Getting Started with MenuLens: Your First Menu, Step by Step

The short answer

MenuLens is an iPhone and iPad app (iOS 17.0 or later) that photographs a restaurant menu and returns each dish as a structured explanation in a language you choose — ingredients, how it is cooked, flavour notes — rather than a word-for-word translation. Getting started takes five steps: sign in with Apple, grant camera and photo access, frame the menu, confirm the target language, read the results.

The free tier gives you 10 menu scans in total. Dish details, dining preferences, menu history and the order chat stay free and are not counted against that quota; the AI "what should we order" flow is the part that needs Premium, at $4.99/month, $24.99/year, or a $4.99 seven-day day pass at US prices.

One thing to know before you rely on it: MenuLens infers ingredients from a photo of words, and the output is informational, not medical or allergy advice. It is a way to know what to ask a waiter, not a way to know what is in the dish. See the MenuLens product page for current availability.

Why you are reading this at the table

The situation is specific and it is not really about language. You are sitting down in a city you have never visited, the waiter has handed you a wall of unfamiliar script with no photographs, and the table beside you is being served something that smells extraordinary. You want that. You cannot name it, and pointing at a random line and hoping is a bad bet when you have an allergy.

That is what makes the obvious tool feel insufficient. A camera translator gives you a clean, fluent, literal rendering of the menu — and it leaves you exactly where you were. You still cannot tell which of six unfamiliar dishes is spicy, which are meat, which are fried, or which two would make a reasonable meal for two people. The problem is not reading the words. It is choosing from them, and the two are different jobs.

So the question a first-time user actually has is not "how do I install this". It is: when I point this at a menu, what will it actually tell me, what will it refuse to tell me, and what happens to my free scans?

What actually happens during a scan

Knowing the pipeline explains most of the behaviour you will observe, including the awkward parts.

• The image is compressed on your device first. You are not uploading the full-resolution photo from your camera roll. That is why a menu shot in dim light still completes.
• The image goes to a backend, which forwards it to a third-party AI model provider. Menu recognition runs on a vision model; dish explanations and recommendations run on separate text-model routes, dispatched independently so one failing does not blank the whole result.
• Your servers keep no menu photo. Only anonymized usage metadata survives: a request identifier, which model ran, token counts, latency, and whether an image was included.
• The response comes back as a structured dish list, which is why you get a dish you can tap rather than a wall of translated text.

Progress tells you this is happening: the camera screen counts characters as it reads, and detected text regions are outlined live, so you can check it is framing the right page before you commit the shot. Results stream in, so the list starts filling before the whole page is processed.

The permissions, and what each one is really for

MenuLens requests five system permissions. It is worth knowing which ones you can decline without breaking the app:

• Camera — photographing the menu.
• Photo library — selecting a menu photo you already took, instead of shooting a new one.
• Microphone and speech recognition — the two-way order conversation: you speak, it translates to the local language; the reply comes back translated. Audio is processed through Apple's Speech framework.
• Location, while using the app — optional, and only ever used to derive a country-level region hint that improves cuisine identification. Your precise GPS coordinates are not uploaded, and the whole thing can be turned off.

An account is required, because menu recognition runs server-side. Sign-in is Apple-only, so the menu screens never touch your contacts or your other photos.

What the free tier actually gives you: ten scans

This is the part competitor pages bury, and it is the first thing that surprises people at a restaurant on the fourth evening of a trip.

So the free version is genuinely complete for understanding a menu. What you pay for is volume and decision-making: unlimited scans, and the flow that proposes a balanced combination of dishes given your party size, budget and dietary needs.

At US prices that is $4.99 for a month, $24.99 for a year — listed as "Save 50%" — or $4.99 for a consumable seven-day day pass, which is the option that makes sense if you are on a nine-day trip and want it for exactly the days you will eat out. Other territories see Apple's own local prices. All purchases are managed through your Apple ID, and cancelling in iOS Settings keeps your access until the current period ends.

Getting a clean read the first time

Recognition quality is decided before the shutter fires, and the difference is mostly geometry and light rather than model quality:

• Move closer rather than zooming. Digital zoom throws away resolution you cannot get back; filling the frame with the menu does not.
• Kill the reflection. Tilt the page relative to the light source rather than using flash — a specular hotspot across the text is the single most common cause of a garbled result.
• Keep the text horizontal and steady. Two hands, or rest on the table.
• Use multi-shot for anything longer than one page. Multi-shot stitching combines several captures into one menu, and it is the standard answer for a laminated four-page restaurant book.
• Handwriting works, at a cost. Legible handwriting is readable; decorative or faded script is not. Printed menus are markedly more reliable.

If a result looks wrong, re-shooting beats editing — a second capture with better light is faster than reasoning about a bad transcription. More on recognition failures is in the FAQ.

Where AI stops and staff take over

Be blunt about this, because it is the one claim in this product that should not be taken on faith.

MenuLens reads what is printed on the menu and infers the rest. A menu that lists "house special with shellfish" and a menu that lists only "chef's catch" produce very different confidence, and the app cannot tell you which one it read. Cross-contamination, hidden broths, sauces, frying oil and seasonal substitutions are all outside what a photo of a page can resolve.

So: AI output from MenuLens is informational and does not constitute medical, allergy, or dietary advice. For allergies, religious dietary law such as halal or kosher, or any health restriction, ask restaurant staff directly and rely on the actual ingredients. What the app is genuinely good at is getting you to the right question — "does this contain peanut?" — instead of no question at all.

One related limit worth stating plainly: dishes with no photograph on the menu get no invented one. Reference pictures are drawn from public encyclopedias and image search when available, and some regional dishes will simply have no image. Missing text is a known failure mode, not a bug you can wait out.

What MenuLens does differently

• Preferences are entered once and applied everywhere. Dietary restrictions, allergens, taste profile, budget and party size live in one place and are injected into recognition and recommendation. A preference is a single tap, not a re-typed sentence on every menu.
• Your reasoning stays visible. When MenuLens narrows a menu for you, the basis for the judgement is shown rather than hidden — which matters, because a filter you cannot inspect is a filter you cannot correct.
• The original name stays next to the translation. You are not relying on a translation alone at the counter; you can point at the original characters and say "this one".
• It ends the interaction, not just the reading. The recommended dishes become an order list in the local language that you can show the server, and if the answer is a conversation rather than a choice, press-and-hold voice translation handles both directions.
• Your photos are not their training set. Recognition data is retained only as anonymized metadata, your sessions stay on your device, and no photo is used to train models without a separate explicit opt-in.

Is it worth setting up before you go?

Decide it on this trip, not on the category. If you are eating out at unfamiliar restaurants on several days, the ten free scans will probably cover it and you can decide about Premium from inside the app, at the table, with real context. If you are doing a long multi-city trip, the $24.99 annual or a single $4.99 day pass removes the counting entirely.

Configure your allergens before you leave — the one step materially easier at home than at a table with a waiter waiting. Then read how translation and region context work before your first foreign meal, and what a dish detail page gives you once the first menu is on screen.