POST /render/diff compares two publicly fetchable HTTPS JPEG or PNG screenshots. Pass the existing media objects verbatim; include both regions-v1 manifests to enable region-aware matching.
perceptual is the default and suppresses minor JPEG, antialiasing, and rendering noise. pixel reports exact pixel differences. Source widths must match; full-page heights may differ and excess height counts as added or removed evidence.
Visualizations
Every mode costs one credit. Generated visualization media uses the same direct storage contract and
X-Scrinly-Storage-URL header as screenshots.
Region-aware matching
When both manifests validate against the downloaded screenshot hashes and dimensions, Scrinly matches logical regions using kind, order, normalized position, height, and visual fingerprints. Matched regions are compared independently, so inserting one section does not make every later section appear changed. Results classify regions asunchanged, changed, added, or removed and report movement and size changes separately.
If either manifest is absent or invalid, the operation falls back to whole-image comparison with regionMatching.status:"unavailable". A manifest whose screenshot hash no longer matches its URL is rejected with diff_source_hash_mismatch; recapture assets that were overwritten before content-addressed storage was enabled.
The result includes recomputed source hashes, dimensions, changed, changedPixels, differenceRatio, similarity, comparison coverage, up to 128 consolidated changedBounds, region results, algorithm version, media, and credits. Source URLs are not echoed.
Sources allow at most three safe redirects, 20 MiB encoded bytes each, 24 megapixels each, 40 megapixels combined, and 32,768 pixels of height. Private/reserved destinations, embedded URL credentials, mismatched content, and invalid images fail with typed diff_* codes and a full refund.