Memoria

Search your photos with words — or with a photo. No cloud. No account. No data leaving your phone.

Screenshots · How it works · Getting started · Architecture · Roadmap · Contributing

---

What is Memoria?

Memoria is an open-source Android app that lets you search your photo gallery using natural language — type "dog at the beach", "birthday cake", or "snowy mountains" and it finds the right photos instantly.

You can also search with a photo instead of words — pick one from anywhere, and Memoria finds visually similar photos already in your gallery. Or open any photo and tap "Find Similar" to see more like it.

Everything runs 100% on-device using Apple's MobileCLIP-S0 model converted to TFLite. No internet connection required after setup. Your photos never leave your phone.

You type:  "my dog with a hat"          You show a photo of a chair
           ↓                                       ↓
    [MobileCLIP Text Encoder]           [MobileCLIP Image Encoder]
           ↓                                       ↓
    512-dimensional vector  ←──── same vector space ────→
           ↓
    cosine similarity search against indexed photo embeddings
           ↓
    ranked results in < 50ms

---

Screenshots

Non-blocking indexing progress	Search results	Full-screen photo viewer

Settings	Search by Image	Filters

---

Features

• Semantic photo search — find photos by meaning, not metadata. "Sunset at sea" finds sunset photos even if they have no tags.
• Search by photo — pick one from anywhere and find visually similar photos in your gallery, no text needed.
• Find Similar — open any photo and find others like it with one tap, reusing the same embeddings your library is already indexed with.
• Filters — narrow search results by date range or folder.
• Full-screen photo viewer — tap any result to view it full-screen, swipe between results.
• Non-blocking indexing — browse while indexing runs in the background, with a real progress bar and pause/resume support.
• Settings screen — live indexing status, storage breakdown, and a link back to this repo.
• Fully offline — MobileCLIP-S0 runs entirely on-device via TFLite. No API keys, no cloud calls.
• Privacy first — photos never leave your device. No account required. No analytics.
• Open source — Apache 2.0. Fork it, extend it, learn from it.

---

How it works

Memoria uses CLIP (Contrastive Language–Image Pre-training), a neural network trained on 400 million image-text pairs. It maps both images and text into the same 512-dimensional vector space — meaning semantically similar things have similar vectors, regardless of whether they came from pixels or words. That shared space is what makes text search, search-by-photo, and Find Similar all the same underlying operation:

Gallery photo  →  [Image Encoder]  →  512 floats  →  stored in Room DB
Search query   →  [Text  Encoder]  →  512 floats  ─┐
Search photo   →  [Image Encoder]  →  512 floats  ─┼─→  cosine similarity search
Find Similar   →  (reuse stored embedding)         ─┘

The model used is MobileCLIP-S0 — Apple's mobile-optimized CLIP variant that achieves the same accuracy as OpenAI's ViT-B/16 while being 4.8× faster and 2.8× smaller.

Component	Details
Image encoder	MobileCLIP-S0 → TFLite float32, ~44 MB
Text encoder	MobileCLIP-S0 transformer → TFLite float32, ~67 MB
Token embedding table	float32, ~100 MB
Vocab	OpenAI CLIP BPE tokenizer, 49,408 tokens
Embedding dim	512 floats per image/query
Similarity	Cosine similarity (L2-normalized dot product)
DB	Room

---

Getting started

Requirements

• Android 8.0+ (API 26)
• ~300 MB free storage (for models + local database)
• 3 GB+ RAM recommended for GPU acceleration

Install

Grab the latest APK from the Releases page. If you're updating from an older version, uninstall the previous version first — the local database format changes between releases while the project is still in alpha/beta.

Build from source

git clone https://github.com/raslenabb12/memoria.git
cd memoria

Model files go into app/src/main/assets/:

mobileclip_s0_image_v2.tflite       44 MB
mobileclip_text_reimpl_v2.tflite    67 MB
token_embeddings_f32.bin           100 MB
vocab.json                           1 MB

Then open in Android Studio and run.

---

Roadmap

• Settings screen (indexing status, storage info, theme)
• Search filters — date range, folder
• Search by photo
• Find Similar
• Camera make/model filter
• Folder selection (index only chosen folders instead of the full library)
• Stop control for indexing (pause/resume already shipped)
• First-run onboarding flow

---

Contributing

Issues and PRs are welcome. If you're reporting a bug, a screen recording or logcat output goes a long way.

---

License

Code is licensed under Apache 2.0 — see LICENSE. The bundled MobileCLIP-S0 model is subject to Apple's ML Research Model Terms of Use.