Windows 11

Windows 11 Narrator’s Image Descriptions Fall Apart on Non-Copilot+ PCs

4 min read Editorial

Windows 11 gives Narrator the ability to look at whatever is on your screen and read back a description of it. Narrator image descriptions are meant to be a huge win for blind and low-vision users who need context about photos, diagrams, or UI elements they cannot see. In practice, the experience depends heavily on the hardware you are running, and that split is becoming hard to ignore.

According to reporting from Neowin, the feature behaves very differently depending on whether you own a Copilot+ PC or a standard machine. On Copilot+ hardware, the image descriptions work as advertised. On a regular Windows 11 PC, the descriptions frequently make no sense at all.

What Narrator’s image descriptions are supposed to do

Narrator is Microsoft’s long-standing screen reader, and the image description capability is one of its more ambitious pieces. When enabled, it uses artificial intelligence to analyze visual content and turn it into spoken or Braille output. A user can ask it to describe a screenshot, a photo inside a document, or an icon on the desktop, and the feature responds with a natural-language summary of what it “sees.”

Advertisement

For readers who cannot rely on vision, this closes a gap that a plain screen reader cannot fill. A traditional screen reader tells you that an image exists and reads any embedded alt text, but it cannot tell you what a chart shows, whether a photo is a meme or a family picture, or what a confusing error dialog is actually depicting. The AI-powered description is meant to fill exactly that hole.

A close-up of a modern laptop's internal AI chip glowing with blue light, representing the NPU in a Copilot+ PC, with a
On-device AI hardware powers Copilot+ PCs' faster, more accurate image analysis.

Why Copilot+ PCs handle it cleanly

The reason for the hardware split comes down to where the analysis happens. Copilot+ PCs ship with a dedicated neural processing unit, or NPU, rated for at least 40 TOPS of local compute. That AI hardware lets these machines run sophisticated on-device models without sending your screen data to the cloud.

When the image descriptions run on that local model, they tend to be accurate and fast, because the processing stays on the machine and does not depend on network conditions. Microsoft has been pushing this on-device AI approach across its Copilot+ lineup, from Recall to live captions, and the image description feature fits the same pattern.

What goes wrong on a regular PC

On a non-Copilot+ PC, there is no equivalent on-device AI accelerator doing the heavy lifting. According to Neowin’s testing, that leaves the feature falling back to something far less reliable, and the resulting descriptions can be confusing, incomplete, or just plain wrong.

The practical symptom is what you might expect from a description generated without strong local intelligence: the reader is told about the image, but the words do not match reality well enough to be useful. A diagram might be described as a cluster of unrelated shapes, or a simple photo might get a garbled summary that misleads rather than helps. For an accessibility tool, that is a serious problem, because a wrong description can be worse than none at all.

A person wearing headphones using a Windows accessibility screen reader interface, viewed from the side in a warmly lit
Users rely on Narrator for on-screen image context, but accuracy depends on their hardware.

What this means for you

If you use Narrator for accessibility, the headline takeaway is that your hardware now determines how well a core feature performs. The same feature, the same Windows 11 install, can be genuinely helpful on a Copilot+ PC and genuinely frustrating on an older machine. That is a new kind of divide: not just whether a feature exists, but whether it actually works for you based on the chip inside your computer.

It also raises a broader question about where Microsoft is taking accessibility. The company is clearly betting that on-device AI is the future of these features, which makes sense for privacy and speed. But it means users who have not upgraded to Copilot+ hardware may be left behind on capabilities that once worked across the board.

What you can do

There is no quick software fix that turns a regular PC into a Copilot+ machine, since the advantage comes from dedicated AI silicon. If you are on a standard PC and the image descriptions are unreliable, your best options are to lean on alt text where it exists, ask a sighted person to describe specific images, or plan a hardware upgrade if this feature is essential to your daily workflow.

For now, the feature is worth testing on whatever you own, but treat its output on a non-Copilot+ PC as a best guess rather than a reliable description. Microsoft has not yet clarified whether it will improve the fallback path for older hardware, so this remains a hardware-gated capability for the time being.

Source: Neowin

Over to you: Are you relying on Narrator’s image descriptions daily, and have you noticed the difference between a Copilot+ PC and a regular machine?

Advertisement
Share:
Editorial
Written by
Editorial

Windows & Microsoft news editor at 9to5Windows. Covering everything from Windows 11 builds to enterprise updates.

Advertisement