Supported files

.jpg .jpeg .png .gif .webp .bmp .tiff .heic .mov .mp4 .dng

What works with what

Type	Scan & dedupe	EXIF	Search	Faces	Near-duplicate
jpg, jpeg, tiff	yes	yes	yes	yes	yes
png, gif, webp, bmp	yes	no	yes	yes	yes
heic	yes	yes	macOS only	macOS only	no
mov, mp4	yes	no	macOS only	no	macOS only
dng	yes	yes	no	no	no

Everything is scanned, hashed and exactly de-duplicated regardless. The gaps above are about what can be decoded, not what is recorded.

.dng is skipped for search because no DNG decoder is available. Its EXIF metadata is still read.

See platform support for why HEIC and video need macOS.

File type detection

Types are identified by the file’s actual leading bytes, not its name, so a mislabeled file is still handled correctly. This costs nothing extra, since the bytes are already being read to hash the file.

A .png that is really a JPEG is treated as a JPEG. A .dng reports as TIFF, which it genuinely is, so DNG is excluded from search by extension rather than by type.

When the bytes match nothing recognised, the file is recorded as application/octet-stream rather than left empty. That is a real answer, not a failure: it records that the question was asked and settled, which is what stops scan --retry-incomplete reopening the same file on every run. Such a file is still processed, falling back to its extension.

Files scanned before type detection existed have no recorded type until you re-scan; those also fall back to the extension.

EXIF fields

EXIF is read from jpg, jpeg, tiff, heic and dng. Fields are empty when the file carries no EXIF data.

Field	Notes
Date taken	DateTimeOriginal, camera-local with no timezone
GPS latitude	Decimal degrees, negative is South
GPS longitude	Decimal degrees, negative is West
Width, height	Pixel dimensions

Dates of 0000-00-00, which cameras with an unset clock produce, are discarded rather than stored.

Near-duplicate detection for video

For .mov and .mp4, the fingerprint comes from a single poster frame, not the video content. So it catches re-encodes and trims that keep the opening frame, but not a trim that cuts it.

That output is review-only and never reaches videre dedupe’s pipeable output, so a false match costs you a noisy group in the report, never a wrong deletion.