Loading page
About this tool
Compare two local screenshot previews through reviewed OCR transcripts with deterministic line and word change counts.
The Screenshot OCR Change Comparator diffs the text of two screenshots by comparing OCR transcripts you paste and confirm you have reviewed, running a longest-common-subsequence diff at both line and word level and reporting additions, removals, replacements, unchanged counts and a similarity percentage. It deliberately does not perform OCR itself — you use the recognition already built into your operating system or device, then correct it — because an unreviewed transcript turns recognition errors into fake changes. Screenshot previews and transcripts stay in the tab, and the exportable report contains counts only, never the text itself.
Open Screenshot OCR Change Comparator on AltFTool — it loads instantly in your browser.
Use 'Choose screenshot' on the 'Before screenshot' and 'After screenshot' panels — PNG, JPEG or WebP up to 20 MB — to load local previews.
Paste each transcript into 'Before transcript' and 'After transcript' from your own OCR, correct it against the preview, and tick 'I reviewed this transcript' on both.
Press 'Compare reviewed text' for the line and word change counts and similarity percentage, then 'Download counts' to save screenshot-ocr-change-counts-only.json.
A single re-wrapped paragraph shows as many line changes but few word changes, which tells you whether the wording or just the layout moved.
Case, blank lines, edge whitespace, repeated spaces and punctuation each have their own switch, so you decide what counts as a real difference before the diff runs.
The exportable JSON carries counts, options and truncation flags under the altftool.screenshot-ocr-change-counts.v1 schema — no transcript text, no filenames, no image data.
No, and that is by design. You paste text produced by your own device's OCR and tick a box confirming you reviewed it, because a raw recognition pass routinely misreads characters and those errors would be reported as content changes. The screenshot previews exist so you can proofread the transcript against the image.
Up to 100,000 characters per transcript, 800 lines and 2,000 words, with the first 200 individual line changes listed in detail. Anything beyond those bounds is truncated and the result flags it explicitly, so split large captures into focused sections rather than trusting a truncated comparison.
No. It means the reviewed transcripts match under the options you selected. Layout, spacing, colour, icons, cropping, images and any text that OCR missed or that is visually obscured are all outside this comparison — use a pixel-level diff if you need to catch visual changes.
Unchanged items divided by the larger of the two counts, at one decimal place — so if the before transcript has 40 lines, the after has 44, and 38 lines match, similarity is 38/44, or 86.4%. Line and word similarity are reported separately because they can diverge sharply on reflowed text.