Every few months, a new AI photo restoration app goes viral. Someone posts a crisp, colorized, sharp version of a grandparent’s faded 1960s portrait, and within days, millions of people are uploading their own family archives to find out what the technology can do for their oldest, most irreplaceable photos.
The results are often impressive. The privacy questions that come with them are worth asking before you upload.
How These Apps Work
AI photo restoration, enhancement, and upscaling apps — the most popular include Remini, VanceAI, PhotoAI, and several others that cycle through viral attention — share a common technical architecture: your photo leaves your device, travels to a remote server, gets processed by a machine learning model running on that server, and the enhanced result is returned to you.
This is not incidental to how the technology works. Running a high-quality image enhancement model on a phone’s processor would produce slow results; running it on powerful server infrastructure produces the sharp, detailed outputs that make these apps worth using. The server-side processing is the product.
Which means your photos — including the faces of your relatives, your parents, your children, and anyone else who appears in your family archive — are being uploaded to a third party’s servers every time you use one of these apps.
What the Privacy Policies Actually Say
The details vary by app, but several patterns appear consistently across the privacy policies of the major players in this space.
What typically gets processed and retained
Most of these apps’ privacy policies state that uploaded images are processed and then deleted from their servers within a short window — often cited as 24 to 72 hours. The processed result is returned to you, and the original upload is not retained beyond the processing period.
That’s the standard claim. It’s worth knowing what it actually covers and what it doesn’t.
What “deleted” usually means: The uploaded image file is removed from active processing queues. Whether copies exist in backup systems, logging infrastructure, or intermediate processing stages during that window is a separate question most policies don’t address specifically.
What the policies say about training data: This is the more material question. Several AI photo apps explicitly reserve the right to use uploaded images for AI model training and improvement — though often with an opt-out or with language specifying this applies only when users have “explicitly opted in.” Remini’s privacy policy, for example, states that datasets for AI development “may come from” user-provided images “where users have explicitly opted in to model training.” The word “explicitly” is doing significant work there — whether a checkbox during initial app setup counts as explicit opt-in varies by interpretation.
Other apps are less specific. Terms that allow use of “anonymized” images for “product improvement” leave substantial room for interpretation about what anonymization means for a face photo and what product improvement covers.
Biometric data and face processing
Face restoration is specifically what these apps do. The AI model identifies facial geometry, reconstructs missing or degraded features, and generates a sharpened or colorized version of the face.
This involves processing what most privacy laws — including GDPR in Europe and the biometric privacy laws of several US states — classify as biometric data. Facial geometry extracted from a photo is considered biometric because it uniquely identifies a person from their physical characteristics.
Under GDPR, processing biometric data for the purpose of uniquely identifying a natural person requires a specific lawful basis, explicit consent, and strict handling requirements. Under Illinois’ BIPA — which has been the basis for numerous class-action lawsuits against tech companies — collecting biometric identifiers without proper consent and data destruction policies can expose companies to significant liability.
Most AI photo apps are headquartered or incorporate in jurisdictions where these specific laws apply to varying degrees. Whether they’ve actually obtained proper consent for biometric processing is a question worth examining before you upload.
The Old Family Photo Problem
Here’s what makes this privacy question particularly important for this specific category of app.
When you upload a blurry photo of your grandmother at her wedding in 1958 to an AI restoration app, you’re uploading the biometric data of someone who:
- Is likely deceased or elderly
- Almost certainly never consented to any AI processing of their image
- Has no way to exercise any privacy rights with respect to that processing
The same applies to photos of living relatives — parents, siblings, children — who appear in your family archive but didn’t choose to submit their face data to a third-party AI processor.
The photo is yours. The faces in it belong to the people depicted. In jurisdictions with strong biometric privacy laws, the question of whose consent is required for AI processing of a face is not fully settled — but the moral question is easier to answer than the legal one.
What Happens When You Restore a Face
The AI model doing the restoration isn’t just sharpening existing pixels. It’s generating new visual information — reconstructing facial features based on patterns learned from whatever training data the model was built on.
