LumiLab Is Coming: A Local-First AI Photo Manager Built for Real Libraries

🌟 A Promise We Made to Ourselves

“Take command of tens of thousands of photos — without uploading a single one.” — This is LumiLab’s core commitment, and the design principle that guided every decision we made.

A few years ago, we faced the same problem ourselves: thousands of photos scattered across multiple drives, burst shots we never got around to culling, EXIF fields half-filled or missing entirely, and no reliable way to find photos of a specific person without scrolling through dates and hoping for the best.

The tools out there either required uploading to the cloud, offered disappointingly shallow features, or demanded weeks of onboarding before you could accomplish anything useful.

LumiLab is the tool we built for ourselves. And now it’s almost ready.

🔭 What LumiLab Is — and Isn’t

Before diving into features, we want to be clear about positioning — because “photo software” covers an enormous and confusing spectrum.

LumiLab occupies the professional management zone.

If you just want to flip through today’s shots, your phone’s built-in app is fine. But if you’re facing tens of thousands of photos accumulated over years — needing systematic cleanup, organization, annotation, and metadata management — LumiLab was designed for exactly that.

👥 Who Needs LumiLab?

Photographers & Creators

You came back from the weekend with five hundred shots. You need to cull, tag the capture info, and archive by project. You don’t need Lightroom’s full editing suite — you need a professional workbench that can import quickly, cull efficiently, and manage metadata in a structured way. That’s LumiLab.

Family Library Keepers

Five years of photos spread across three hard drives and two phone backups. You know it needs organizing, but every time you open a folder you don’t know where to start. Duplicates eating up storage, childhood photos impossible to find, dates scrambled across devices. LumiLab’s library diagnosis feature maps the entire situation first, then guides you through resolving it step by step.

Privacy-Conscious Users

You don’t trust cloud analysis. You won’t upload family photos to any server for AI processing. Every AI capability in LumiLab — face recognition, similarity clustering, aesthetic scoring — runs entirely on your local machine. Photos and feature vectors never leave your device.

🔬 Deep Feature Breakdown

1. Library Diagnosis: Understand the Problem Before You Fix It

LumiLab’s entry point isn’t “import photos.” It’s “diagnose your library.”

Add one or more photo folders (multiple drives supported), and LumiLab scans in the background with live progress reporting. The result is a professional diagnostic report:

• Library overview: Total count and storage used, time span, format distribution, camera and lens statistics
• Action items: Duplicate/similar groups with reclaimable space, quality defects (blur, over/underexposure), missing metadata (GPS, date anomalies)
• People: Number of recognized individuals and photos containing faces

This isn’t a “one-click tidy up” for beginners. It’s a professional system’s diagnostic dashboard. A photographer sees the camera and lens breakdown and thinks “this tool understands my workflow.” A family user sees the gigabytes that could be reclaimed and immediately wants to act.

2. Cleanup: Precise Deduplication, No Accidental Deletions

LumiLab’s cleanup isn’t a simple MD5 comparison.

Exact duplicate detection uses content hashing to precisely identify identical file copies. Reclaimable space is calculated down to the byte.

Similar photo clustering uses perceptual hashing (pHash) combined with quality signals to surface blurry shots, overexposed frames, and slightly-off duplicates from a burst sequence. Results are presented as groups, and you decide which one to keep — LumiLab doesn’t decide for you, but it provides quality scores to inform your judgment.

Safe deletion: Rejected photos move to an in-library trash folder by default. They’re never immediately and permanently deleted. You choose the deletion method in settings; permanent deletion is always an additional explicit step, never the default.

3. People Recognition: Face AI That Runs On Your Machine

LumiLab’s face recognition runs entirely on-device with inference accuracy matching commercial-grade face recognition systems.

People Wall: Once analysis completes, your portrait photos are organized into a person gallery — as intuitive as a contacts list.

Confirmation Inbox: LumiLab doesn’t force-merge similar faces. Instead, it builds a confirmation queue — asking you “Are these two the same person?” one pair at a time. Your answers continuously refine the clustering.

