§ · AI Photo Scan
Yes, the photoleaves your phone.
If an app can look at a picture of your plate and tell you the calories, that picture is being uploaded. Photo estimation runs on a large vision model, those models run in a data centre, and no amount of privacy copy changes the physics of it. So the question worth asking is not whether the image is sent. It is what gets sent with it, who does the processing, and what is kept afterwards. Here are BiteDeck's answers, in the specific.
§ · The short answer
BiteDeck's AI Photo Scan does send your meal photo off the device, because that is how the recognition works, and any photo calorie app running a large vision model is doing the same thing. What differs is the size of what travels with it. A BiteDeck photo scan is a request with exactly one field in it — the image — downscaled to 1,024 pixels on its long edge before it leaves the phone. No account goes with it, because BiteDeck has no accounts; no device identifier; no part of your food log. It is relayed to Google Gemini for recognition, the item list comes straight back, and the images are not retained by BiteDeck or used to train models. Photo Scan is one of the four Premium features and it is optional — search, barcode, nutrition-label OCR and freeform entry all log a meal without sending anything anywhere.
§ 01 · Why every photo calorie app uploads the photo
Identifying a burrito bowl and estimating how much of it is on the plate is a job for a large multimodal model. Those models are tens of gigabytes and they run on server hardware. A phone can run a small vision model — Apple ships several, and they are why your photo library can find pictures of dogs — but recognising an arbitrary meal, matching it to a named restaurant item, and estimating grams from perspective is a different weight class.
So when a photo calorie app implies the analysis happens locally, that claim deserves one specific follow-up question: which model runs on the phone? A real on-device pipeline can be named. If the claim appears in a landing-page headline but not in the privacy policy, the privacy policy is the document that counts.
BiteDeck does not make that claim, and this page exists so that the answer is on the record rather than inferred. The app is local-first in a way that is checkable: your log lives in SwiftData on the iPhone and syncs through your own private iCloud rather than a BiteDeck server. The AI features are the stated exception, and Photo Scan is the one people ask about.
§ 02 · What a BiteDeck photo scan actually sends
You take the photo. Before anything is transmitted, BiteDeck resizes it to 1,024 pixels on the long edgeand re-encodes it as a JPEG at quality 0.8. Your camera's full-resolution original never leaves the phone — it stays in BiteDeck's own storage on the device, alongside the rest of your meal photos.
The request that carries it has one field in it. Not one field of interest, or one field plus the usual metadata: one field, named image. There is no user identifier to attach, because there is no account. There is no device identifier on this call. There is no slice of your log, no weight, no goal, no history.
What is sent
One downscaled JPEG of the plate you photographed.
Where it goes
Through BiteDeck's own relay to Google Gemini for recognition. The relay exists so the API key lives on a server instead of shipping inside the iOS binary, where anyone could pull it out.
What comes back
A list of items with grams, calories and macros. It lands in the review screen, where you edit it before it is saved.
What is stored in between
Nothing. The relay is a stateless function — it forwards the image, returns the result, and keeps no copy. There is no BiteDeck database of meal photos to search, breach, or hand over.
BiteDeck does not retain your images and does not train on them. That is stated on the privacy page, which also names the processor rather than hiding it behind “our AI partner.”
Photo scan when you want it. Barcode, label, search or freeform when you don't. Both are the same app.
Download on App Store§ 03 · Why there is nothing to attach it to
The reason a BiteDeck photo travels alone is structural rather than a policy someone could quietly revise. Most trackers require an account because your log lives on their servers, and once a server holds your history, every request you make can be associated with it. BiteDeck has no such server: the log is in SwiftData on your iPhone and syncs through Apple's CloudKit private database, tied to your Apple ID, which BiteDeck cannot read.
So there is no profile for a meal photo to join, no “photos you have scanned” list on our side, and no account you could ask us to delete, because you never made one. The full architecture is on how offline tracking works and what private actually means here.
§ 04 · The other three AI features, stated plainly
Photo Scan is the tightest of the four. The others send more, and a page about what leaves your phone is worthless if it only covers the flattering case.
Plate Check
Sends two images — the plate before you ate and the leftovers afterwards — plus the item list from the first scan, because the model has to know what it is subtracting from.
AI Coach
Sends your question along with context about your day: recent macros and meals, your weight trend, body fat. That is a slice of your log, and it goes off-device. A coach with no context gives advice that could apply to anyone.
Voice logging
Sends the transcript of what you said, which is the food description you spoke.
All four
Premium only ($14.99/month or $69.99/year), all optional, and none of them run unless you invoke them. Nothing in the free tier makes an AI call.
