On-Device Skin Analysis vs Cloud: Why Where the Photo Goes Matters

On-device skin analysis means your phone finds your face, maps it and scores it without the photo leaving the device. Cloud analysis uploads the photo to a server that runs a larger model. Pori, for example, uses the cloud and deletes the photo once the result is back. Face detection and landmark mapping run easily on a phone; detailed skin scoring by large AI models usually needs a server.

We make Pori, so we've put it first; here's what each app does well. Below is how a skin scan breaks into steps, which of those steps a phone can handle, what really changes for your privacy and speed, and how to tell where any app sends your selfie.

What is the difference between on-device and cloud skin analysis?

The difference is where the computation happens, and therefore who holds a copy of your photo while it does. With on-device analysis, a model stored inside the app runs on your phone's processor. With cloud analysis, the app sends the image over the internet to a server, which runs the model and sends back the result.

Many apps mix the two. A common pattern is to check the photo and find the face on the phone, then send the image to a server for the detailed read.

Which steps of a skin scan can run on the phone?

The early steps, which find and map the face, run on a phone without trouble; the later steps, which interpret skin, are where size starts to matter. Here is a typical pipeline, as described in our guide to how AI skin analysis works.

Face detection

Phones ship with this. Apple's Vision framework includes a face landmarks request that first locates all faces in an image, then finds features like the eyes and mouth. Google's MediaPipe Face Landmarker starts with a face detector optimized for mobile GPU inference.

Landmarks and the face mesh

Landmark mapping is also light enough for a phone. The Google research behind the widely used face mesh describes a model that estimates a 3D mesh of 468 vertices from a single camera and runs at 100 to 1,000+ frames per second on mobile GPUs, depending on the device and model variant. The current MediaPipe version outputs 478 3D landmarks and scores for 52 facial-expression blendshapes. For what those points are used for, see face mesh landmarks explained, and for how they split your face into scored areas, face zones in skin analysis.

Skin scoring and explanations

This is where the trade-off appears. Apple's Core ML lets apps make predictions with machine learning models, and even train or fine-tune them, all on a person's device, so a compact skin classifier can run locally. But the strongest general-purpose vision and language models are much bigger than what a phone holds. For scale, Apple describes its own on-device language model as about 3 billion parameters, paired with a larger server-based model for requests the phone can't handle.

Is on-device skin analysis more private?

Usually yes for the photo itself, though not automatically for everything else. Apple's security team calls on-device processing the cornerstone for the security and privacy of user data, because data that exists only on user devices isn't subject to any centralized point of attack. If the photo is never uploaded, it can't be breached on a server, reused for training or shared with a partner.

Two caveats matter. First, an on-device app can still upload what it learned. Apple's App Store rules say data processed only on device isn't "collected", but anything derived from it and sent off the device has to be considered separately. Scores, face measurements or crops sent to an analytics service are still data about you. Second, the reverse holds too: cloud processing can be designed to keep almost nothing. Apple's Private Cloud Compute, for instance, keeps user data on its servers only until the response is returned, with the goal that personal data leaves no trace. What protects you in the cloud is a short retention period, encryption in transit and a firm no on training. Our skin analysis app privacy checklist walks through how to check those in a policy.

Is on-device analysis faster?

For simple steps it is, because nothing has to travel. A face mesh that runs at hundreds of frames per second gives instant feedback while you line up a selfie. For the full scan, the answer is less clear-cut. A compact on-device model returns a score with no network delay, but a cloud scan with a larger model may read more detail and still finish in seconds on a decent connection. On-device analysis also keeps working on a plane or with weak signal, which a cloud app can't.

Is cloud skin analysis more accurate?

It can be, mainly because server models can be larger and can be updated without waiting for you to install a new version of the app. Size isn't everything: a small model trained carefully on skin images may beat a large general one at a narrow task. What matters more for your results is consistency. The same lighting, angle and bare skin each time will do more for accuracy than where the model runs; our guide on how to take a skin selfie for analysis covers the setup.

How can I tell whether an app analyzes my photo on the device?

1. Try the airplane-mode test. Switch on airplane mode, turn off Wi-Fi and run a scan. If you get a full result, the analysis ran on the phone. If the app shows an error or waits, it needs a server.
2. Read the privacy policy. Look for where photos are processed, how long they're kept, and whether any AI provider receives them.
3. Check the App Store label, with care. Apple's definition of "collect" covers data kept longer than needed to service the request in real time, so a photo that's uploaded and deleted at once may not be listed.
4. Watch what happens after the result. An app that works offline but syncs a cloud history is storing something on a server; the policy should say what.

Where does the analysis run in real tools?

Where does Pori fit, and how does a selfie scan help?

Pori is a cloud-analysis app. According to its privacy policy, your selfie is sent over an encrypted HTTPS connection to an AI model, isn't stored on Pori's servers beyond the time needed to produce a result, and isn't sold or used for advertising or to train third-party models. Your scan history, diary and progress charts stay on your phone. The scan reads a 468-point face mesh region by region and returns an overall score from 0 to 100, scores for Hydration & Moisture, Pores & Texture, Pigmentation & Tone, Anti-aging & Elasticity and Overall health, and a skin-age estimate. The download and first scan are free; Pro is $9.99 a week with a 3-day trial for eligible Apple IDs, or $49.99 a year (US, as of October 2026). It needs a connection to scan, so it won't pass the airplane-mode test.

Whichever approach an app takes, a skin scan describes how your skin looks in a photo. It isn't a medical diagnosis and it can't check moles, so see a dermatologist about any spot that's new, changing or bleeding.