Seba Exhibit Management: a guide

Getting in

This tool needs an admin key before it will do anything — it's what stops anyone else from deleting exhibits. There are two ways to provide it:

• A bookmarked link. If you were given a link that already ends in ?key=…, just use that — you won't see any prompt.
• The admin key prompt. If you open the page without the admin key, you will be asked for it. Enter it once and it's remembered for that browser tab, so you won't be asked again until you close the tab.

Don't have the admin key? Ask whoever manages the server. Only they can see or change it.

Getting oriented

A black bar reading Seba Manager sits at the very top, then a row of three fixed selections underneath it: Exhibits, where you'll spend most of your time, Venue, for settings that apply to the whole venue and the usage report, and Identify, for testing photo matching on its own (see Venue settings, Analytics and Test photo matching).

The Exhibits view has two buttons, + Add New Exhibit and Search (see Search exhibits), above a list of every exhibit that's been added, in alphabetical order, with a running total of exhibits and photos at the top. Click "+ Add New Exhibit" to create one, or click any exhibit in the list to open it. An open exhibit shows its name as a heading, a one-line summary (photos, description, latest photo date), and four tabs: Photos, Description, Rename and Delete. Opening an exhibit — whether by clicking it in the list or finishing Add a new exhibit — always lands on the Photos tab first, whatever tab you had open last time.

On a narrow screen (a phone), the list and the open exhibit share one column instead of sitting side by side: opening an exhibit replaces the list with its Photos/Description/Rename/Delete view, and a "‹ Exhibits" link at the top of that view brings the list back. Nothing is lost by going back this way — it just hides the view, the same exhibit is still open if you return to it.

Unsaved work is protected. If you have photos staged but not yet uploaded, or a description you've edited but not saved, and you then click Venue, Identify, Exhibits, + Add New Exhibit, or a different exhibit in the list, you'll be asked to confirm first — continuing discards that unsaved work. The "‹ Exhibits" back button on a narrow screen is the one exception: it never asks, because it never discards anything.

Every exhibit in the list has a number. An exhibit that's ready shows in normal colors, like 2. above. If it has no description or no photos, its number, its name and the line under it all turn red, like 1. above — red means the exhibit isn't complete and usable as yet. With no photos, a visitor's photo can never match it; with no description, a photo can match it but Seba won't have anything to say. The line under the name shows what's there (for example "0 photos" or "No description"), and hovering over a red row spells out what's missing. A red row goes back to normal as soon as both are in place.

The numbers are just each exhibit's place in the alphabetical list, so they shift when exhibits are added, renamed or deleted. Refer to exhibits by name, not by number.

When an exhibit is open, its name appears above the tabs as a heading, with a summary line underneath: how many photos it has, whether it has a description (and how long it is), and the date of its latest photo.

Add a new exhibit

An exhibit needs two things before it works for visitors: reference photos, and a description. Either can be entered first.

1. Click "+ Add New Exhibit" It's above the exhibit list on the left (on a narrow screen, it replaces the list with the form below).
2. Name itEnter the name in the "Exhibit name" box. Use whatever you'd naturally call the piece — this name becomes the exhibit's permanent identity, so it's worth getting close to right the first time. (You can still rename it later — see Rename an exhibit — that's safe and doesn't break anything.) As you type, a line underneath tells you whether the name is "✓ Available." or "✗ … already exists."
3. Click "Continue" The tool double-checks the name against the server, then opens the new exhibit on its Photos tab, with its name at the top. The exhibit isn't actually created until you upload its first photo or save its first description — until then the summary line reads "Not saved yet".
4. Add photosYou're already on the Photos tab — see Add more photos for the ways to do this, and what makes a good set before you do it for the first time.
5. Write the descriptionClick the Description tab: choose the language it's written in, write or paste the text, and click "Save description". This is what Seba actually reads from to introduce the exhibit and answer questions — write it the way you'd want a knowledgeable guide to talk about the piece. See what makes a good exhibit description for what to include.

An exhibit with photos but no description will show up as an error on the visitor's phone if they scan it. Add the description before you expect visitors to encounter it.

Open an existing exhibit

Click its name in the list on the left (or, on a narrow screen, in the list that's showing). Its photos and description load automatically — nothing to fetch or confirm — and it always opens on the Photos tab, whatever tab you had open for the previous exhibit.

