DuoBolt
AI culling · on your machine

AI photo culling that shows its working

A wedding leaves you eight hundred frames and forty of them are the same moment. DuoBolt groups those frames, marks the best one, and tells you what it measured to get there — sharpness, exposure, whether the eyes are open. Then it stops.

Free to use · One-time purchase from $34.99 · 2 seats · 30-day refund
v2.1.0 · macOS & Windows · Docs
DuoBolt's AI Curation workspace: groups bucketed under Copies to remove, the open group showing its best shot beside two frames marked as copies, and a Why this one panel giving face sharpness, face exposure and resolution against the shot it was compared with

The two things nobody trusts about culling AI

That it will quietly bin a keeper, and that it wants your whole wedding on somebody's server first. DuoBolt is built against both. It selects rather than deletes , and every selection is a comparison you can read and reverse in one click — the panel names the runner-up and the margin, not a confidence percentage. And the models are in the app, so a 900-frame shoot never leaves the drive it is sitting on, never waits on an upload, and never needs an account.

It names what it measured

Sharpness, exposure, resolution and format fidelity on every group, plus an aesthetic score on the 1–10 scale photo panels use. Where DuoBolt finds faces, eyes open and smile join in — and the sharpness and exposure it reports become the face's, not the frame's.

Only the measures that mattered

A group lists the factors that actually separated its shots and stays quiet about the rest. Two frames at the same resolution do not get a resolution row just to fill the panel out.

When it is not sure, it says so

Groups it can separate become copies to remove. Groups where the shots are genuinely different go to Needs your eye instead of into a selection, and groups with nothing to do say so and get out of the way. Copies that are identical byte for byte are not scored at all: one is kept by rule, and the card names the rule.

It marks. It never deletes.

The sweep selects; it does not act. Everything then goes through the same review screen the rest of DuoBolt uses — full paths, sizes and previews, and a choice between Trash, an Archive folder, a copy elsewhere, or deleting outright.

How a keeper gets chosen

Four stages between a folder of frames — or a Photos library, on a Mac — and a selection you can disagree with.

  1. Step

    The frames become groups

    Before anything is ranked, the shots of one moment are gathered — by what the picture shows, not by file name or timestamp. Bursts and retakes land together; a RAW and its exported JPEG land together separately.

  2. Step

    One group is analyzed at a time

    Models bundled with the app score each frame on your own machine. Nothing is queued to a server and there is no wait for an upload — the analysis runs at the speed of your CPU, and a group you have already reviewed stays scored.

  3. Step

    A best shot, and the reason

    One frame is marked best and the others are compared against it in plain terms — 3.5× sharper, better exposed, eyes more open, same 19.6 MP. The comparison always names the shot it was measured against.

  4. Step

    You sweep, review, and decide

    Select the detected copies across every analyzed group in one action, or work group by group and disagree wherever you like. Then the ordinary review screen, and nothing has moved until you say so.

AI photo culling FAQ

Does DuoBolt delete photos automatically?
No, and it cannot. The engine ranks and explains; the app renders and you decide. AI Curation selects the frames it believes are copies of a shot you are keeping — selecting is all it does. Removing anything means going through the review screen, which shows full paths, sizes and previews and offers Trash, Archive, Copy or permanent delete.
Do my photos get uploaded for analysis?
No. The models ship inside the app and run in-process on your machine. There is no account, no library upload and no queue. The only network calls DuoBolt makes are the license check and the update check.
What does it actually measure?
Sharpness, exposure, resolution in megapixels, format fidelity, and an aesthetic score on the 1–10 scale photo panels use. When faces are found it adds eyes open and smile, and measures sharpness and exposure on the face rather than the whole frame. Each group shows only the measures that actually separated its shots. The seven measures, with the scale each one uses →
Does it work on RAW files?
Yes — the RAW files themselves, so the originals get culled too instead of piling up behind the exports. Canon, Nikon, Sony, Fujifilm, Olympus, Panasonic, Leica, Pentax, Samsung, Sigma and Adobe DNG are all read on macOS and Windows. Worth knowing how: DuoBolt reads the full-color preview your camera writes inside the RAW rather than developing the sensor data itself, which is what keeps it fast and free of a RAW-converter dependency. That preview is what the grouping and the measures see, so treat the result as a first pass over the take — it is not a check of critical focus at 100%.
What if the shots in a group are genuinely different?
Then it says so. Those groups go to Needs your eye rather than into a selection, and groups where every frame is distinct are reported as having nothing to do. A tool that always produces a winner eventually throws away a keeper. How a group is built in the first place →
Is it a subscription?
No. Culling tools in this category run about $100–250 a year. DuoBolt Pro is $64.99 once, for two machines, and includes every 2.x update. Every download starts with 7 days of Pro, and below Pro the first 8 groups of a scan stay whole with AI Curation working inside them.