Skip to content
DuoBolt

AI Curation

In a group of twelve burst frames, the question is which one to keep. AI Curation measures the frames and says which ones are the same shot written twice; the choice between distinct shots stays with you.

The AI Curation screen with a three-photo group open. The kept shot carries a BEST SHOT badge, the two others are selected and one is labeled COPY, and the WHY THIS ONE card compares the kept shot with the runner-up: 1.2× sharper, slightly better exposed, same resolution at 19.6 MP. The Recommendation pane reads: 1 detected copy can be selected safely.

Opening it

Section titled “Opening it”
  • AI Curation in the Similars results header, for the whole scan.
  • Review with AI on a single group — the button on the tile, or its right-click menu.

Analysis runs group by group in the background and can be paused (Pause analysis) and resumed (Continue analysis) at any point. The progress line reads Analyzing 14 / 32 groups.

What it decides

Section titled “What it decides”

Every group ends up in one of three states:

Outcome Meaning
All distinct “Every shot here is distinct — nothing to remove.”
{n} copies Shots DuoBolt is confident are the same frame stored again
Needs your eye The top candidates are too close to separate

The header sums the scan up in one line — 9 copies of 4 shots · 214.8 MiB · 2 groups need your eye.

The groups come in the order the results have them, whatever sort you chose there, so the list you work here is the list you left.

Best shot, copies, distinct shots

Section titled “Best shot, copies, distinct shots”

Inside a group each photo gets one of three verdicts:

  • Kept. The best of the 7 shots in this group. — badged BEST SHOT.
  • A copy of the kept shot. Same scene, written again — nothing the kept shot lacks.
  • A distinct shot. A different capture — it stays unless you decide otherwise.

A group of copies that are identical byte for byte skips the scoring altogether — there is nothing to judge between them. One is kept by rule and badged KEPT rather than BEST SHOT, and the card names the rule: Identical copies — kept: first import, 16 Jun 2023. When no rule tells them apart — the same import, none edited, none a favorite — any one is as good as another: the first is kept, the rest are selected, and the card says so: Identical copies — nothing tells them apart, so any one can go.

AI Curation over a folder scan, Painted_Romanian_Easter_Eggs (1).jpg selected in the Groups list — 2 files · Identical, 1 copy. The stage shows the kept photo of painted eggs with a KEPT badge, two thumbnails under it labeled KEPT and COPY with the copy selected, the file line reading 1280 × 942 · 442 KiB, and the Why this one card: Kept. One of identical copies, kept by rule rather than by eye. Identical copies — nothing tells them apart, so any one can go. Compared with Painted_Romanian_Easter_Eggs.jpg. The header reads 19 copies of 18 shots · 48.3 MiB · 8 groups need your eye.

That third one is the important category. A burst of seven different expressions is seven distinct shots, and DuoBolt will tell you it found nothing to remove rather than pick a favorite for you.

Why this one

Section titled “Why this one”

Select any two photos and the WHY THIS ONE card compares them measure by measure, naming each one:

Measure What it is
Sharpness — or Face sharpness where faces were found Detail contrast. It has no absolute scale — only one shot against another.
Exposure / Face exposure How much of the frame is neither crushed to black nor blown out. 0 to 1.
Resolution (MP) Pixel count, in megapixels.
Format fidelity How much the file’s format keeps of the original. 0 to 1.
Eyes open How open the eyes are, across the faces found. 0 to 1.
Smile How much of a smile the faces show. 0 to 1.
Aesthetic A trained model’s opinion of the picture, on the 1–10 scale photo panels use.

Each row reads as a verdict rather than a number you have to interpret — 2.3× sharper, Slightly better exposed, Bigger, 24.2 MP, Eyes more open. The raw value is there beside it.

Two rules decide what the card shows:

  • A group only lists the measures that separated its shots. Eyes open and Smile appear only where DuoBolt found faces.
  • When nothing separates them, it says so: “Too close to call on every measure — your eye decides.”

Acting on it

Section titled “Acting on it”

Select the copies marks the detected copies across the group, and the tooltip is exact about what that means: “Marks them — you review before anything happens.” Distinct shots stay untouched.

Press S to select or unselect the focused group’s copies without leaving the keyboard. Everything then goes through Review Selection like any other batch.

What runs, and where

Section titled “What runs, and where”

Five small models, shipped inside the app. They run offline, and none is downloaded on first use.

Faces

A detector finds faces and their landmarks; a second model reads whether the eyes are open; a face-mesh model reads the mouth. Together they produce Face sharpness, Face exposure, Eyes open and Smile.

The picture as a whole

An aesthetic model trained on human photo ratings supplies Aesthetic on the 1–10 scale. Two embedding models decide whether two frames really depict the same thing, which is what separates a copy from a distinct shot.

Sharpness, exposure, resolution and format fidelity are measured directly from the image — no model involved.

Tiers

Section titled “Tiers”

AI Curation itself is not a paid feature: it runs on the groups you can see. What the tier changes is how many groups that is — below Pro a scan shows its first 8 groups, and the curation list says 9 more groups · Upgrade to Pro at the end.

Before this

How grouping works — what has to be true before two photos end up in the same group at all.