Google launched Guided Vision within Gemini Live on compatible Android devices on October 1, 2026. The feature verbally describes what the camera sees and helps users aim it, for example when the requested item is outside the frame. Its intended users include blind people and people with low vision. Alongside its potential usefulness, Google makes clear at launch that the tool can make mistakes and is not intended for navigation, safe walking or obstacle detection.
The camera receives instructions as well as answers
According to Google's announcement, users can ask for descriptions of objects, the reading of small text or help finding an item nearby. If the image is insufficient, Gemini can suggest tilting the camera, moving it or changing the distance. The addition therefore also concerns collecting the information the model needs to answer.
The difference is practical. An unclear photograph of a label is not necessarily a problem with understanding the question; sometimes the words are simply not in the picture. Instructions to improve the framing may allow a more focused second attempt. Even then, a clear answer and a confident voice are not proof that every letter was read correctly.
Testing a tool like this can begin with a question that is easy to verify, rather than an item the user does not know at all. That makes it possible to identify whether the difficulty comes from the photograph, reading the text or interpreting the image, before moving to a more complex task.
How to get started, and what still needs checking
In the Gemini app's profile settings, enable Use Guided Vision in Live, open a Gemini Live conversation and share the camera. Google also describes access through Android accessibility shortcuts and the TalkBack menu. Android 9 or later is required, and availability depends on the regions and languages in which Gemini Live is supported.
Users in Israel should distinguish between the toggle appearing in their account and the quality of work on local material. The announcement does not provide dedicated test results for reading Hebrew. An initial trial can use familiar packaging, with text for which the correct answer is already known. This allows the description to be compared with the source before relying on the tool for a new task.
A useful test will also include different lighting conditions and the difference between large printed text and small letters. These are suggestions for personal testing, not capabilities we have verified. If the answer changes when the camera moves slightly, that should be treated as a reason to check again, rather than automatically choosing the more convenient version.
The boundary Google has set should remain part of everyday use: Guided Vision is not a medical device, a mobility aid or a replacement for a white cane. It may help with understanding a visual detail, but that does not imply it can decide whether it is safe to move through a space. The first trial question could be what is written on a familiar object; deciding whether to cross a road is not a suitable experiment.