v1.23.1 "Grip"

Facet 1.23.1, "Grip", after the crew member who builds and steadies the camera rig, for a patch that gets the CUDA image standing again.

Fixed

• The CUDA Docker image builds again. 1.23.0 published only the cpu and cuda-legacy images: the restricted lock's setuptools 84.1.0 conflicted with the CUDA base image's torch 2.11.0 (setuptools<82). The lock is held at setuptools 81.0.0 and CI now resolves each lock on its image's base torch, so this cannot ship again. ghcr.io/ncoevoet/facet:1.23.1-cuda (and 1.23-cuda, latest-cuda) is published again.
• --upgrade-db no longer re-measures every no-subject photo on each run (#257); photos still carrying the neutral 5.0 of a version before 1.21.0 are rewritten once.

Security

• virtualenv 21.14.13 (GHSA-4f8h-vr72-p32c, GHSA-cx6f-vjhg-g3cc). virtualenv is only pulled in by pre-commit; Facet's runtime is unaffected.

Thanks

• @ch1y0q for their first contribution: --upgrade-db no longer re-runs BiRefNet on every no-subject photo (#257), found and measured on a real 13,720-photo library.

Upgrading

• CUDA users who stayed on 1.22.x because the 1.23.0 CUDA image was missing can upgrade directly; read the 1.23.0 notes first.

Not verified

• The cuda-legacy lock resolution on Python 3.11 ran in CI only.

Full changelog: v1.23.0...v1.23.1