Live clustering adjustment: Similarity thresholds can be dragged in the UI and results update in real time, without rerunning the model. Within seconds you can compare strict versus relaxed matching and find the balance that fits your library.

False positive exclusion: Faces on animals, statues, and artwork are common false positives. LumiLab supports one-click “Not a person” marking to permanently remove them from the People Wall.

4. Metadata Writeback: Two-Stage Safety, Full History

This is where LumiLab differs most from competitors — and where we invested the most engineering effort.

Structured parsing: LumiLab fully supports EXIF / IPTC / XMP metadata protocols, with parsed fields including: title, description, creator, copyright, city, country, tags, capture time, camera model, lens model, ISO, aperture, shutter speed, and focal length.

Batch draft editing: Select multiple photos, edit a field (such as “copyright” or “location”), and LumiLab preserves each photo’s original value while generating a draft patch plan — you can preview the before/after for every photo before committing.

Two-stage writeback:

1. Draft stage: All edits are stored as drafts. The original files are untouched.
2. Confirm stage: You review all pending changes. Files are written to disk only after explicit confirmation.

Writeback priority: XMP Sidecar by default (written alongside the original, never modifying it), with optional JPEG embedded metadata and a copy-backup mode (backs up the original before writing).

Full history and rollback: Every writeback operation is recorded in a global change log with per-field diff views and one-click rollback to any previous state. Metadata is an asset, not a liability.

5. Browse, Search, and Filter: A Keyboard-Driven Professional Workbench

LumiLab offers four photo views:

• Grid view: High-density thumbnails for fast browsing of large libraries
• Timeline view: Auto-grouped by month (August 2026 · 347 photos), restoring the chronology of capture
• Loupe view: 100% pixel-level magnification for checking focus and detail
• List view: Tabular view with metadata columns for bulk operations

Facet filters support combined filtering across: rating (1–5 stars), Pick/Reject flags, color labels, camera model, lens, tags, people, date range, folder source, and collection membership.

Full-text search supports fuzzy matching on filename, title, caption, camera model, and lens model, with live result counts displayed as you type.

Core operations have complete keyboard coverage: rating (1–5), color labels (W/R/G/Y/P), flags (P/X), navigation (arrow keys), select all (⌘A), batch confirm (Return).

⚙️ Architecture: Local-First Is a Choice, Not a Compromise

LumiLab is built on Tauri v2 + React, with all core business logic implemented in Rust. This isn’t arbitrary:

• Scanning 100,000 photos doesn’t block the UI — background async processing with live progress events
• EXIF parsing and writeback have strict boundary handling — format anomalies don’t crash the app
• AI inference runs through a standard inference runtime, achieving near-native performance
• Local vector database stores face features for efficient clustering, with no network required

AI model weights are not bundled into the installer. They download to the system AppData directory by default, with a configurable path in settings. Models run on-device, and photo data never leaves your machine.

Dual-platform: LumiLab supports macOS and Windows, with path and permission handling optimized for both platforms independently.

🗺️ How LumiLab Compares

Versus Excire Foto (the closest current competitor): LumiLab’s two-stage writeback is safer (Excire writes immediately), the Free tier lowers the trial barrier, the interface is more modern, and the writeback history and rollback are more complete.

Versus Mylio: LumiLab’s AI depth is greater (deduplication + cleanup + face recognition as a complete system), and in-place indexing means your files never move.

🚀 Coming Soon

LumiLab is in the final sprint before launch.

Core feature development across all phases — library management, cleanup, people recognition, and UX polish — is complete and has passed full automated testing. All test suites are green.

Over the coming weeks, we’ll complete:

• [x] Core feature development
• [x] Full automated test suite
• [ ] macOS and Windows packaging and signing
• [ ] Launch preparation and early access
• [ ] Public release

If you’re a photographer or someone with a large local photo library, stay tuned for our launch announcement. When LumiLab ships, something interesting will happen to your photo library.

LumiLab = Professional photo management + Local-first + AI-driven + Two-stage safe writeback

Nothing uploads. Nothing moves. We don’t make decisions for you — but we help you bring tens of thousands of photos under control.