Here's what macOS 26.7 tells us about Apple's upcoming AirPods with Visual Intelligence
You've probably seen the leak by now. macOS 26.7 RC includes an intro video showcasing a person presenting a book cover to their AirPods and saving information for later. Then, a localization table included strings about setting up AirPods with Visual Intelligence, and how AirPods cameras will handle being covered by hair.
But that's only half of the story. macOS 26.7 RC also includes AccessorySensorManager, which controls the image capture on the new devices.
TLDR:
- A color camera on each AirPod, can take up to 1 megapixel images.
- The two cameras capture synchronized image pairs.
- The system supports one-shot and periodic still-image capture, with an active and background mode.
- There's person detection, and a hardware indicator with controllable brightness.
Here's what I've been able to find based on code that ships in macOS 26.7 RC.
Do the upcoming AirPods have cameras?
Code in macOS 26.7 suggests that each AirPod has its own camera sensor. Apple refers to the hardware sensor as "Kino," and identifies left and right CVCAM (Computer Vision camera) devices. [1] macOS also tracks the left and right camera hardware independently in code. [2]
Are they color cameras, IR, or another type of sensor?
macOS receives color images from the AirPods. The sensor output uses the CoreVideo rg3h format, which contains three 16-bit red, green, and blue planes. [3] CoreVideo explicitly describes the format as ContainsRGB = 1 and ContainsGrayscale = 0. [4]
Still, the code doesn't rule out that AirPods themselves may have additional sensors that do not participate in the image capture pipeline.
Do both AirPods take a picture simultaneously?
Yes, the cameras are designed to act as a synchronized pair. Each AirPod earbud captures a still frame with a specific frame ID. macOS considers a picture complete only when the left and right frame IDs match, explicitly treating a mismatch as an error. [5]
If only one earbud is able to produce a picture, macOS can create an empty placeholder for the missing still. [6]
What resolution are the cameras?
The active capture mode starts with a 640×640 capture and returns a processed 1024×1024 image. That's about 0.4 megapixels of sampled image data, and a little over 1 megapixel for the output frame.
There is also a lower-resolution passive capture mode with a 320×320 capture that can return 320×320 or 512×512 output, depending on the capture mode. [7]
These are capture-specific resolutions, and the physical sensor's native resolution may be larger.
Are these still or video cameras?
AccessorySensorManager supports both one-shot and periodic capture of still frames. [8] There is no implementation for recording conventional video. This ties nicely with the current implementation of Visual Intelligence, which is optimized for image understanding.
In active mode, the framework has two options. It can request a synchronized pair of still images with matching frame IDs from each earbud. Or it can provide a capture rate, and AirPods will continue sending pairs of images at the specified rate. In the one-shot case, the capture rate is 0. [9]
What's active and passive capture?
Active mode is a deliberate, higher-resolution capture. The code doesn't identify what user action or system process starts it, it is easy to imagine that a Siri request may trigger an active capture if the user is wearing AirPods.
Passive mode is lower-resolution and supports environmental awareness. In fact, macOS 26 includes a "contextual alarm" configuration with 5 possible conditions:
- Ambient speech, where speech is detected around the user. The code does not specify whether AirPods classify the speech or respond to any detected speech.
- Audio scene change, where the surrounding audio environment changes. Once again, the code doesn't specify the specific triggering conditions, but possible scenes may include speech, traffic, a room with loud music, or a quiet place.
- Posture change
- Head rotation
- Circular region, where the user has moved out of a defined spatial radius.
How are the cameras calibrated?
Because the earbuds may sit at different angles and move continuously while worn, the new devices have extensive options for sensor calibration. The framework includes:
- Intrinsics, which describe the camera's field of view, optical center, and how the lens bends the image.
- Extrinsics, which describe the camera's position and angle relative to the earbud, its counterpart, and the user.
- Distortion correction, which compensates for the curved image produced by a particularly small or wide-angle lens. This may give us some more indication about the physical sensor.
- Camera-to-IMU calibration, which allows the framework to account for head movement and gravity.
- Surface models, which help transform the raw image into a user-relative view.
The framework can also set aside images with excessive motion blur or occlusion. [14] [15]
Is there any processing done on the AirPods themselves?
There's at least one mention of inference done on the earbuds themselves. The framework includes mentions of peripheral inference, which is used to detect if a person is in the camera's field of view. [16] [17]
Are there any privacy features included?
The framework contains code that allows it to remotely control a hardware indicator and its brightness. [18] [19] It's easy to imagine that this is some sort of camera privacy light. I did not find any indication that an unobstructed view of the privacy light is required for capture, but this may be handled by the AirPods firmware instead.
I could not find any code that showed that AirPods may suppress images when another person is detected in the image capture. In fact, the person detection mentioned above has a "person not detected, sent the image anyways" result, which points in the opposite direction. [17]
The links throughout the walkthrough have references to code added in macOS 26.7 so you can verify the claims or explore further on your own. They're from blacktop/ipsw-diffs on Github. They do amazing work and you should go give them a star.