# How to find and remove duplicate photos

> Use AI > Identical copies for byte-for-byte duplicates and AI > Duplicates for resized or re-saved copies. Nothing is deleted until you decide what goes.

Source: https://firelet.media/help/find-and-remove-duplicate-photos
Published: 2026-09-15 · Updated: 2026-09-15
Applies to: Windows, macOS, Linux

To find duplicate photos in firelet.media Gallery, open the **AI** panel ([[Alt]]+[[3]]). **Identical copies** finds byte-for-byte copies saved under other names or folders, and **Duplicates** finds resized or re-saved copies. Neither tool deletes anything. They only flag the extras as rejected, so you can review them and move them to the Recycle Bin yourself.

## Two kinds of duplicate: identical files and near-duplicates

Duplicates pile up in two ways, and the gallery has a tool for each:

| | Identical copies | Duplicates |
|---|---|---|
| Finds | The very same file, byte for byte, under another name or in another folder | Copies that were resized, re-compressed or lightly edited, which a checksum would miss |
| How it checks | File size, then the first 64 KiB, then a full SHA-256 checksum | How alike the pictures look, from the AI index |
| Needs | Nothing extra | The AI model and **Index library** |
| Then | **Keep one of each, reject the rest** | You review each pair and reject the copy you don't want |

Online-only cloud files are never downloaded just to be compared.

## Steps: find identical copies

1. Press [[Alt]]+[[3]] to open the **AI** panel.
2. Expand **Identical copies**.
3. Click **Find identical copies**.
4. Look through the groups it lists. Each group is one picture that exists more than once.

![The Identical copies section after Find identical copies, showing one group and the Keep one of each, reject the rest button](https://firelet.media/help/media/shot-ai-identical-copies.webp "Each group is the same file saved in more than one place")

## Keep one of each, reject the rest

Click **Keep one of each, reject the rest** to tidy every group at once. In the app's words: "The copy kept takes the others' stars, keywords and albums; the rest are flagged, not deleted".

The gallery chooses which copy to keep in this order:

- the copy with the **highest star rating**;
- then the one with **more keywords and albums**, the copy you have been working with;
- then one **not named like a copy**, such as "Copy of…", "- Copy", "(1)" or a folder called "Backup";
- then the one with the **shorter path**, nearer the top of its folder.

If any copy was marked as a pick, the kept copy is marked as a pick too. Nothing leaves your disk at this stage.

## Steps: scan for duplicates (near-identical)

Near-duplicates need the AI model and index. If you haven't set them up, see [how to index your library for AI search](https://firelet.media/help/index-library-for-ai-search).

1. In the **AI** panel, expand **Duplicates**.
2. Choose a threshold (see the table below). **Near-identical (0.97)** is the default.
3. Click **Scan for duplicates**.
4. Click a pair in the list, shown as "{x}% match · same folder" or similar, to open it in the viewer.
5. Press [[X]] on the copy you don't want to keep. That marks it as rejected.

![The Duplicates section listing pairs with their match percentage, and the grid filtered with the chip duplicates ≥ 0.97](https://firelet.media/help/media/shot-ai-duplicates.webp "Click a pair to compare it in the viewer")

While the scan's result is shown, the toolbar chip reads, for example, "duplicates ≥ 0.97". Click its **✕** to return to the whole library. To look at two copies side by side, select both tiles and press [[C]] for **Compare**.

## Choosing the threshold

| Threshold | How strict |
|---|---|
| Identical (0.99) | Strictest: only pairs that look practically the same |
| Near-identical (0.97), default | Catches resized and re-compressed copies |
| Very similar (0.95) | Looser: more pairs, including lightly edited copies |
| Similar (0.90) | Loosest: also pictures that are merely alike |

Start with the default. Lower thresholds find more, but they also list pictures that are merely alike, such as two frames from the same burst. For those, [pick the best shots with the quality check and bursts](https://firelet.media/help/pick-best-photos-culling) is the better tool.

## Reviewing rejects and moving them to the Recycle Bin

When you're happy with what has been rejected, remove the extras in one go.

1. Press [[Alt]]+[[2]] to open **Filters** and expand **Rating & labels**.
2. Set the flag drop-down to **Rejected**. Only rejected pictures remain in the grid.
3. Look through them once more. To keep one after all, select it and press [[U]] to clear its flag.
4. Press [[Ctrl]]+[[A]] to select them all.
5. Press [[Del]], or right-click and choose **Move to Recycle Bin**, and confirm.

The files go to the Recycle Bin, along with any XMP sidecar, so you can still restore them from there. More about flags is in [how to rate, pick, reject and colour-label photos](https://firelet.media/help/ratings-flags-colour-labels).

> [!WARNING]
> Network drives and some USB drives have no Recycle Bin. On those, Windows deletes files permanently, so check the selection carefully before you confirm.

> [!TIP]
> For a longer, step-by-step clean-up plan, read [too many duplicate photos? How to find and remove them safely](https://firelet.media/blog/too-many-duplicate-photos).

## Questions

### Will it delete my photos automatically?

No. The duplicate tools only list and flag. Keep one of each, reject the rest marks the extras as rejected; you delete them yourself later, and deleting sends them to the Recycle Bin.

### What is the difference between Identical copies and Duplicates?

Identical copies are byte-for-byte the same file under another name or folder. Duplicates uses the AI index to find re-encoded, resized or slightly edited copies.

### Which copy is kept?

The one you rated highest, then the one with more keywords and albums, then one not named like a copy, then the one with the shorter path. It takes the others' stars, keywords and albums; the rest are flagged, not deleted.