Photool

A local, free photo library for photographers who shoot RAW: import, cull, develop and export, the parts of Lightroom you actually use, with no subscription and no cloud.

Photool runs on the computer that holds your photographs and opens in your browser. Your files stay where they are, in ordinary folders. Ratings and edits go into a local catalog and into standard XMP sidecars, which Lightroom can read too.

It's built for the everyday work of someone who shoots a lot of events and family days: get a card onto the disk, get through two thousand frames quickly, make the keepers look right, and share them.

Shared as-is. Photool is a personal project published in case it's useful to others. It's used every day and well tested, but there's no support and no roadmap. Bug reports with clear steps are welcome; pull requests may not be reviewed.

What it does

• Import from a memory card into date folders, with checksums and duplicate detection.

• Cull fast, by keyboard: stars, pick/reject, colour labels and tags. Bursts are grouped automatically, with "keep the sharpest" and side-by-side compare. Soft-focus and blown-highlight warnings, and a 1:1 loupe for checking focus.

• Develop RAW files in linear light, starting from the camera's own look, which Photool learns from each camera's JPEGs:

  • exposure (±5 EV), highlights and shadows, white balance, tone curve, colour mixer, black & white
  • local masks: gradients, brushes, subject, sky, people and faces, and any object you point at or name (a box or clicks, chosen by Segment Anything), with zoom and pan to work on small things
  • depth read from the picture: a lens blur, and masks that reach a band of distance
  • double exposure: one photo laid over another in linear light, RAW or JPEG
  • spot healing, noise reduction (AI), sharpening, lens correction, upright
  • virtual copies, presets, copy/paste and "match this look"

  Edits are non-destructive and written to XMP.

• People: faces are found and grouped on your own computer. Name a group once and search by name.

• Search the whole library in one line (rating:>=3 iso:>3200 date:2025-06), or by what's in the picture ("kids at the beach").

• Places, a year in review, collages and a slideshow.

• Export sized JPEGs with your credit in the EXIF, and with location stripped if you want.

• Other devices at home: browse and cull from a laptop or phone you've paired. Nobody else on the Wi-Fi sees anything.

• AI agent control (optional): the library is also available as MCP tools, so an agent like Claude Code can search, propose a cull with reasons, develop and export. Proposals wait in the grid for you to accept.

Cameras: Canon CR3 and CR2, DNG (Ricoh GR and others), camera JPEGs, and most other makes' RAW files through LibRaw (NEF, ARW, ORF, RW2, RAF, PEF). The camera-look learning works best on cameras that write RAW+JPEG.

Platforms

	Status
Windows 10/11	Supported. What it's developed and used on daily.
macOS (Apple Silicon or Intel)	Expected to work; less tested. No tray icon or start-at-login yet.
Linux	Expected to work; untested.

No NVIDIA card needed. Everything runs on the CPU. The one GPU-accelerated step, noise reduction, uses any graphics card on Windows (DirectML) and the Apple Silicon GPU and Neural Engine on a Mac (Core ML).

Getting started

You need Python 3.11+, Node.js 18+, and ideally ExifTool.

git clone https://github.com/FoRJu/Photool.git
cd Photool
setup.bat              # macOS / Linux: ./setup.sh
# edit backend/config.yaml: set library_root to your photo folder
serve.bat              # macOS / Linux: ./serve.sh

Then open http://localhost:8000.

The full walk-through, covering every setting, starting at login, other devices, updating and moving to a new computer, is in docs/setup.md.

Documentation

• docs/setup.md: installing and configuring it on a new computer.
• docs/guide.md: how each feature works and why, with the measurements behind the design and the keyboard shortcuts.

Privacy

Faces, descriptions and edits are computed and kept on your computer, and no photograph or catalog data leaves it. Photool goes online for only two things:

• the optional models below, downloaded the first time you use a feature that needs one
• map tiles from OpenStreetMap, when you open a map of where your folders were shot

Models and their licences

Optional models are downloaded on first use. They are not part of this repository, and each has its own licence. Check them before any commercial use. In particular, the ArcFace weights (InsightFace) and SegFormer (NVIDIA) are released for non-commercial use only. SAM 3 is under Meta's SAM License, which is not an open-source licence; it is fetched only if you choose it and agree to it.

Used for	Model	Source
Finding faces	YuNet	OpenCV model zoo
Grouping faces	ArcFace (InsightFace buffalo_l / buffalo_sc), or SFace	InsightFace releases, OpenCV model zoo
Smart search	CLIP ViT-B/32 (ONNX)	Hugging Face, Xenova/clip-vit-base-patch32
Noise reduction	TreeNet denoise (RawForge)	github.com/rymuelle/RawForge
Subject mask	U²-Net-p	rembg releases
Sky and people masks	SegFormer-B0 (ADE20K)	Hugging Face, Xenova/segformer-b0-finetuned-ade-512-512
Parts of people (hair, skin, eyes, lips, clothes)	SegFormer-B5 face parsing (CelebAMask-HQ), SegFormer-B2 human parsing (ATR), if chosen	Hugging Face, jonathandinu/face-parsing, mattmdjaga/segformer_b2_clothes
Object masks	SAM 2.1 Hiera base+ (Apache-2.0); SAM 3 if chosen (SAM License), our ONNX export of Meta's weights	Hugging Face, onnx-community/sam2.1-hiera-base-plus-ONNX, wW-Ww/sam3-ONNX
Depth (lens blur, depth masks)	Depth Anything 3 Mono Large, 4-bit (Apache-2.0); Depth Anything V2 Small (Apache-2.0) without a graphics card or as backup	Hugging Face, Jens-Duttke/Depth-Anything-3-MONO-ONNX, onnx-community/depth-anything-v2-small
Place names	GeoNames cities5000	geonames.org (CC BY 4.0)

Built with

Python, FastAPI, rawpy/LibRaw, OpenCV, ONNX Runtime, SQLite, React, Vite and Tailwind CSS. Developed with a lot of help from Claude Code.

Support and thanks

Photool is free. If it saves you time and you'd like to say thanks, you can buy me a coffee on Ko-fi.

Thanks to @antVRC for reports and ideas that improved the gradient tool, the slideshow's look, and masking.

Licence

MIT. See LICENSE.