LightCraft is an open-source photo library and raw photo editor, built as a clean-room reimplementation of Adobe Lightroom in pure Rust. It is aimed at photographers who shoot raw and want the catalog, the edits and the original files to stay on their own disk, and at developers who want a photo editor that a script or an AI agent can drive. Be warned up front that it is new: the repository was created at the end of September 2026, and the project's own status badge reads "young and moving fast".
What it does
The project describes itself as photo library plus raw development, the two halves of Lightroom that most people actually use. From the README and the project's own description:
- Raw development with full color grading and tone curves.
- A photo library for browsing your shots; the README's hero screenshot shows an Edit view with a single image in a loupe.
- Masking that can select the sky, the subject or the background.
- Native builds for macOS, Windows and Linux, plus a browser build via WebAssembly.
- An MCP server, so an AI agent can drive the whole application end to end rather than just suggest edits.
All of that is the project's own account of what it does; this channel does not run the tools it covers. The visible traction is easier to verify: GitHub's counters show roughly 8,400 stars and 2,500 forks on a repository about ten days old.
How it works
The clean-room part matters. The authors say they rebuilt Lightroom's behavior from scratch rather than porting any Adobe code, and the badge on the README claims pure Rust, with no second language holding up the core.
Edits run through a 32-bit float pipeline, and the project states that your originals are never touched: editing is non-destructive and the raw file on disk stays as the camera wrote it.
Because the engine is Rust, the same code compiles both to native desktop apps and to WebAssembly, which is how one editor covers macOS, Windows, Linux and the browser.
AI subject masking is the one part that needs a model. It uses SAM 3, and the weights are a separate download rather than something shipped inside the app. Nothing in the materials points to a hosted inference service for it, so that work happens on your own hardware.
Getting started
Start at the repository, github.com/storytold/lightcraft. The README also links to a project page at getartcraft.com/apps/lightcraft — LightCraft is one of several crafting apps from the ArtCraft project — and to a Discord server for the community.
Two honest gaps. First, the project does not list hardware requirements: no minimum memory, no GPU or VRAM figure, nothing about Apple Silicon versus Intel. Raw development and a segmentation model are both demanding, but the authors have published no numbers, so plan on testing it yourself. Second, the copy of the README available here is cut off before the install section, so there is no command to quote. If you want a prebuilt binary rather than a build from source, check the repository and the project page for where the current download lives; do not assume a one-command install until you have read that section.
On money: the code is open source, Apache-2.0 on GitHub, with the README badge offering MIT or Apache-2.0, and nothing in the materials asks for an API key, an account or a paid cloud service. The SAM 3 download is the only extra, and it runs locally, so the cost is disk space and compute on your own machine rather than a bill.
When to use it / when not
The main catch is age. This is a ten-day-old reimplementation of an application Adobe has shipped for close to two decades, so read the feature list as a starting point, expect gaps, and do not point it at a catalog you could not afford to rebuild. Keep your raw files backed up.
Use it instead of Lightroom when you want local, non-destructive raw editing with no subscription and no cloud catalog, or when you want an agent to perform the edits over MCP, which Adobe's product does not offer.
Don't use it if you depend on a mature plugin ecosystem, on printing and soft-proofing workflows, or on round-tripping to Photoshop; and don't count on AI subject masking until you have fetched the SAM 3 weights.
Alternatives
The reference point is Lightroom itself: mature, subscription-priced and tied to Adobe's cloud. Older open-source raw developers also exist, and the materials here make no comparison against them, so judge rendering quality on your own files rather than on anyone's claim.
Photographers who are uneasy about keeping a career's worth of images inside a subscription should take this repository seriously, but as something to watch and test rather than migrate to this month. The interesting bet is the architecture: one Rust engine that runs natively, runs in a browser, keeps the originals untouched, and exposes itself over MCP so an agent can do the dull half of a cull or a grade. If the project keeps the pace its first ten days suggest, that combination is not available anywhere else.