When you restore a degraded old photo, the sharp version you get back is partly your relative’s actual face and partly the model’s inference about what that face looks like, based on its training data. The result looks convincingly real because modern face restoration models are very good at generating plausible facial details.
This means the “restored” photo is, to some degree, AI-generated content. The face might look sharper and more realistic than the original, but some of the visual information — the exact texture of skin, the specific shape of an eyebrow, the definition of a jawline — is synthesized rather than recovered from the original image.
This is worth knowing if you’re using these restored photos for family history purposes, where people might assume they’re seeing what their ancestor actually looked like.
The Encryption Picture
Most major AI photo apps encrypt uploads in transit using HTTPS. That’s the standard for any competent modern web application and prevents the image from being intercepted in transit between your phone and their servers.
What they encrypt at rest, and how, varies significantly and is often not disclosed in detail. When a photo is sitting on a processing server between upload and the return of the result, the specifics of key management, server access controls, and logging retention are typically not addressed in consumer-facing privacy policies.
Server-side encryption without detailed disclosure of key management practices means you’re trusting the provider’s internal security posture. That’s not the same as encryption with keys the provider doesn’t hold.
Red Flags to Look for Before You Upload
Before using an AI photo restoration app, look for these specific things in the privacy policy:
Does the policy explicitly state how long uploads are retained? Vague language like “we delete your data when it’s no longer needed” is not the same as “uploads are deleted from our servers within 72 hours.”
Does the policy address model training? Look specifically for whether your uploads can be used to improve the AI model. If the policy grants a broad license to use uploaded content for product improvement without a clear opt-out, that’s worth weighing.
Is the company subject to privacy laws with real enforcement? Companies incorporated in the EU or operating under GDPR have legally enforceable baseline requirements. Companies without any regulatory anchor have less external accountability.
What data does the app request beyond the photo? Some of these apps request contacts, location, or device identifiers that have nothing to do with photo processing. Excessive permissions beyond what’s needed for the stated purpose are a signal.
Is there a way to request deletion of your data? GDPR and CCPA require deletion-on-request for users in covered jurisdictions. If you can’t find a data deletion mechanism, the policy is either incomplete or the company doesn’t take these requests seriously.
Alternatives That Don’t Require Uploading
For users who want AI photo enhancement without server-side processing, the on-device option is the meaningful alternative — though the gap in output quality is real.
Apple Intelligence on supported iPhone models includes some photo cleanup features that run on-device. These are less powerful than dedicated restoration models but don’t require sending photos off your device.
Topaz Photo AI (desktop) processes photos locally on your computer. The quality is high and competitive with cloud-based options, and nothing leaves your device.
Adobe Photoshop (desktop) with Lightroom includes AI enhancement tools that run locally, though some features use cloud-based processing depending on your settings.
The trade-off is that on-device and desktop options typically require more computing time per photo and may produce less dramatic improvements on very degraded images than the purpose-built server-side restoration models. Whether that trade-off is worth it depends on the sensitivity of the photos and your comfort level with the apps’ terms.
What a Storage-Only Approach Looks Like
There’s also a simpler version of the problem. If you want to preserve your family photo archive without AI processing or restoration, you just need somewhere to store the original files privately.
daftei stores photos and documents without running any AI features on them. There’s no face recognition, no biometric analysis, no AI enhancement or processing of any kind, and no training of third-party AI models on your content. Files are encrypted in transit with TLS 1.3 and at rest with AES-256. Five gigabytes are free; unlimited storage is available on Pro.
This doesn’t answer the question of whether to use AI restoration tools — that’s a separate choice. But it does mean your baseline archive can live somewhere that isn’t doing AI processing on your relatives’ faces.
The Practical Takeaway
AI photo restoration produces genuinely impressive results and there are legitimate reasons to use these apps. The privacy concerns aren’t a reason to never use them — they’re a reason to use them deliberately.
Read the privacy policy before uploading, specifically for model training language and retention periods. Understand that you’re uploading biometric data of people who may not have consented. Consider whether on-device alternatives meet your needs for the most sensitive family photos.
And keep the originals somewhere that isn’t processing them.