Organise

Live camera

In the store, at the loans shelf or while unpacking, you stand in front of an object with a label and want to see its record. The live camera reads QR codes, barcodes and inventory numbers, printed or handwritten, for this and, while the camera runs, shows the matching objects of the project. It also reads a title on a wall label and searches it in a field of your choice. Objects without a label it recognises by their look, once the project’s images are indexed. You can type a number as well. Opened with the camera icon in the top right corner, or on a phone with Camera in the bottom bar outside a project, the camera searches all projects you may read.

Live camera in the project Sammlung. On the left the camera picture with a painting and three labels, the reading frame in the middle, below it Reading codes and numbers and at the bottom the button Read number, at the top the counter of the work basket. On the right at the top Search in with All fields with unique values and the field Type a number. Under In view the codes read: Portrait of Dora Wheeler with the inventory number 1921.1239 and the note Barcode, Stag at Sharkey’s with 1922.1133 and the note QR, and Code read: ALT-00417, for which there is no object. Below them the history with the same three entries.

Opening the camera

In the Filter tab of the sidebar, the camera icon stands next to the full-text search field. It opens the camera; the first time, the browser asks whether the page may use it. On phones and tablets the rear camera starts.

On a phone the camera picture fills the screen and the hits sit in a sheet below it. Dragging the sheet’s handle upwards makes it larger. On a computer a large dialog opens, with the hits to the right of the picture. Two buttons at the bottom left of the picture switch the light on and change the camera, where the device offers either. With two cameras the button switches straight to the other one; with more it opens a list of their names, with the one in use marked.

The camera picture stays in the browser. Moliri receives the text read from a code or a number and, for matching by look, a small still every two to three seconds, which the server compares and does not store (see Recognising objects by their look ). Read number passes on the part inside the frame, when your text recognition place is not the browser (see Read number ). The camera stops when you close the view, open an object or switch to another browser tab.

Searching all projects

When you do not know which project an object is in, open the camera with the camera icon in the top right corner. It is there on every page, on the project overview as well as in a project. On a phone Camera in the bottom bar opens it, across all projects outside a project. Where the project’s name usually stands, it then says All my projects, and codes, numbers and the look find objects of every project you may read.

Live camera from the main menu, All my projects and the counter of the work basket at the top right. On the left the camera picture with a painting and three labels. On the right SKH-0118 in the field Type a number, under In view Portrait of Dora Wheeler with 1921.1239 and Stag at Sharkey’s with 1922.1133, each in Sonderausstellung 2026, Sammlung and Highlights (3D), and Rehe im Walde II with SKH-0118 in Leihanfrage Moderne 2026. In the history also Code read: ALT-00417, for which there is no object in any of your projects.
  • An object in several projects appears once, and the row names all of these projects you may read below its title
  • Tapping the row opens the object in the one of these projects you used last, otherwise in the first by name
  • In every project its own fields with a uniqueness condition count as numbers. A field that is unique in one project only finds the objects of that project by its number
  • Objects of projects you may not read, objects in the trash of a project and numbers in fields you may not see do not appear
  • The title of a row is the first text field your cards show in the object’s project
  • By their look the camera recognises objects of all projects you may read in which the image search is on and whose images are indexed. An object appears only in the projects in which the recognised image is linked to it. The Image search icon is missing here: indexing happens in the camera of a project

What the camera reads

It reads QR codes, DataMatrix and the common barcodes (Code 128, Code 39, EAN-13, EAN-8, ITF), anywhere in the picture, and inventory numbers inside the frame in the middle (see Reading inventory numbers ). A code leads to the object like this:

The code holdsResult
the address of an object of this Moliri instance, such as …/p/900/o/5 or its permanent address …/id/object/5That object, also when it lies in another project you may read. The row then names the project
a numberThe objects of the project that carry this number
anything elseCode read: … with the text read and a note that no object matches it

Numbers are the fields that carry a uniqueness condition in the project (unique in the project or unique in the database, see Fields ), such as the inventory number. Like any other field, the RegNr only counts when it carries a uniqueness condition. You also find every object by its ID: #123, typed or read as a code, leads to object 123. A match without a matching number shows its ID in its place, for example #123.

