• DarkRoom 0.13.4
    Benchmarks / CPU and I/O (per commit) (push) Failing after 6m22s
    Benchmarks / Frame budget (on demand) (push) Skipped
    Build and test / Desktop (Linux) (push) Failing after 46s
    Build and test / Layer separation (push) Successful in 29s
    Traceability / Requirement traces (push) Failing after 40s
    🐳 Android image / Build and push (push) Successful in 1s
    Build and test / android-image (push) Successful in 1s
    🐳 Windows image / Build and push (push) Successful in 1s
    Build and test / windows-image (push) Successful in 1s
    Build and test / Android (aarch64) (push) Failing after 2m22s
    Build and test / Windows (x86_64, cross) (push) Failing after 3m5s

    dtourolle released this 2026-09-20 12:04:59 +00:00 | 405 commits to master since this release

    The white balance picker works on raw files, and a well-named library
    groups faster.

    • A neutral picked off a raw frame comes back neutral. Pressing
      "pick" and clicking a near-grey wall on a Canon 6D frame set tint to
      -77 and turned the photograph green. The white balance gains multiply
      camera RGB, before the body's base curve and matrix; the probe was
      read after the matrix, off a display render, and the solve treated
      that sRGB triple as if the gains multiplied it directly. On a JPEG the
      two spaces coincide, which is why it worked there and nowhere else.
      The probe now reads the camera-space tap a merge stitches from, under
      the edit's own framing, with the as-shot balance put on — exactly what
      the operation is about to multiply.
    • A blown highlight is refused as a neutral, the way black already
      was. Sensor white with the as-shot multipliers on is strongly magenta,
      and sampling the overcast sky drove tint to -100 for a patch the
      canvas showed as pure white. The probe declines any channel at or
      above the onset the shader fades clipped pixels from; one constant
      serves both.
    • Each person stands in the grouping pass by at most 100 references.
      Every confirmed face entered as an anchor, so a person with 750
      confirmations cost 750 comparisons against every other face, and the
      scan grew with how well the library was named. A person now enters
      through the 100 faces spanning the greatest volume of their
      embeddings — pivoted Gram–Schmidt, so near-duplicates are passed over
      and the one profile shot is taken early. The rest keep their
      confirmations and are simply not compared. At 3,851 confirmations the
      scan shrinks by about a fifth; at 15,000 it is a fifth of what it was.
    • The manual's white balance recording clicked "reset" instead of
      "pick"; it is aimed and recorded again.

    darkroom-0.13.4-arm64-v8a.apk is the release-signed Android build (API
    28+, arm64). darkroom-desktop-0.13.4-x86_64-linux is the desktop binary.
    SHA256SUMS covers both.

    Downloads