Redacting embedded images
Documents carry sensitive pictures as well as sensitive words: an ID photo on a CV, a scanned signature on a claim, a photo of a car’s number plate in an accident report. When the model analyzes a document with vision, it surfaces those images so you can deal with them.
The Images section
Section titled “The Images section”If a document has flagged images, an Images section appears in the entries panel, above the text entries, with a thumbnail and a blur switch for each. The header shows how many were found and has a switch of its own that blurs every image at once (or none: switch it off). Click the header to fold the thumbnails away.
The same picture shown several times, like a letterhead logo on every page, gets a single thumbnail with a ×N badge, and its switch blurs every copy. Click the badge to see each copy with its page number and switch them one by one; click it again to group them back. When only some copies are blurred, the thumbnail has a dashed outline and its switch is half-tinted.
Flip a switch on, and that image is blurred in the regenerated output — heavily enough that a face, a plate, or a signature is no longer legible. Flip it off to leave the image as-is.
Because there’s no in-document preview for a blur (unlike text, which you can instant-redact), toggling an image switch nudges the anonymized view to regenerate so you can actually see the result.
Images vs. drawn redactions
Section titled “Images vs. drawn redactions”Two tools touch images, for two situations:
| Situation | Use |
|---|---|
| Piixie recognized the image (it’s a thumbnail in the Images section) | The blur switch here |
| A sensitive area the image pass didn’t flag — part of a page, a stamp, a region of a scan | Draw a redaction box |
Blur softens the whole recognized image; a drawn box paints a hard patch over an exact rectangle. For something that must be provably gone — a signature on a legal PDF — a Black box is the firmer choice. For a CV photo, blur is usually enough.
Format notes
Section titled “Format notes”- PDF — embedded images are analyzed and can be blurred or covered with a drawn box. This is the richest case.
- DOCX — Word pictures (JPEG, PNG, GIF, BMP, TIFF, WebP) are analyzed and blurred like a PDF’s, and listed here with their switches. A picture Word shows in several places, like a header logo on every page, is one picture with one switch, and its blur applies everywhere it shows. With the Page N filter on, the panel lists the pictures on that page. Drawings Word stores as EMF or WMF aren’t read and stay as they are, and the drawing tool is PDF-only. (See file formats.)
Example: a CV with a photo
Section titled “Example: a CV with a photo”A redaction-style CV PDF has the candidate’s headshot top-right and their name, email, and phone in the header. The model synthesizes (or redacts) the text and lists the headshot in the Images section. Flip its blur switch on, Save to output, and the delivered CV has anonymized contact details and an unrecognizable photo. The full walkthrough is in Anonymize a CV with a photo.