Help Series

This page is generated from docs/aide/en/: it is also the admin’s online help (“Help” menu). Edit it in the repository, not here.

Series group the photos together (Street, Portrait…). They form the site's index and serve as a filter. Each series is a plate: its first three photos, its name and its number of photos.

• Create: ＋ New series at the top of the page puts the cursor in the creation card. Type the name, then Enter (or Create); Esc clears the input. The new series takes first place, with an address derived from the name.

• See its photos: Photos › opens the list of photos filtered on this series.

• Rename: Rename opens the name in the card, then Save. Under the name, “Name in English and German”: the version shown on the English and German pages of the site (optional; empty: the French name). The address (/serie/<adresse>) is tucked away under “Series address (advanced)”: changing it breaks links already shared.

• Order: this is the order of the series index on the site.

  • with the mouse: grab the ⠿ Move handle and drag the card to its new place;
  • with the keyboard: Tab to the handle, then ↑ / ← (move forward one place) or ↓ / → (move back one place);
  • everywhere, including on a phone: ⋯ menu, Move up one place or Move down one place.

  The order is saved straight away, and a message gives the new place.

• Delete: ⋯ menu, Delete…. A window opens.

  • Empty series: confirm.
  • Series with photos: they are never deleted. Choose the series to move them to, then Move the n photos and delete.

To file photos in a series, the quickest way is the multiple selection on the Photos page. On the site, a series with no published photo is greyed out and cannot be clicked.

Suggested series

At the bottom of the Series page, Suggest a series helps you file incoming photos. It is off by default.

• Switch on: Switch on suggestions. The analysis model (about 90 MB, an open model called CLIP) is downloaded once into the data folder. Existing photos are then analysed in the background, with a progress bar; each imported photo is analysed straight away.
• What it suggests: on import, each photo gets the series chosen in the form. Until you have chosen it yourself, the server compares the photo with the photos already filed (yours and the other photographer’s: likeness is worked out per series). If another series looks clearly more like it, that series is suggested, and two are suggested when they are almost as close. A series is only suggested once it holds 3 filed photos. A published photo never gets a suggestion.
• Where to see it: a ✦ star and a dotted outline mark it, without relying on colour.
  • on the thumbnail in the contact sheet, and with the ✦ Suggested series filter;
  • in the inspector: Move to “…” applies it at once, Keep “…” confirms the current series;
  • on the photo’s details: Choose fills in the Series field, and Save confirms; the small thumbnails show the photos of the series that look most like it;
  • on an outing page, under each photo;
  • in Today, a line “n photos with a suggested series”.
• One photo at a time: nothing is ever filed without your click, and there is no “accept all”. As soon as you choose a series (suggested or not, or by dragging the photo onto a series), the suggestion disappears and the photo becomes an example for the next ones.
• Switch off: Switch off and delete removes the model and all image fingerprints.

The analysis runs on your server: no image is sent anywhere else. A photo’s fingerprint (a series of numbers) is only used to compare images with each other. It cannot recognise a person, and it is deleted with the photo.

Describing photos

Below Suggest a series, the Describe the photos panel suggests a short, factual sentence for each photo (“a red lighthouse on a small island”), shown on the photo’s page under the caption. It is off by default.

• Turn on: choose the model, Light (Florence-2 base, the default: about 270 MB to download, 2 GB of memory while computing, about 10 seconds per photo) or Precise (Florence-2 large: about 850 MB, 4 GB, a little more detailed). Precise needs the site to have 6 GB: add SITE_MEM_LIMIT=6g to the stack’s .env, then Compose Up. The panel shows the available memory and refuses Precise if it is not enough. Also choose the night hours (1 am to 6 am by default, Paris time), then Turn on descriptions. The model downloads once, at the first computation.
• When: photos without a description are described during the night hours, one at a time, on a single processor core, so as not to slow the rest of the server. The model only takes memory while computing.
• On the photo’s page: Describe this photo puts it first, right away (about a minute). Then Use as caption copies the sentence into the Caption field, to review, rewrite and save. Describe again recomputes it. Nothing is ever copied by itself.
• Language: the model writes in English. If the site has a translator (LibreTranslate), the sentence is translated into French and German; otherwise it stays in English.
• Turn off: Turn off and delete the model removes the model from disk; existing descriptions remain.

The analysis happens on your server: no image is sent elsewhere. The model describes what it sees, without ever naming people.