are just as informative. One claim places the transmitter and receiver on opposite lateral sides of the body. Another uses an array of receivers. Another describes the support as a C-shape, U-shape or horseshoe, which is a description of an open collar resting on the shoulders rather than a headset.
Step 4: notice what the AI loop adds. Claims 25 through 30 are the part that makes this an AI product rather than a microphone. Vocal data gets transcribed to text, fed into a machine learning model or a natural language model, and the response is communicated back to the wearer. Power comes from an internal lithium cell with wireless charging, and communication can run over Bluetooth or WiFi with encryption.
Read together, the filing describes a voice interface: ultrasonic sensing of speech, a nearby language model, an answer spoken back. The background section states the motivation plainly. Contact throat microphones work, but they are uncomfortable, so Midjourney wants the same signal from a distance.
What the filing does not say. It does not name a product, a price, a ship date or a launch country. It does not promise accuracy, battery life or latency figures. The medical classification codes on the document, which include speech analysis for diagnostic purposes, are not a regulatory clearance. A pending PCT application is a placeholder for a family of national filings, not proof that a factory is tooling up.
What to do with this today
Nothing in your prompt workflow changes because of a patent. The value here is strategic, and it fits in a short checklist.
- Stop treating Midjourney as a website company. Its project page already lists hardware and medical work, and this filing shows the sensing layer those projects would need.
- Watch the alpha site, not the press. Alpha changelogs are where Midjourney announces features first, and the most recent one landed on October 9.
- If you build voice-driven creative tools, file this under competitive risk. A voice interface from an image lab points at prompt-by-speech, not a new image model.
- Give any hardware claim a two year horizon. Filing, publication, national phase and manufacturing are separate gates.
That is the honest takeaway. A 2024 provisional became a published PCT in June 2026, and its claims describe a non-contact ultrasonic collar that hears your voice through your neck and answers using a language model. It is the most concrete thing Midjourney has published about hardware, and it takes ten minutes to read.
Comments