If you have unsaved photos or description text for the exhibit you're currently on, you'll be asked to confirm before it's replaced — see Getting oriented.

Search exhibits

With more than a handful of exhibits, scrolling to find one by eye stops being practical. Click Search, next to + Add New Exhibit, to search every exhibit's description text on the server.

1. Click "Search" A dialog opens with a text box, an "Advanced (regular expression)" checkbox, and two buttons, Cancel and Start.
2. Type what you're looking for By default this is a plain text search — it looks for that exact text, ordinary punctuation included, anywhere in a description. Check Advanced only if you deliberately want to search with a regular expression (for example bronze|porcelain to match either word) — with it unchecked, characters like ., ( and ) are searched for literally, so typing something like "Q. (Qing)" finds that exact text rather than being misread as pattern syntax. Either way the search is not case-sensitive.
3. Click "Start" The dialog closes, the exhibit list empties while the server searches, and a "Matching: …" line appears above the list with a Cancel search button beneath it. (Clicking Cancel instead just closes the dialog without searching — whatever you typed is not remembered for next time, unlike Start.)
4. Review the results The list now shows only exhibits whose description matched. Click one exactly as you would in the full list. If nothing matched, the list says "No matches" instead.
5. Done searching? Click Cancel search, next to the "Matching: …" line, and every exhibit is listed again.

The search stays active until you clear it. Switching to Venue or Identify and back to Exhibits keeps showing the same search results — it does not quietly revert to the full list. Only clicking Cancel search brings every exhibit back. Opening Search again, even after that, prefills the box with the text (and the Advanced setting) you last used, so refining a search is quick.

View, enlarge and remove photos

Open the exhibit — it opens on the Photos tab already, and "Existing photos" is the first thing on it: what's already on file for the exhibit. The thumbnails are small square crops, so to see a whole photo properly, enlarge it.

Enlarge a photo

Click the small magnifier in the top-left corner of any thumbnail. The whole photo opens at full size, with its position in the set, the date it was added and its file name underneath. Use the ‹ and › buttons (or the ← and → keys) to step through the exhibit's photos, and close it with ✕, Esc, or a click on the dark background. Enlarging a photo never selects it for removal.

Remove photos

Click the photo itself (anywhere except the magnifier) to select it — a checkmark appears and the image dims slightly. Click again to deselect. Once at least one is selected, a "Delete N photos" button appears; click it, confirm, and they're removed.

Add more photos

Below "Existing photos" on the same tab is where you bring new ones in. There are four ways — use whichever is quickest for the photo you have:

1. Use the camera. The "Use camera" button sits at the top of this section — see below.
2. Choose files. Click the dashed box below it and pick one or more photos from your device.
3. Drag and drop. Drag photos from a folder onto the dashed box.
4. Paste a copied image. On any web page, right-click the picture and choose Copy image; then, anywhere on the Photos tab, press Ctrl+V (⌘+V on a Mac). No need to save it to a file first. Only JPG, PNG and WebP images can be pasted (Chrome usually hands over a copied image as PNG, whatever the page's original format was). The exhibit needs a name first. Choose Copy image, not Copy image address — an address is only a link, and you'll be told so.

Whichever way you add them, photos are only staged at first — they appear as thumbnails below the dashed box, with the "Upload photos" and "Clear" buttons underneath them. Nothing is saved until you click Upload photos; "Clear" throws the staged ones away instead. A staged photo can also be left out of that upload without removing it: click the photo itself to mark it excluded (its caption says so), and click again to include it. Click its magnifier to see it at full size first, the same as for a photo already on file. Switching to a different exhibit — or navigating away with unsaved changes and confirming — drops every staged photo, so one can never end up on the wrong exhibit. You can stage multiple photos from different sources at once.

Taking a photo with the camera

1. Click "Use camera" Your browser will ask permission the first time. The live picture appears with zoom controls underneath.
2. Zoom to frame the piece Drag the slider, use the − and + buttons, pinch on a touch screen, or scroll the mouse wheel over the picture. The label under the slider says which kind of zoom you have: camera zoom is the device's own zoom (full quality — available on some phones and browsers, mainly Chrome on Android), while digital zoom crops and enlarges the picture instead (fine for framing, but a little softer). What you see framed is exactly what gets saved.
3. Take the shot Tap the picture itself, or press "Take photo". The button is greyed out for a moment while the camera warms up, so you don't capture a frame that isn't ready yet.
4. Check it The photo appears as "Just captured", with a Cancel button to discard it. Take more if you like, then click "Upload photos" to save all the pictures you have added.

