krites¶
Take the 3,000–4,000 frames from a full day's wedding shoot and judge them — cull the unusable, rank the rest, then auto-correct and lightly retouch the keepers — straight from a local browser studio.
“KREE-tees” · from Greek κριτής, “the judge” · a bespoke, private alternative to AI culling services, tailored to the photographer's own taste
Status — design in progress, build not started
krites is scaffolded and its design is captured in the specs below; feature implementation hasn't begun. The master design spec is the source of truth. These docs describe the intended product and grow with the build.
What krites does¶
It runs the whole post-shoot pipeline — cull → develop → retouch → remove → export — over a single shoot:
- Culls at scale — flags out-of-focus / motion blur, bad exposure, closed eyes & blinks, and near-duplicate bursts, giving every frame a keep / maybe / reject verdict with reasons for you to confirm.
- Auto-corrects the keepers — straighten, composition crop, a colour look, and light retouching.
- Removes unwanted objects from a scene by inpainting.
- Stays non-destructive — originals are never touched; every decision is a reversible record, rendered only on export.
- Local-first & private — a wedding's photos stay on your Mac by default.
- Fits Lightroom — reads/writes XMP sidecars, so a krites cull shows up in your Lightroom catalog (cull in krites, finish in Lightroom).
It's built on go-tool-base, ships as a single Go binary, and keeps no state of its own — the shoot directory owns the originals, decisions, and exports.
-
The pipeline, the non-destructive shoot workspace, pluggable providers, config-driven cull profiles + looks, the studio, and the phased roadmap.
-
The CLI, studio, and MCP contracts — the
R-*requirements that anchor the tests. -
Ways of working, the design decisions, and how the pieces fit as the build progresses.