What Is Ente Photos?
Ente Photos is an end-to-end encrypted photo and video backup app from Ente Technologies, Inc., with optional face recognition and natural-language search that run on the device. The App Store lists it in Photo & Video, with a first release on 2 December 2020. It is a cloud service first: your library is stored, encrypted, on Ente's servers, and the AI is an opt-in layer on top.
Ente also makes a separate on-device chat app, Ensu, which has its own review. This page covers only the Photos app and its machine-learning features.
Get It
Ente Photos installs from the app stores on phones and from downloadable installers on computers; the machine-learning features need one of those apps.
iPhone and iPad
- How to get it:
- Ente Photos on the App Store, free download, iOS 15.1 or later per the listing
Android
- How to get it:
- Google Play, F-Droid or the APK on the GitHub release page
macOS, Windows, Linux
- How to get it:
- Installers on ente.com/download: macOS DMG, Windows EXE, Linux AppImage, DEB, RPM and Flathub
Web
- How to get it:
- photos.ente.com, nothing to install
Plans
- How to get it:
- Free: 10 GB. Paid, in USD as listed on ente.com: 50 GB $2.49 per month, 200 GB $4.99, 1000 GB $9.99, 2000 GB $19.99; yearly prices are listed too, and a family plan covers up to 5 members at no extra cost per the homepage FAQ
Website
- How to get it:
- ente.com
Privacy policy
- How to get it:
- ente.com/privacy
Source code
- How to get it:
- github.com/ente-io/ente, AGPL-3.0 per GitHub
Version reviewed: 1.3.66 on mobile, released on 21 October 2026 per the App Store and GitHub; the desktop app is on 1.7.30 (6 October 2026). The iPhone app is 242.6 MB before any model download. Prices are the ones shown on ente.com; PromptQuorum did not check regional store prices.
What Runs Locally and What Runs in the Cloud
The face and search models run on your device, but the photos live in Ente's cloud, and the AI only indexes photos that were backed up to it.
Face recognition
- Where it runs:
- On your device
- What the sources say:
- Faces are detected, turned into numeric fingerprints and grouped into people locally. It is off by default and the privacy policy says it is enabled after your consent. It works in the mobile and desktop apps, not on the web; mobile can merge and split groups, desktop shows them read-only.
Magic search
- Where it runs:
- On your device
- What the sources say:
- Queries are matched against a local index built with a CLIP-style image model; the source code names MobileCLIP S2. Once indexing has finished, search works offline.
Detect Text (OCR)
- Where it runs:
- On your device
- What the sources say:
- Long-press text in a photo to copy it. The iPhone app uses Apple's Vision framework and Android downloads offline models on first use. It is mobile only, runs on demand and builds no searchable index.
Model download
- Where it runs:
- Network, once
- What the sources say:
- About 300 MB according to the help pages. The source code points to models.ente.com, and the privacy policy lists Hugging Face as a model host.
Indexing input
- Where it runs:
- Cloud to device
- What the sources say:
- The app first downloads your originals from Ente in order to index them, and only photos backed up while you are signed in are included; mobile Gallery Mode can index local photos while signed out.
Search indexes
- Where it runs:
- Cloud, encrypted
- What the sources say:
- Finished indexes are encrypted and synced to your other devices, so one device can do the work for all of them. The desktop app indexes faster than a phone.
Photo storage
- Where it runs:
- Ente's cloud, encrypted
- What the sources say:
- Files are stored encrypted on Ente's servers, which the security FAQ places in Amsterdam, Paris and Frankfurt, with three copies at three providers.
These statements come from Ente's help pages, privacy policy and the source code on GitHub. PromptQuorum did not inspect the app's network traffic or run the models.
How to Turn On the On-Device Search
The steps follow the help pages; PromptQuorum has not run them.
- 1Switch on machine learning
Why it matters: On mobile open Settings, Machine learning; on desktop open Settings, Preferences, Machine learning. Enable Machine learning and, if you want, Local indexing. - 2Let the first index finish
Why it matters: For a large library start on the desktop app, which indexes faster, and let the indexes sync to the phone. Enabling it before a big import avoids downloading photos twice. - 3Name people and search
Why it matters: Open People, name a group, then type the name or a description such as "beach sunset" into the search bar. - 4Spare a low-RAM phone
Why it matters: On phones with 4 to 6 GB of RAM and a big library, turn off Local indexing and let the desktop app do the work.
Encryption and Data Flow
Per the maker, files and their metadata are encrypted on your device before upload, and only you hold the keys. The help pages say the keys derive from your password with Argon2id, files are encrypted with XChaCha20 and XSalsa20 plus Poly1305 through the libsodium library, every file and album has its own key, and metadata such as capture time, location and description is encrypted too. Shared albums are unlocked with the recipient's public key; for public links the decryption key sits in the URL fragment and is not sent to the server, according to the FAQ.
