What you saw and what you meant
“This is the layout the client rejected” is different from “this image contains a layout.” seenX keeps human notes separate from OCR and automatic descriptions. A model-generated guess should never become a claim you made.
One linked record
The design links an original image, extracted text, editable notes, source details, and search representations with a stable record identifier. Each part remains inspectable. Combining everything into one opaque vector would make corrections and source inspection harder.
Edit, search, export
Changing a note should update its searchable representation. Notes should remain readable and exportable without access to a particular model. Source moves, deletions, and index removal need explicit lifecycle behavior.
In this preview
Take a pretend screenshot, click a sample text box or drag a highlight, and save it with an optional note. Press Command/Ctrl K to return to search. Saved text and notes are searchable. Pins and regions use coordinates normalized to the full image, independent of preview size. Copy for assistant and Export JSON include saved notes, selected text, source details, original image dimensions, normalized targets, and pixel bounds. The Text tool uses authored text boxes; it does not run OCR. Clearing browser storage removes edits and sample captures. Older single-note edits migrate as whole-image notes.
Automatic source context
Capture time and source details can make an unannotated image useful. Active-app identity and browser URLs require a suitable capture integration and permissions. A generic image dropped in a watched folder does not reliably reveal the originating app, tab, or URL. Unknown context stays unknown.