A still capture is a two-step flow: choose the region on the page, then annotate and name it in the editor.
1. Selecting the region
Start a capture from the popup's Capture region button or the keyboard shortcut. The page dims, and the region you drag stays bright.
| Action | How |
|---|---|
| Draw a region | Click and drag anywhere on the page |
| Resize | Drag any of the eight handles on the selection edge |
| Move | Drag from inside the selection |
| Confirm | Press Capture Image, or press Enter |
| Cancel | Press Cancel, or press Escape |
The pill above the selection shows the size in CSS pixels and the native pixel size you will actually get — on a Retina display a 600×400 selection is captured at 1200×800.
2. Capture full tab
Full tab skips the selection step and captures the whole visible area of the tab. It does not capture below the fold — only what is on screen.
3. The private-data check
As the editor opens, Lens looks over the region it just captured for things that usually should not be shared, and shows a bar above the image listing what it found. Each one is outlined on the image so you can see exactly what would be covered.
It looks for password fields, fields named for a credential that have something in them, and text shaped like an API key, a JWT, a bearer token or a PEM private key. Unless you turn it off, it also flags email addresses and card numbers — card numbers are checksum-checked, so an order number is not mistaken for one.
- Tick or untick anything in the list, then press Blur all to cover the ticked ones in a single step that a single Undo reverses.
- Dismiss hides the bar and changes nothing.
- High-confidence findings are ticked for you; the rest are listed but left unticked.
The check runs on the structure of the page in your own browser. It uses no AI and sends nothing anywhere, and it is on by default. If you have connected an AI provider you can additionally let a model sort real credentials from documentation examples — see Settings.
Treat it as a safety net, not a guarantee. It recognises known shapes, so it will not catch a secret that does not look like one, or a name or figure that is sensitive only in context. Read the capture as well.
4. Annotating
After confirming, the editor opens with a toolbar above the image.
| Tool | What it does |
|---|---|
| Pen | Freehand drawing |
| Text | Click to place a text label |
| Rectangle | Outline a box |
| Ellipse | Outline a circle or oval |
| Arrow | Point at something |
| Blur | Pixelate a region — applied to the image data, not drawn over it |
| Undo | Step back one annotation |
| Reset | Discard all annotations and start from the original capture |
Blur is irreversible and that is the point. It rewrites the underlying pixels rather than covering them, so the original content cannot be recovered from the saved file by removing a layer. Use it for tokens, names and anything else that should not leave your machine — but check the result before saving, because Undo after a blur restores the pixels only within the current editing session.
5. Naming and saving
Below the image are the file name, an optional note, and the page title and URL that will be recorded alongside the capture.
- The name is prefilled from your file name pattern — see Settings.
- The note is stored in project_metadata.json for that capture. Use it for context a file name cannot carry.
- Press Save to write the file. A toast confirms where it went.
If you have connected an AI provider, a Suggest name & description button appears above the name field. It reads the text inside the region you captured and fills in a file name and a one-line description of what the screenshot shows. The description is added to your note rather than replacing it, and everything it writes is yours to edit.
This is the one feature that sends the captured region's text to your provider, because describing a screenshot is the task. It only ever runs when you press the button — never as part of saving. If your provider is the browser's built-in model or one on your own machine, that text does not leave the device either way.
The toast tells you the truth about where the file landed. If it says *Downloaded to Downloads*, the project folder was unavailable and the reason is in the message — most often folder access needs reconnecting.