Managing face clusters

When face recognition is on, Photogala automatically groups photos by person – each detected person becomes a cluster. As a moderator you organise these clusters: you give people a name, merge duplicate clusters, split wrongly grouped ones, hide irrelevant clusters and set a preview image. Well-kept clusters make the guest filter by people accurate.

When does the tab appear at all?

The Face Clusters tab only appears when your plan includes AI face processing and face recognition is switched on for the event. The host controls both under AI processing and Settings: Features. If the tab is missing, one of the two is off.

Reaching the cluster manager

Moderation lives in two places – with the same interface. Take the route your role gives you:

1. As a host or admin: admin sidebar → Gallery group → Review & Reports. Face Clusters is also a sub-entry you can click directly.

   As a moderator: profile menu top right → Moderation.

2. At the top, choose the Face Clusters tab.

Phone

Computer

The cluster manager: search, the filter for unlabeled clusters, the counters and the clusters as tiles.

Understanding the cluster overview

The top row controls the display:

1 the Search by name or ID field – jumps straight to people you already named, 2 the Only unlabeled checkbox – hides everything that already has a name. Together they let you work through the open clusters without losing track.

• Search by name or ID – filters the clusters live by name or cluster ID.
• Only unlabeled – hides every already-named cluster so you can focus on the open ones.
• Total clusters and Showing – how many clusters exist and how many are visible after filtering.

Each cluster tile shows:

• a preview face and the number of faces in the cluster;
• a selection checkbox for merging (see below);
• the name – Unlabeled at first – and below it the shortened cluster ID;
• a Hide or Show button. Hidden clusters carry a red Hidden badge and a dashed border.

Naming a person

A name makes the cluster clear for you and for the guest filter. There are two ways to do it.

Way A – right on the tile (fast)

The fastest route when working through many clusters in a row – you don’t open a dialog at all.

1. On the tile, click the name row (tag icon; on unnamed clusters it reads Unlabeled). It turns into a Person Name (optional) input field.

2. Type the name and confirm with Enter or with OK. Cancel discards the change.

3. A toast Label updated confirms it; the name appears on the tile right away.

Phone

Computer

Way A: clicking Unlabeled turns the name row into the Person Name (optional) field, with Cancel and OK below it. The other tiles stay untouched.

Way B – in the detail dialog

1. Open the cluster by clicking the preview face. The Cluster details dialog opens.

2. At the top, click Unlabeled - Click to name (pencil icon), enter the name in the Enter person name field and confirm with OK.

Tip

Name the most important people first (the couple, family, best man/maid of honour). Then use the Only unlabeled filter together with Way A to work through the remaining clusters in one go.

Merging clusters

When the AI detects the same person as several separate clusters (different lighting, glasses, angle), merge them.

1. Tick the checkbox on at least two tiles of the same person. A bar appears at the top with the number of selected clusters and a Merge button. With fewer than two selected, a toast says “Select at least 2 clusters to merge”.

   Phone

   

   Computer

   

   Two tiles are ticked – they get a coloured ring and a check mark. The bar above counts 2 selected and offers Clear and Merge.

2. Click Merge. The Merge Face Clusters dialog opens and names how many clusters are affected. You can optionally set a Person Name (optional) for the result right away.

   Phone

   

   Computer

   

   The dialog names the count – 2 clusters here – and offers the Person Name (optional) field. Its note calls the source clusters *hidden*; in fact they vanish from the list entirely (see the warning below).

3. Confirm with Merge.

Merging is a one-way street

Merging creates a new cluster that all the faces end up in. The source clusters disappear from the list afterwards – they come back neither normally nor as Hidden, and their old cluster IDs are gone for good.

There is no undo. Split does not restore the original state either: splitting creates yet another new, third cluster holding the selected faces, while the rest stays in the merged cluster. So you can separate two accidentally combined people again – but the result is a new split, not the old state.

That’s why it pays to check before confirming that it really is the same person.

Splitting a cluster

If the AI put two people into one cluster, extract the wrong faces into a new cluster.

The Cluster details dialog: pick faces (with a confidence percentage) and extract them into a new cluster; the star sets the preview image.

1. Open the cluster. The Cluster details dialog reads “Select faces to extract into a new cluster”. Every face carries a confidence percentage – how sure the match is. Low values (red or orange) are good candidates to sort out.

2. Select the faces that don’t belong to this person. Select All and Clear help; the counter shows Selected: N / total.

3. Optionally enter a New Cluster Label (optional) and click Split at the bottom. The chosen faces form their own cluster.

At least one face has to remain

You cannot extract every face of a cluster. The attempt is rejected with the message “Cannot split all faces. At least one must remain.” In practice that means Select All followed straight by Split never works – deselect at least one face.

The other way round applies too: with nothing selected nothing happens either, and the page says “Please select faces to split”.

Example

The AI picked up Lena twice: once with glasses (18 faces), once without (7). You tick both tiles, click Merge, set Person Name (optional) to “Lena” and confirm – afterwards a single tile “Lena” with 25 faces sits in the list and the two old ones are gone. In the Cluster details dialog you notice three faces at around 40 % confidence that actually show her sister. You select exactly those three (the counter reads Selected: 3 / 25), enter “Mira” as the label and click Split. What remains is “Lena” with 22 faces and a new cluster “Mira” with 3.

Setting a preview image

In the Cluster details dialog every face carries a star icon with the tooltip Set as preview. A click makes that photo the cluster’s preview image – handy when the automatic preview is blurry or not descriptive. The currently set face is marked Current preview, and a toast Preview image updated confirms the change.

Refreshing the list in the dialog

At the top right of the Cluster details dialog there is a Refresh button (circular arrow). It reloads the faces of this cluster – useful when photos are still being processed in the background, or when you just split something and want to see the new state.

Hiding and showing clusters

Not every cluster is relevant (background guests, blurry faces, strangers).

• Hide removes the cluster from the people selection in the guest filter without deleting any photos. The cluster stays here with a Hidden badge. A toast confirms: “Cluster is now hidden”.
• Show makes a hidden cluster available again (“Cluster is now visible”).

Note

Hiding only affects the people selection in the filter – the photos themselves stay in the gallery untouched. To hide photos, use the Media tab under Approving & moderating photos.

Phone

Computer

What the clusters are for shows up here for guests: the people selection in the gallery filter.

Privacy

Face data is biometric, especially sensitive data. Guests have a right to erasure and can remove their selfie and face data themselves – see Your privacy. As the host you must enable face recognition deliberately and provide a contact for data-subject rights; background under AI processing.

Troubleshooting

• No “Face Clusters” tab → AI face processing not in your plan, or face recognition off for the event.
• No or few clusters → processing runs while photos are uploaded. New faces appear gradually; in the detail dialog use Refresh to reload.
• Merge does nothing → fewer than two clusters are selected. Tick at least two tiles.
• Split is rejected → you selected every face. At least one has to stay in the original cluster.
• A cluster vanished after merging → that is intended: the source clusters are absorbed into the new cluster and do not reappear.
• Wrong person in a cluster → use Split to extract the foreign faces into a new cluster.

Related help

• Moderator: Approving & moderating photos · For moderators
• Guests: Finding photos of you (face search) · Your privacy
• Host: AI processing (face recognition, GDPR)