What the feature does, and what it does not
The meal scan looks at one photo and suggests what is on the plate. For each item it gives a name, a portion with an approximate weight in grams, calories and the four macros PepLabo tracks: protein, carbohydrate, fat and fibre. It then adds them up into a total for the meal.
It does not plan meals, set a calorie target or tell you what to eat. The instructions given to the model rule out health, diet or medical advice, judgements about the meal and any remark about weight loss, medication or peptides. Your daily goals are numbers you enter, ideally ones agreed with your clinician or dietitian. The estimate simply saves you typing; the decision about whether the numbers look right stays with you.
Where to find it
There are three ways in: the Scan meal button on the Lifestyle screen, Scan meal in the + menu, and the Meal button when you log food. The camera screen has two modes at the bottom, Meal and Barcode, and you can switch between them before you take the picture.
Taking a useful photo
The screen asks you to keep the whole plate inside the frame, and that is the single most helpful habit: a cropped plate means missing items. You can flip between the front and rear camera and turn on the light for a dim table. You can also pick an existing picture from Photos, which is handy when you ate first and remembered to log later.
For packaged food, the barcode mode is usually the better choice. It reads the product code and looks it up in Open Food Facts, and it does not use PepLabo AI. If a barcode is not found there, you can type the digits yourself or enter the food by hand.
What happens before anything leaves your phone
The first time you send a photo, PepLabo asks “Send this photo to AI?” and explains what will happen. If you tap Send photo, the photo is shrunk on your phone to at most 1,280 pixels on its long side and a smaller copy is sent once, through PepLabo’s server, to Google Gemini, the model behind PepLabo AI. Only that copy and your language go with the request [1]. If you tap Not now, the photo is not sent, you can enter the meal by hand, and the question comes back next time.
One consent covers both PepLabo AI chat and meal photos. If you chose to continue without AI in the chat, meal estimates stay off until you switch AI back on under Settings › Privacy.
According to PepLabo’s privacy policy, the server does not keep the photo in its database, and data you import from Apple Health or Health Connect is never part of the request [1][2]. The policy also notes that anything visible in the picture travels with it, so a photo of just the plate is better than one that shows a prescription label or your face. On Google’s side, the Gemini paid-service terms exclude using the photo to improve Google’s products but allow limited logging to detect misuse [7]. Estimates need a signed-in account.
Reading the result
The result screen lists each item it recognised, with its portion, weight in grams and calories, and shows the meal total at the top. A confidence label (low, medium or high) appears for the meal, and an individual item can carry its own low confidence note. Under the list, a short line from the model names the biggest uncertainty, such as hidden oil or the portion size, followed by a reminder that this is an estimate, not medical or diet advice.
Treat a low label as a prompt to look closely, not as an error. If the scan cannot see food or a drink at all, it says so and offers manual entry instead of guessing.
Correcting it before you save
- Smaller or larger: each item has minus and plus buttons that change its portion in quarter steps, from a quarter of the suggested portion up to four times it.
- Leave out: remove an item that is not on your plate, such as a sauce you skipped. You can put it back with one tap.
- Edit items: opens the regular meal editor, where you can rename a food and change its calories and macros directly.
- Add meal: only this button writes anything to your log. Until you tap it, the estimate exists only on screen, and right after saving a short notice offers Undo.
Why a photo estimate can miss
A camera sees the surface of a plate. It cannot see oil or butter cooked into a dish, sugar in a drink, or what sits under the top layer, and it has to guess depth and portion size from a flat image. Mixed dishes and restaurant meals tend to be the hardest.
Published tests point the same way. One 2025 study gave 52 standardised food photos, in small, medium and large portions, to three general-purpose AI models, earlier versions and not the model PepLabo uses. The two closest models were off by about a third on average for weight and energy, the third by more, and all three underestimated more as portions grew. The authors judged them not yet precise enough where exact amounts matter clinically [5].
A small 2026 validation study in 20 women with obesity compared a dedicated photo app with energy expenditure measured by doubly labelled water. The app underestimated intake by about a quarter on average and agreed poorly for individuals, while 24-hour recall interviews underestimated by about half [6]. Both studies are limited, and neither tested PepLabo, but they explain why the result is labelled as an estimate and why your own correction matters.
Checking against a label
When a package is at hand, its label usually beats any photo. Open Food Facts is a non-profit, volunteer-run open database [3], so PepLabo marks its values as community data, asks you to compare them with the package, and says which nutrients were missing and counted as zero.
The FDA’s guide to the Nutrition Facts label makes a point that applies to photo estimates too: every number on the label refers to one serving, and if you eat two servings you get twice the calories and nutrients [4]. When you correct a photo estimate, matching the portion is usually what moves the numbers most.
When AI is off, or the daily try is used
If AI is off, the photo is not sent and you are offered manual entry. Once the free daily try is used, the scan shows a Pro card and you can still add the meal by hand. With Pro, there is also a daily allowance that renews every day.
Several ways of logging never use AI: searching the built-in sample food list, scanning a barcode, entering a food by hand, and Describe what you ate, which matches your words against the sample food list on the phone. Pro gives access to meal estimates alongside PepLabo AI chat, and nothing you logged is removed if Pro ends.
Where estimates fit in your day
Saved meals sit next to your doses, water, weight and side-effect notes, so a week of entries can show patterns, for example small portions on days with nausea. If you use a GLP-1 medication, the food diary guide covers what is worth noting beyond calories.
Logging every bite is not required for that. A quick photo on the days you want a record, corrected in a few taps, is often enough to bring a useful picture to your next appointment.