The comparison ignores spaces, full stops, slashes, hyphens, underscores, colons and upper and lower case. G 1923/44 therefore finds g1923.44. The letters O, l and I also count as the digits 0 and 1, because they are easily confused on labels. If a number of five or more characters differs in exactly one character, the hit appears with a tilde on its note (Barcode ~, Text ~). It is a suggestion, not a certain hit.

A code that matches no object stays visible, so a label from an earlier system is not passed over unnoticed. The icon on the right copies the text read, for instance to look for it in the full-text search or in the field Frühere Inventarnummern.

Reading inventory numbers

An inventory number without a code, printed on a label or written on a tag or on the object, is read as soon as it stands inside the frame. Hold the number completely inside the frame and close enough to fill about half of it. About once a second the browser’s text recognition reads that part of the picture and looks up every number in it like a typed one. The hits carry the note Text, with a tilde when one character differs. Words cut by the frame do not count, so that a half-read 1921.1 does not suggest other numbers. A number in several parts, such as G 1923/44 or Inv.-Nr. 1923.44, is found all the same.

Printed numbers are usually read at once, handwritten ones often only with a tilde or not at all. That is what Read number is for.

Live camera in the project Sammlung. In the frame a painting with a handwritten tag 1927.1984, at the bottom the button Read number with the place Browser. On the right at the top the text read, 1927.1984, in an input field with the note to correct or shorten it to search again. Under In view The Biglin Brothers Turning the Stake with 1927.1984 and the notes Text and Read. In the history below also Stag at Sharkey’s with 1922.1133 and the note Text, read earlier from a printed label.

Read number

The button Read number at the bottom of the camera picture stops the picture and reads the part inside the frame once more thoroughly, at your text recognition place from the profile (Text recognition ). The button shows which place that is:

PlaceWhat happens
BrowserThis browser’s text recognition reads the part in two ways, as a label and as scattered text. It does not leave the computer
Transcribe on this computerThe desktop app Transcribe reads it. It does not leave the computer
AI serviceAn image model of the AI account reads it, handwriting too. This can take up to a minute. The part goes to that service through the server
ServerThe server’s text recognition service reads it

The AI service and the server read only for people with write access to the project. Without it the browser reads. If your place cannot be reached right now, the browser reads as well.

While it reads, the picture stands still and a circle with the place turns in the list of hits. Then the camera runs on, and the top of the list shows what was read, such as 1927.1984 under Read (Browser), or No object of this project has this number. The hits carry the note Read. They stay until you press Read number again or close the line with the cross.

What was read stands in an input field. If the text recognition caught too much or misread something, change or shorten the text there: as soon as you pause, Moliri searches again with the changed text, and the hits with the note Read replace the earlier ones.

Reading a title or other text

A label often carries no number but a title. Above the field Type a number, Search in chooses where Read number and the field search:

Search inWhat is searched
All fields with unique values (default)The fields with a uniqueness condition, as described above. Besides the numbers, Read number also searches each line read as a whole, so a unique text field matches too
a text field, such as ObjekttitelThe text read or typed, in this field, by its words

With a text field the button says Read text. An object matches when all words of its field are in the text read, or all words read are in its field. A wall label with artist, title and date thus finds the object by its title alone. Case and punctuation do not count, a word may differ in one character, and the start of a word is enough, such as Portra for Portrait. When only some of the words match, at least two longer ones, the hit appears as a suggestion with a tilde. Words of fewer than three characters do not count. Moliri shows the ten best hits.

The project’s text and long text fields are offered. The choice applies to this project and is kept in this browser. The camera goes on reading codes and inventory numbers live. Search in is not offered in the camera from the main menu: there it searches the unique fields of each project.

