Local AI photo finder · Mac · Windows
Your photo library. Type it. Find it.
Type “sunset” — get your sunset photos. Type “mom holding the cake” — get that one. RetinaTag reads every photo on your Mac or PC with on-device CLIP and local Ollama models. Nothing leaves your machine. Ever.
FreeNo account in the appOffline by default
Before you installN°04
- Mac
- macOS 11 or later · Apple Silicon only (M1 and newer) · about 190 MB. It won’t run on Intel Macs.
- Windows
- Windows 10/11 · 64-bit · about 210 MB. The installer isn’t signed with a Microsoft certificate yet, so a blue warning may appear the first time — click More info → Run anyway.
- Update
- Installed before September 2026? Download it once more — later updates arrive automatically.
Your turn — type anything, or try
The problem → the fix
Years of photos. Zero of them findable.
Apple Photos searches what's tagged. Lightroom searches what you tagged. The cloud charges you monthly to keep it lost. Meanwhile the photo of your dog on the beach is in there — somewhere. RetinaTag fixes that in four steps.
- 01
Install & point
Run the installer and show RetinaTag where your photos live — Pictures, an external drive, a NAS share. Add as many folders as you like.
- 02
Models fetch themselves
It checks your hardware and suggests a CLIP tier and an Ollama model that fit. One click; files come straight from Hugging Face and Ollama.
- 03
Indexing in the background
~10 ms per photo for CLIP on Apple Silicon, 1–2 s for a caption. Search works from minute one; results keep streaming in.
- 04
Search by feeling
Type a scene, a colour, a year, a place — in English or Turkish. It understands photos you never tagged in your life.
01 · Find
A search box that knows what you mean.
ASemantic search · CLIP
Search in plain words. “sunset over the sea.”
CLIP — the model built to connect pictures and words — runs locally. Pick the tier your hardware likes: Fast (~155 MB, any Mac), Balanced (~600 MB) or Best (~1.7 GB, M1 Pro and up). Index once; the box answers while you type.
- Three tiers, matched to your machine.
- No tags needed — CLIP reads pixels, not filenames.
- Multilingual — type Turkish, find anything.
BAI tags · 16+ per photo
Every photo gets a written description.
Open a photo and read what the local model saw: sixteen-plus named attributes and a one-sentence caption. Rate it, favourite it, re-tag it with another model. Everything is searchable and exports as XMP — your tags travel with the file.
- Mood, scene, subject, colour — all named.
- Captions in your language — English, Turkish, anything.
- XMP sidecars — Lightroom-compatible.
CLocal Ollama · vision models
Pick your model. Pick your VRAM.
RetinaTag checks your GPU and suggests a model that actually fits — Moondream 2 (1.5 GB) on a thin laptop, Qwen2.5-VL 7B on a 16 GB rig, 32B if you own a 4090. Or point it at your own Ollama server.
- Six curated models — Moondream, Gemma, Qwen2.5-VL.
- VRAM-aware — know before you download.
- Your own server — remote Ollama over HTTP.
02 · Browse
Seven views. One library.
Gallery, Timeline, Map, Cleanup, Dashboard, Calendar and Vault — seven lenses on the same photos, with folders, collections and people one click away.
The whole library in one grid, live counts in the corner — photos, tagged, pending, and a cost meter that stays at <$0.01 with local models.
Folders, collections, people (cluster or recognise) and your AI provider — local Ollama by default.
Every geotagged photo on a globe, a flat map or a heat map. Click a pin: photo, place name, date, EXIF and tags.
Years on top, months below, counts under each. It reads EXIF date-taken, so the chronology survives backups and migrations. Arrow keys jog month by month.
03 · Protect
Photos that don't show up. Even on the same machine.
The vault is invisible until you unlock it. Files are encrypted with a key derived from your PIN; locked photos never appear in search, timeline or cleanup. Ten wrong PINs and it wipes itself — on purpose.
- Argon2id + AES-GCM encryption.
- Auto-lock after the idle time you choose.
- Touch ID on Mac, Windows Hello on PC.
- Recovery code — optional, your call.
04 · Keep it tidy
The boring stuff. Automated.
Plug in. Sorted.
Insert an SD card or USB drive and files land in YYYY / YYYY-MM folders by date-taken. Watch folders index new photos on their own.
Duplicates out. Gigabytes back.
Perceptual-hash duplicates, blur scores and the exact space you'd reclaim. Photos you mark as keepers are never touched.
One button, one job.
Fix sideways photos, import tags from XMP, run a health check, re-extract GPS or dominant colours. All reversible — except clearing all tags, on purpose.
Updates you barely notice.
Signed updates from our own server. A small card offers one-click install and relaunch — never a blocking modal. Or switch the check off.
Privacy, in one line
Hundreds of thousands of photos. Only you can see them.
No cloud upload, no telemetry on photo content, no licence server, and no account inside the app. CLIP runs locally, Ollama runs locally, the vault encrypts to your PIN. We literally cannot see your library — and we like it that way.
- No cloud sync
- No photo telemetry
- No account in the app
- Offline by default
- Open format (XMP)
FAQ
Asked & answered.
Do you store my photos?
What models actually run? Are they bundled?
Do I need a powerful computer?
What's the catch? How is it free?
Can I install it on all my computers?
How secure is the vault?
Where does my library actually live?
Stop scrolling for that one photo. Type how it felt.
Free. No account inside the app.
Before you installN°04
- Mac
- macOS 11 or later · Apple Silicon only (M1 and newer) · about 190 MB. It won’t run on Intel Macs.
- Windows
- Windows 10/11 · 64-bit · about 210 MB. The installer isn’t signed with a Microsoft certificate yet, so a blue warning may appear the first time — click More info → Run anyway.
- Update
- Installed before September 2026? Download it once more — later updates arrive automatically.
From the same studio