What happens automatically

• Exact duplicates are skipped, not saved twice. If you (or someone else) already added a photo, uploading it again just gets quietly skipped — you'll see it listed as skipped in the confirmation, not an error.
• Near-duplicates are flagged, not blocked. A photo that looks very similar to one already on file — including one filed under a different exhibit — gets a warning so you can double-check it landed on the right exhibit, but it's still saved either way. This is the main safety net against accidentally uploading a photo to the wrong exhibit.
• Files that aren't photos are skipped, with a message. Only JPEG, PNG, WebP and GIF pictures are accepted — the tool checks what's really inside the file, not just its name. Anything else (a PDF, an iPhone HEIC photo) is listed in the confirmation as skipped and not saved.

Update the description

Open the exhibit and click the Description tab. Choose the language the text is written in ("Written in"), write or paste the description, and click "Save description". A character count under the box shows how long it is, and the same figure appears in the exhibit's summary line and in the list on the left. This text is what Seba introduces the exhibit from and answers visitors' questions with, so it's worth keeping accurate — you can come back and revise it any time. For advice on what to write, see What makes a good exhibit description.

Rename an exhibit

Open the exhibit, go to the Rename tab, type the new name, and click Rename. Only the display name changes — its photos, description, and identity all stay exactly linked to it. Nothing needs to be re-uploaded or re-saved. Rename the old exhibit first if the new one's name is nearly the same.

You can't rename one exhibit to the exact name of another — you'll get a clear error if you try, rather than ending up with two exhibits sharing a name.

Delete an exhibit

Open the exhibit, go to the Delete tab, and click "Delete this exhibit". You'll be asked to confirm once — after that, its photos and description are gone for good.

Test photo matching (Identify)

Identify is one of the three fixed selections at the top of the page, alongside Exhibits and Venue — it doesn't depend on any exhibit being open. It lets you try a photo the way a visitor would — without Seba answering or speaking anything. It only shows which exhibit the photo matches, using exactly the same matching that visitors' phones use, so it's the best way to check that a newly added exhibit is recognized.

1. Click "Identify" at the top of the page.
2. Give it a photo Click "Choose photo file", or "Use camera" and frame the piece with the same zoom controls described under Add more photos — tap the picture or press "Take photo".
3. Read the result A match shows the exhibit's name with its "Confidence" and "margin" (as percentages) figures, and lists the closest candidates. If the tool isn't sure, it says "No confident match" and gives the reason.
4. Say whether it was right (optional). Under "Was this right?", click "✓ Correct match", or type the exhibit it should have been and click "✗ Submit correction". Your answer is recorded to help tune the matching; it doesn't change any exhibit.

How "confident" is decided. Confidence (the score) is how similar the photo is to the closest exhibit's reference photos — higher is better, and it's shown as a percentage. Margin is how far ahead that exhibit is of the second-closest one, also as a percentage. A photo is accepted only when its confidence reaches the Minimum score and its margin reaches the lead required for its score band: the better the score, the smaller the lead needed. With the standard settings, a score of 45% or more needs a margin of at least 5%, 60% or more needs 4%, and 75% or more needs 3%; below 45% is never accepted. That's why two look-alike exhibits can give "no confident match" even with a high score: neither clearly beats the other. You can change these numbers under Venue → Options → Matching Thresholds.

A second try gets a little more room. If the same visitor photographs the same exhibit again right after a "no confident match," the margin required eases slightly — a real visitor re-centering, moving closer, or catching better light often clears it on the second or third attempt without needing a looser setting for everyone. It only applies while the closest candidate stays the same exhibit across those attempts; photographing something else, or succeeding, starts fresh. It still requires a real margin, just a smaller one, so a different but similar-looking exhibit won't be waved through by simply trying again.

Photos you test here are recorded like a visitor's photos, so they appear in the Analytics report too.

Venue settings

Click Venue in the bar at the top of the page. These settings apply to every exhibit and every visitor at your venue. The first three tabs (Header, Theme and Options) each have their own Save button; the last tab, Analytics, is a report and changes nothing (see Analytics).

Header

The welcome line visitors see before they identify an exhibit. Edit it and click "Save header".

Theme

Background material about your venue that Seba draws on for every exhibit, alongside each exhibit's own description. Edit it and click "Save theme".