Live camera in the project Sammlung. In the frame a wall label with Frederic Edwin Church, Twilight in the Wilderness and 1860 · Oil on canvas, at the bottom the button Read text with the place Browser. On the right at the top Search in with the field Objekttitel, below it the text read in an input field with the note to correct or shorten it to search again. Under In view Twilight in the Wilderness with the inventory number 1965.233 and the note Read.

Recognising objects by their look

The camera recognises an object without a label by its look. Every two to three seconds the sharpest picture of that time goes to the server, made small. An image model on your own server compares it with the images of the project’s objects. An object appears only when it is clearly more similar than the next most similar one, with the note Image and the similarity, for example Image 85 %. Blurred pictures and an unchanged picture are not sent. Below the frame it then says Reading codes and numbers · and the object by its look.

Flat objects such as paintings, prints and documents are recognised reliably. Three-dimensional objects seen from another angle than the catalogue photo are often not found. Then the label or the number helps.

Live camera in the project Sammlung. In the camera picture a painting lying at a slant, without a label, below the frame Reading codes and numbers · and the object by its look. On the right the hit The Large Plane Trees (Road Menders at Saint-Rémy) with the inventory number 1947.209 and the note Image 85 %, recognised by its look alone.

Switching the image search on and off

Administrators decide whether the camera searches by look:

  • Under Settings › Image search, Image search in all projects switches it on or off for the whole instance. At first it is off
  • In the camera of a project, the Image search icon at the top right opens a dialog. There In this project offers As in the settings, On and Off. The page Settings › Image search lists the projects that differ under Projects with their own setting; As in the settings lets a project follow the switch for all again
  • Also under Settings › Image search, Images decides which images go into the index: All images, Except these or Only these. With the last two a filter describes the images, with the same conditions as the file export : keywords, file types, file extensions, copyrights, copyright owners and creators, each with an operator. An image has to meet all conditions. An image that no longer belongs after its details changed leaves the index

When the image search is off in a project, the camera only looks for codes and numbers there, and its images do not go into the index.

Indexing the images

For the camera to recognise an object by its look, the image model must have looked at the project’s images once. Where the image search is on, this happens by itself:

  • New images go into the index within minutes: uploaded ones, ones attached to an object and ones whose details change so that they now belong
  • Every night in the server’s time window a run catches up with what is still missing: the images that were there when it was switched on, changed images and changed details

The Image search dialog at the top right of the camera shows where it stands:

  • It names how many objects with how many images the project has, how many of them are still to be indexed and about how long that takes, and how many images the filter keeps out
  • Index now starts the run at once instead of waiting for the night, Cancel run stops it

While a run is under way, the dialog shows what it is doing: whether it still waits in the queue, whether the server is loading the image model (the first run downloads it, which can take several minutes) or how many images it has checked and indexed so far, with a progress bar. Below it you see how long the run has been going and when it last made progress. When an attempt failed, the dialog names the error and when the next attempt starts; after three attempts the run counts as failed. When a step makes no progress for a long time, the dialog says so.

You can close the dialog: the run goes on on the server. Its progress also shows under Tasks & Uploads in the bar at the bottom of the project, where you can cancel it as well. It lists the runs you started; administrators see the runs of all projects.

Anyone with write access to the project can press Index now; only administrators switch the image search on and off. No image leaves the server for this. Every image is indexed once, even when it belongs to several objects or projects. The camera from the main menu searches all projects in which the image search is on.

The list of hits

The list beside the camera picture has two parts. That way it stays calm while you move the camera.