While we are being exact about what leaves the phone: the free tier is not silent either, and it would be dishonest to imply it on this page of all pages. A barcode that is not already in the bundled database is looked up against Open Food Facts or USDA, and food search can run a remote phase for the same reason. Those requests carry the barcode or the search term and nothing else, and Open Food Facts is queried directly from your phone rather than through our relay, so it sees your IP address the way any website you visit would. Everything already in the bundled database resolves with no connection at all.
The free tier is the whole tracker, offline, with nothing sent anywhere. Start there and never turn the AI on if you'd rather not.
Download on App Store§ 05 · How to check this for any app, including ours
Cal AI, SnapCalorie, MyFitnessPal, Cronometer, MyNetDiary and FoodNoms all offer some form of photo or AI-assisted logging, and we are not going to characterise anyone else's pipeline from the outside — the only honest source for that is each company's own privacy policy, read on the day you read it. What we can give you is the checklist we would want someone to apply to us.
Does the policy name the processor?
“A third-party AI provider” is not an answer. If the company will not tell you whose model sees your food, you cannot go and read that model's terms.
What accompanies the image?
An account ID turns a meal photo into a row in a profile. Ask specifically, because “we don't sell your data” is a different claim from “we don't attach your identity to it.”
Retention, in a number
How long the app keeps the image, and how long the processor keeps it. These are two different answers and most policies give neither.
Training
Whether your photos can be used to improve a model. Look for the word explicitly; silence is not a no.
Policy or marketing?
If a privacy claim only appears on the landing page and not in the policy, it is positioning rather than a commitment.
Applying that list to BiteDeck: the processor is named, the payload is one image with nothing attached, we do not retain the images and do not train on them. The gap in our own answer is the third item — we publish no retention window for the processing itself, because that is the processor's term rather than ours to state. Naming Google Gemini is what lets you go and check it rather than take our word for it.
And if you would rather sidestep the whole question, that option is real here: the accuracy limits of photo estimation are covered in how accurate photo calorie apps really are, and every meal in BiteDeck can be logged four other ways without a single byte leaving the phone.
§ 06 · FAQ
Do AI calorie apps upload your meal photos to a server?
Yes, essentially all of them do. Estimating calories from a picture takes a large vision model, and those models run in a data centre rather than on a phone, so the image has to travel. An app that offers photo calorie scanning and also claims your photo never leaves the device is describing two things that do not fit together — unless it can name the on-device model doing the work.
Does BiteDeck upload my food photos?
Yes, when you use AI Photo Scan, and only then. The photo is downscaled and sent through BiteDeck’s own relay to Google Gemini for recognition, and the item list comes straight back to your phone. BiteDeck does not retain your images and does not train on them. Photo Scan is a Premium feature and it is optional — search, barcode, nutrition-label OCR and freeform entry all log a meal without sending anything anywhere.
What is sent along with the photo?
Nothing else. The request that carries a photo scan contains a single field — the image itself. There is no account identifier, because BiteDeck has no accounts; no device identifier; and no part of your food log, your weight, or your profile. The relay adds the API key it holds server-side and forwards the image. That is the whole payload.
Is my photo sent at full resolution?
No. BiteDeck resizes the image to 1,024 pixels on its long edge and re-encodes it as a JPEG before anything is sent, so the full-resolution original never leaves your phone. That is enough for a vision model to identify a plate and estimate portions, and it is a fraction of what your camera captured. The original stays in BiteDeck’s own storage on the device.
How long is my photo kept after it is processed?
BiteDeck does not keep it. The relay is a stateless function: it takes the image, passes it to the model, returns the item list, and stores nothing — there is no BiteDeck database of meal photos to search, breach, or hand over. What BiteDeck cannot state for you is the retention window inside the processor’s own pipeline, because that is Google’s term rather than ours. That is exactly why the processor is named on the privacy page instead of hidden behind "our AI partner" — you can go read their terms directly.
Do the other AI features send more than the photo?
Two of them do, and it is worth knowing which. Plate Check compares a before photo with a photo of your leftovers, so it sends both images plus the item list from the first scan. AI Coach sends your question along with context about your day — recent macros and meals, your weight trend, body fat — because a coach with no context gives generic answers. Voice logging sends the transcript of what you said. Photo Scan is the one that sends the image and nothing else.
Can I use a calorie tracker without any AI at all?
Yes, and in BiteDeck that is the free tier. Logging, food search, nutrition-label OCR, Apple Health, fasting, weight tracking and export all work with no internet connection at all, and none of them involve AI. Two free-tier lookups do reach out when you are online: a barcode that is not already in the bundled database is looked up against Open Food Facts or USDA, and food search can run a remote phase for the same reason. Those send the barcode or the search term and nothing else. The four AI features — Photo Scan, AI Coach, Plate Check and voice logging — are the paid tier.
§ · Get it
Know what you sent.
Download on App StoreiPhone · iOS 17+ · Free tier · AI features are Premium