Options

Click "Save options" after changing any of these.

• Silent Mode — Allow visitor voice inputs (the usual choice) lets visitors speak their questions with the "Touch to Speak" button. Disable visitor voice inputs hides that button for every visitor and mutes the error alarm, for venues where speaking aloud isn't appropriate. Visitors have no switch of their own for this. (On Firefox and Amazon Fire browsers, which can't do voice input reliably, the Speak button is hidden whatever this is set to.)
• Default response length — 15, 25 or 40 seconds. This is only the starting point for a new visitor; they can still change it themselves.
• Knowledge scope — LIMITED keeps the responses to the material you've provided; GLOBAL lets it add general knowledge. This is a fixed venue policy, not something visitors can change. Either way, Seba stays focused on the exhibit, the museum, and the visit — a request unrelated to any of those (homework help, personal advice, unrelated news) gets a polite redirect rather than an answer. A creative request tied to what's in front of the visitor, like a short poem about the piece, is treated as part of the experience and is welcome under both settings.
• Matching Thresholds — how sure Seba must be before it accepts a visitor's photo as an exhibit. There are three rows (Minimum, Medium and High), each with two boxes: the best score the photo must reach, and the lead over the runner-up, meaning how far the best exhibit must be ahead of the second-closest one. Better scores need a smaller lead, and a photo scoring below the Minimum is never accepted. The standard settings are a 45% score with a 5% lead, 60% with 4%, and 75% with 3%. Enter percentages; decimals are fine (for example 75.5). The three scores must be in order (Minimum, then Medium, then High, all between 0 and 100), or the page won't save. "Restore values" immediately replaces all six with the standard settings and saves them, after a confirmation — a quicker way back if custom values aren't working out. The change takes effect for the next photo a visitor takes.

Change the thresholds with care. Lowering a required lead makes photos match more easily, but it also makes a wrong exhibit more likely to be picked, especially between look-alike exhibits. If the right exhibit keeps failing on a small lead, adding more varied photos of both exhibits is the safer fix. After any change, try a few photos on Identify.

Analytics

Click Venue at the top of the page, then the Analytics tab (the last one). It shows how visitors are using Seba: how many photos were identified and how many failed, which exhibits are viewed most, and how visitors ask their follow-up questions. It is a report only; nothing on this tab changes an exhibit or a setting.

How visitors are counted

When someone opens Seba, the server gives their browser a random ID number, kept in a cookie. Every request they make (a photo, an introduction, a question) is added to a log with that ID and the time. The log holds no names, no network addresses, no photos and not the text of anyone's questions. Touching New Visitor in the app removes the ID, so the next person to use the phone counts as a new visitor.

Choosing the period

A line at the top says when the log begins and when it was last updated ("Log available from … to …"), in your computer's local time. The From and To boxes start filled in with that whole span. Change them to look at a shorter period, for example to leave out the weeks of testing before real visitors arrived, and click "Full range" to go back to everything. The minute shown in the To box counts in full.

Identification data

Click Identification data for the total identifications in the period, how many succeeded, and how many failed, broken down into why:

• Threshold failures — the score never reached the Minimum in Matching Thresholds. The photo didn't really resemble anything on file.
• Margin failures — the score was fine, but the lead over the next-closest exhibit wasn't (see Test photo matching). Often two similar-looking exhibits, or one that could use more reference photos.
• Other failures — an unrelated reason, most often the matched exhibit having no description yet.

Failure rate is all three of these together, as a percentage of every identification attempt in the period.

Exhibit data

Click Exhibit data for three tables:

• Follow-up questions (all exhibits). How often visitors touched a suggested question, and how often they asked a question of their own, spoken or typed (with the split shown underneath), plus the total. "Not recorded" counts questions from phones that still had an older version of the app open.
• Visitor sessions. How many visitors scanned at least one exhibit, how many sessions (visits) that made, and the average session time. A session is a run of one visitor's requests with no gap of more than 6 hours. Its time adds up the time between the visitor's requests, leaving out any pause longer than 15 minutes because the visitor was elsewhere. Visitors who only opened the app without taking a photo are not counted. Times show as minutes:seconds, or hours:minutes:seconds above an hour.
• Exhibits, most viewed first. A scrolling list with one row per exhibit, so scroll down to see the least-viewed ones. Exhibits nobody viewed in the period are at the bottom with 0 views, and an exhibit you have since deleted is shown with "(removed)". The columns are: Views, the number of successful identifications; Avg dwell time, how long visitors spent at the exhibit: the time from it being identified until the visitor took another photo or touched New Visitor, again leaving out any pause longer than 15 minutes. The number in brackets is how many views the average is based on; Avg follow-up questions, the questions asked about the exhibit divided by its views, so a view with no questions counts as zero; and Avg time to last question (minutes:seconds), the time from the exhibit being identified to the visitor's last question about it, ignoring any pause longer than 15 minutes. Only views with at least one question count towards that time, so it is "—" for an exhibit whose visitors never asked anything. Avg dwell time can also be "—": the last exhibit of a visit has nothing after it, so when the visitor stopped looking is not known and that view is left out rather than guessed.