  • In view at the top shows up to three objects the camera recognised in the last few seconds. An entry keeps its place while the camera sees it and leaves this part about four seconds later. When all three places are taken, an entry only gives way to a clearly better hit: a certain hit by code or number, an object found in more ways, or one that is clearly more similar to the picture. The hits of the typed number and of Read number always get a place here and stay while they apply
  • History below shows every recognised object once, the most recent on top, with the count in the heading. When the camera sees an object again, only its notes change; the entry does not move. Nothing disappears from the history while the view is open. When you close the view, it is empty
  • Certain hits, objects whose address a code held or whose number matches exactly, have a green background under In view, a code without an object an orange one. Suggestions carry a tilde at their note
  • Each row shows picture, number and title and, on the right, where the hit comes from: QR, DataMatrix, Barcode, Text (a number read live), Read (Read number), Image with the similarity (recognised by its look) or Typed. When more than one way found the object, all notes stand together
  • Tapping the row opens the object. Back in the browser returns to the camera, with the history still there

Pinning

The pin on the right of each row puts the object into your work basket . It is the same pin as on the cards of the gallery: the object is then pinned in its projects too. A second click takes it out again. At the top of the camera picture a counter with a pin shows how many objects the basket holds.

So you can recognise newly arrived objects one by one with the phone, pin each of them and then have them all together at the PC: the basket shows them as soon as you open it there or reload the page. While the camera is open, it reads the basket again every three seconds. When you unpin an object at the PC, its pin in the camera goes a little later, and the counter drops.

Opening hits on the computer from your phone

When your phone follows a Moliri tab on the computer (Phone as companion ), Scan and open there opens the live camera on the phone. A tap on a hit then opens the object on the computer, not on the phone. The camera stays open, so you can point it at the next label straight away.

  • At the top of the picture, under the project’s name, stands the device the phone follows, for example Desktop · Chrome
  • After the tap the row says Sent to Desktop · Chrome … until the computer shows the object, then, in green, Shown on Desktop · Chrome. The object the computer shows at the moment is marked the same way, whoever opened it
  • The camera searches the project the tab on the computer shows when the camera opens. If it shows the project overview or a page without a project, it searches all your projects (Searching all projects ). If a hit opens another project on the computer, the camera keeps its choice; to change it, close the camera and open it again
  • The pin puts the object into the work basket as usual
  • If the tab on the computer is closed, the camera closes and the phone offers the other tabs

If you are typing in a field on the computer, opening waits until you leave it.

Typing a number

The field Type a number at the top of the list looks up a number like a code read, as soon as you pause typing. When Search in names a text field, it says Type a text from … and searches the text in that field (see Reading a title or other text ). Its hits stay until you change the input or Search in. This helps with labels without a code, with damaged codes and where no camera is available.

If the camera does not start

The view says in the camera picture what is wrong. The number field stays usable in every case.

MessageCause
The camera is not allowedThe page may not use the camera. Allow it in the browser’s settings for this site (usually the padlock in front of the address)
No camera foundThe device has no camera, or it is switched off
The camera is in use by another appA video call or the camera app holds it. If the device has other cameras, the switch button stays at the bottom left, so you can choose a free one
This browser offers no camera hereThe browser offers web pages no camera access
No camera icon next to the search field or in the main menuLikewise: the browser offers no camera access
A code is not readToo small, blurred or reflecting. Move closer, switch the light on or type the number
An inventory number is not readIt is not completely inside the frame, too small or handwritten. Hold the number completely inside the frame, move closer or press Read number
Below the frame it says Reading codes · loading text recognitionThe first time, the browser loads the text recognition. Codes are read already
Below the frame nothing is said about the lookThe image search is off in the project, for the camera from the main menu in all of your projects (Switching the image search on and off ). Codes and numbers are still read
Below the frame it says Image matching unavailableNo image of the project is indexed yet (Indexing the images ), for the camera from the main menu none in any of your projects. Or the server’s image service does not answer; the camera asks again after a while. Codes and numbers are still read
An object is not recognised by its lookThe project is not indexed yet, the filter under Settings › Image search keeps its image out of the index, the picture shows the object from another angle, or a similar object is too close. Hold the whole object steadily in the picture or use the label and the number
A title read finds nothingSearch in does not name the field with the title, or the text recognition misread too many words. Choose the field or correct the text read
An inventory number finds nothingThe field carries no uniqueness condition in the project, or the object is in the trash. With the camera from the main menu also: the object is only in projects you may not read

Further reading