The weights were MIT. The training data wasn't.
Accuracy came second when I picked models for krites. The licence file got there first, and one of those models is still in my lockfile with no licence at all...
Categories
11 pages
krites
Krites is ancient Greek for 'the judge', and judging photographs is the whole job. I built it for an audience of one, my wife: a simple, UI-driven tool (not a command line in sight) that helps her rip through culling 4,000-plus wedding images to her own standards. It leans on local ML rather than a cloud-AI subscription, so it works wherever she is, signal or not. It's since grown an optional AI critic as well, expert feedback on composition, colour and technique, there to help her learn rather than to do the judging for her.
The release that OOM-killed its own compiler
A wedding-photo culler ran out of memory compiling an LLM SDK it has never once called. The release had been running for thirty-eight minutes at that point.
Ten lines of XML so a cull shows up in Lightroom
A cull is worthless if it cannot follow you into Lightroom. Writing XMP sidecars out of krites, and being careful about what not to write.
The version pin you can't lower
A standard-library security fix forced a Go version bump, and then the release failed because the build image shipped an older toolchain.
The feature your users are already faking
The one real user of my photo tool was already faking a feature with a workaround. That workaround was the specification.
The grid that would not behave
A desktop-first photo tool reviewed mostly from a phone, and the CSS grid that refused to behave on the one screen it was being judged on.
About that 'no AI'...
A correction to my own post: the culler does contain machine learning, and the distinction I drew between ML and AI does not survive scrutiny.
Why my photo app won't build for Windows
Dropping the Windows build of a photo tool rather than fixing it, and why that is the right call for a one-person project.
You can't reshoot a wedding
A photo tool that is non-destructive by construction rather than by intention, because the shoot it works on can never be taken again.
There's no AI in my photo culler
A wedding photo culler that contains no machine learning at all, by deliberate choice. Blur, exposure and eyes, done with plain arithmetic.