A view of an exhibit lasts until the same visitor identifies another exhibit, touches New Visitor, or goes quiet for more than 6 hours (after that they count as starting a new visit).

Reading the numbers sensibly

• Browsers, not people. A visitor who clears their cookies, browses privately or switches to another browser looks like a new visitor, and two people sharing a phone look like one unless someone touches New Visitor.
• Your own testing is included. Photos taken on Identify and any use of the app by staff are counted like a visitor's. Use the date boxes to leave out the period before real visitors started. If your developer imported earlier history from the old debug log, that period is mostly testing too.
• Times are approximate. The report can only see when requests arrive. Dwell time and session time are the time between a visitor's requests, and a pause of more than 15 minutes is not counted, so they are best used to compare exhibits and periods rather than as exact stopwatch readings. "Time to last question" is measured only up to the visitor's last question, because nothing shows how long they stayed after it.
• Tell your visitors. Because a cookie is used, check that your privacy notice mentions it. Your developer or legal adviser can help with the wording.

What makes a good reference photo set

This is what visitors' phone photos get compared against, so the set matters more than any single photo does.

• Use at least five photos per exhibit. More is fine — there's no upper limit worth worrying about.
• Vary the angle and distance. A visitor won't stand exactly where you did — a few different angles make the match far more reliable than five near-identical shots.
• Include one high-quality shot if you have one — a professional or catalog photo, if it exists — alongside your own phone photos of the piece as it's actually displayed.
• Match real gallery conditions for at least some of the set: the lighting and case glare visitors will actually be dealing with, not just ideal studio conditions.
• Keep other objects out of frame where you can — a crowded shot with several pieces visible gives the matching system less to go on for this one specifically. If you can't get close, zoom in on the piece when you take the photo.
• Test it afterwards. Once the photos are uploaded, photograph the piece the way a visitor would on Identify and confirm it comes back as the right exhibit.

What makes a good exhibit description

The description is the material Seba works from when it introduces an exhibit and answers visitors' questions about it, so what you put in it decides what visitors hear.

• Plain writing is fine. Everything is free-form. Seba reads the text directly, so plain, unformatted prose is completely fine. No headings, bullet points or layout are needed.
• Start with the basics, then go deeper. A good description typically includes the title, date, culture or period, artist or maker, medium, dimensions and provenance. Then it goes deeper: history, significance, symbolism, stories worth telling, restoration, controversies and cultural context.
• Aim for 300–1,000 words; more detail is generally better. Seba automatically adjusts its own vocabulary and depth to each visitor's stated age and knowledge level, so richer source material simply gives it more to draw on for visitors who want to go deeper, without changing what a casual visitor hears.
• Mention related exhibits by name. Nearby pieces, similar works elsewhere in the collection, and pieces from the same period or maker. This lets Seba draw comparisons and point visitors toward other things worth seeing. It's worth cross-referencing in both directions: mention exhibit A in exhibit B's description, and B in A's.
• Keep it honest. Don't state uncertain or disputed facts as settled. Seba is instructed to say plainly when something isn't documented rather than guess, so it's better for a description to leave a gap than to fill it with speculation.
• Include only what you'd want said aloud. Don't include anything the museum wouldn't want spoken aloud to a visitor of any age or background, since Seba may reference or repeat this material if asked.
• Any of four languages. The description can be written in English, German, French or Spanish; choose the matching "Written in" language when you save it. Whichever you use, Seba translates and speaks to each visitor in their preferred language.
• Keep a copy. Deleting an exhibit deletes its description text too, so keep the text in your own files if you may want it again.

Troubleshooting