What if your wearable could capture the moments that matter and then recreate them with AI?
OpenAI has filed a patent for a system that selects photos from wearable devices and generates synthetic versions based on their visual content, metadata, and surrounding context.
Manually capturing photos requires the user to physically stop and pull out their device to take a snap. This can disrupt one’s experience and even lead to poorly framed, obstructed, or blurry photos on top of the large selection of photos. OpenAI’s patent application is designed to solve these problems by creating a software pipeline that can automatically filter the best photos, analyze their metadata, and even create a more expressive version of the original photo. Additionally, voice commands can be used to automatically capture images or record videos, eliminating the need for users to physically stop for photos.
How OpenAI’s synthetic image system works
The system can be implemented in wearable devices such as smart glasses and earbuds. The software inputs raw captured images and their metadata into an image selection engine. This engine filters the images and selects the most relevant photos based on the image itself and additional information such as location, timestamps, and camera orientation.
The selected photo is then used to guide the creation of a synthetic version. The system generates text summaries describing the image and its surrounding context using the image and its metadata. It also assigns salience scores to subjects based on their prominence across the collection of original images. Furthermore, it isolates image segments that need to be accurately represented in the synthetic image, such as a person’s face. A secondary optical character recognition model can extract text captured within an image so it can be included in the text summary.
The prompt generation engine then uses the outputs from the image analysis engine to create a prompt for the final synthetic image. Finally, the image generation engine uses the prompt and, optionally, the isolated image segments and metadata to generate a synthetic version of the selected image. This generated version can be stored and displayed on the user’s device like a conventional captured photo.
Why it matters
This application could fit into the context-aware wearable ecosystem reportedly being explored by OpenAI and Jony Ive. Although the details of the anticipated device remain limited, reports have described a device that could use cameras and microphones to gather information from its surroundings and communicate with other smart devices.
U.S. Pat. App. Pub. No. 2026/0268535, titled “Synthetic Image Generation,” was filed on March 4, 2026 and published on September 10, 2026. The listed inventors are Douglas Stuart Brodie, Matthew Allen, and Adam Cue.