The system checks every photo for sharpness, exposure, and colour fringing (chromatic aberration), then splits them into two batches: the ones worth a second look come first, one at a time, so you're not clicking through photos that are already fine. Once you're through those, you can optionally double-check the rest too, if you want the extra reassurance.
Stage 2 — more automated
Photos taken in quick bursts are grouped together automatically. The strongest frame in each group is pre-selected to keep, with the rest set aside — you can still adjust any group before moving on.
Stage 3 — a closer AI pass
What's left is checked for genuine duplication across the whole shoot, and anything redundant gets flagged. You dive in a bit further here, keeping only what's actually diverse and impactful.
Your memory card is never touched. Loupe only ever works with copies of files already on your computer — it's best used before you import anything, straight off the card (or a copy of it), so the filtering happens first. Once you're done, the whittled-down set downloads as a zip, ready to import into Lightroom or whatever you use. After that, format, back up, or reuse the card however you'd normally do.
This only works reliably on original camera files — not photos you've already edited or exported (e.g. via Snapseed, Lightroom exports). Editing tools often rewrite or strip the metadata this tool relies on, so duplicate detection especially won't be accurate on already-processed images.
What this tool is for
This is a pre-import sift — run it on your original camera files, straight off the card, before you bring anything into Lightroom or your cataloguing software. It's built to help you decide what's worth importing out of a large, repetitive shoot (bursts, wildlife, sport). It is not designed to re-sort photos you've already edited, exported, or catalogued — run it on those and the checks below (especially duplicate detection) won't work reliably.
Round 1 — technical triage
Fast pass. Upload your shoot (JPEG, or RAW — CR2/CR3, NEF, ARW, RW2, RAF, DNG all work). Photos are checked for focus, exposure, and colour fringing, then split into two batches — flagged ones first, so you're not clicking through every single photo.
Olympus ORF isn't supported yet — the only preview data that format exposes is too low-resolution to judge reliably (confirmed on real files, not a guess). ORF uploads will be rejected with a clear message rather than run through anyway.
Based on the RAW preview, not the actual file — if you do noise reduction in post (e.g. Lightroom's Denoise AI), a noisy-looking high-ISO shot may still turn into a clean keeper. Off by default for exactly that reason.
Runs a small face-detection model in your browser (first use downloads a few MB, cached after that — nothing is sent anywhere). Only relevant if people are actually in frame.
Drop your photos here, or click to choose
Nothing uploads anywhere — this all runs in your browser.
0 photos loaded
0 keep0 reject0 left
01 / 00—
Green = no issue found (e.g. sharp) · Red = worth a second look before you decide
Reject X
Keep K
Round 1 complete.
Flagged photos reviewed.
Round 2 — bursts & redundancy
Near-identical frames are grouped together. The strongest one in each group is pre-selected — click any thumbnail to add or remove it. Keep as many or as few as you want from each group.