The privacy policy, last updated on 16 October 2026, says Ente collects your email address, public keys, anonymized crash reports and server logs including the IP address, and collects no analytics about how you use app features. Subprocessors named include Scaleway, Hetzner, Backblaze, Wasabi, Cloudflare and Stripe. Face-based search data is described as derived on your device and encrypted before it leaves it.
📌Note: These are the maker's declarations and documentation, not audit results from PromptQuorum, which did not inspect network traffic or the cryptography in the code. The policy states that the company is headquartered in the United States.
What We Could Not Verify
- Quality and speed. PromptQuorum did not install the app or run the models, so the accuracy of face grouping, the relevance of search results and indexing time are not assessed; the help pages themselves say face models struggle with babies and toddlers.
- Encryption and network traffic. The claims come from Ente's own documents; PromptQuorum did not inspect traffic or the code's cryptography.
- Metadata scope. The architecture document does not say whether thumbnails, file names, file sizes or IP addresses are covered, while the FAQ names capture time, location and description as encrypted.
- Free-plan sharing. The homepage lists album sharing and public links in the free plan, whereas the security FAQ says only paid plans can share albums.
- Unexplained third parties. The policy lists Anthropic, OpenAI and xAI under "Research and documentation" without saying what data, if any, they receive.
- Undocumented models. The source code lists dog and cat face and body models that the help pages do not describe.
- Store prices. Plan prices in local currencies and inside the app stores were not checked.
Competitors and Alternatives
PixlStash
- Platforms:
- Windows, macOS, Linux
- Price / license:
- Free / GPL-3.0
- Key difference:
- Self-hosted image library manager with local tagging, face and semantic search; you run and store everything yourself, with no managed cloud backup
Articles about PixlStash (1)
- PixlStash Review: Self-Hosted Image Library Manager with Local AI TaggingUpdated October 17, 2026
pcAI Photo Retouch
- Platforms:
- iPhone, iPad, Mac
- Price / license:
- Free + in-app purchases / Not stated
- Key difference:
- Offline photo restoration and art styles on the device; edits single images and does not back up or organise a library
Articles about pcAI Photo Retouch (1)
- pcAI Photo Retouch Review: Offline Photo Restoration and Art Styles on iPhone and MacUpdated October 16, 2026
No other photo-library app with on-device AI has a PromptQuorum review yet, so these are the closest directory entries; confirm each app's current price, license and platforms on its own listing.
Frequently Asked Questions
Can I use face recognition or magic search in the web app?
No. The help pages say machine learning is not available on photos.ente.com, which offers basic search by date, file name, description and album only.
What happens if I lose my password and my recovery key?
Ente says it cannot recover the account or decrypt the data. The 24-word recovery key is shown in Settings while you are signed in; if you are signed in on another device you can reset the password from there.
Is my face data used to train AI models?
The help pages say photos and ML data are never used to train AI models by Ente or third parties. The privacy policy adds that sharing a photo lets the recipient's device extract face information from it.
Can I self-host Ente instead of paying for storage?
The documentation describes self-hosting with a quickstart script, Docker Compose or a manual setup, and the server code is open source. PromptQuorum has not tried it, and you would then run and back up the storage yourself.
Has the encryption been audited?
The maker cites a first audit by Cure53 with Symbolic Software and a later one done for CERN, both published on its blog. PromptQuorum has not read the reports.
Verdict
Ente Photos is an actively maintained, open-source way to back up a photo library with end-to-end encryption and to search it by person and description without handing the analysis to a server. Against that, the library still lives in the maker's cloud, the AI works only through the mobile and desktop apps, only on photos you have backed up, and after an initial index that can take a while on a large library. It suits people who want a private alternative to mainstream photo clouds and accept a paid plan beyond 10 GB; anyone who needs independent proof of the encryption or the local-only claims should read the audit reports and test the app's network traffic first.
Sources
- Ente Photos on the App Store — version, release date, size and requirements, checked 24 October 2026.
- Ente Photos release 1.3.66 on GitHub — release date and Android APK.
- Ente website and download page — plans, features, platforms and desktop version.
- Ente privacy policy — data collected, subprocessors and consent for face data, last updated 16 October 2026.
- Ente Help: machine learning, magic search and Detect Text — what runs on the device.
- Ente Help: security and privacy FAQ and architecture document — encryption design, storage locations and audits.
- Ente source code — licence and the machine-learning model list, read on 24 October 2026.