The homography behind the four handles
A photograph of a document taken by hand is never square on. The page leans, the far edge looks shorter than the near one, and the text runs at a slight angle. That keystone distortion is the single thing that makes a phone snapshot look amateur next to a real scan, and it is not fixable by rotating or cropping.
What fixes it is a projective transform. Once you have marked the four corners, the tool sets up an eight-by-eight linear system from those four point correspondences and solves it by Gauss-Jordan elimination with partial pivoting, producing the three-by-three matrix that maps your leaning quadrilateral onto a clean rectangle. The flattening then works backwards: for every pixel of the output page it applies that matrix to find where the pixel came from in your photo and samples there with bilinear interpolation. Going output-to-source rather than source-to-output is what guarantees the result has no holes in it. Anything that maps outside your photo comes back white, so a slightly generous corner placement gives you paper-coloured margin rather than garbage.
Capture, drag, add page, export
- Allow the camera when your browser asks. The live preview appears and Capture becomes clickable.
- Fill the frame with the document and click Capture to freeze the frame.
- Drag the four corner handles onto the corners of the page, choose an Enhancement, and click Add page.
- Repeat for the rest of the document, then click Download PDF for one combined file or Download JPGs for separate images.
Retake throws away the current frame and puts you back on the live preview without adding anything.
Four enhancement modes, and what each one does to a pixel
The Enhancement menu is not a set of filters with vague names. Each option is a specific per-pixel function, and the default is Auto-enhance.
- Original (color) applies the perspective correction and nothing else, so a colour letterhead or a highlighted passage stays exactly as photographed.
- Auto-enhance (magic color) converts to Rec. 601 luma and then applies a fixed contrast boost, pivoting around mid grey. A paper pixel sitting at 220 clips to a clean 255 white, a duller 180 lifts to 204, and ink at 90 darkens to 73. Mid grey at 128 is the fixed point and never moves.
- Grayscale is the luma conversion on its own, no contrast change, which is the safest choice for a page with photographs on it.
- Black & white measures the flattened page’s own mean brightness and forces every pixel above it to white and every pixel below it to black. That adapts to how bright your particular scan came out, but it is one global cut for the whole page rather than a per-region adaptive threshold, so a hard shadow across one corner can wash that corner out entirely.
The queue, the page sizes and the file names
Each page you add joins a strip below the preview with numbered badges and per-page arrows to move it left or right plus a cross to delete it. The camera goes straight back to live so you can keep going. Nothing is committed until you export.
Download PDF builds one document with your pages in the order shown, each page given its own dimensions at one point per captured pixel, and names it from today’s date in the form scan-2026-08-01.pdf. Download JPGs saves each page separately as scan-2026-08-01-page-01.jpg and so on; with a single page the button reads Download JPG and the file name skips the suffix. If you would rather control the paper size yourself, take the JPGs to Convert JPG to PDF, which accepts the whole set at once and offers A4, US Letter or fit-to-image.
Camera permission, rear cameras and the retry button
The camera request asks for the environment-facing lens at roughly 1920 by 1080 as a preference, never as a hard requirement, so laptops and desktops with a single front-facing webcam are handled without a separate code path. While the browser is deciding, the preview area says so and the Capture button stays disabled, which is why clicking too early does nothing.
If you refuse the permission, or the browser has remembered a refusal, you get an explicit explanation and a Retry camera access button rather than a dead screen. Leaving the page stops the camera track and releases every preview image. Unlike the shared tools on this site, this widget surfaces real messages: capture before the video is ready and it tells you to wait a moment, and a degenerate corner selection tells you to spread the handles into a proper quadrilateral.
Lighting beats every enhancement mode
No transfer function recovers detail the sensor never captured. Lay the page flat, kill any hard shadow crossing it, and fill as much of the frame as you can, because the flattened page can only ever be as detailed as the region you selected. Hold the camera roughly parallel to the paper even though the perspective correction will handle the rest, since an extreme angle means the far end of the page is sampled from very few real pixels.
For a document that is going to be read by a machine rather than a person, scan it here and then pass the PDF to OCR PDF to add an invisible, searchable text layer. If one page came out too dark, Adjust Image Brightness & Color will take the exported JPG and give you live sliders. And if you only want a plain photo with no page flattening at all, Take a Photo with Webcam is the simpler tool. The rest of the document toolset lives on the PDF tools hub.