Skip to content

RawLab 0.9.0 (build 15)

Choose a tag to compare

@dancancer dancancer released this 28 Oct 05:05
82f028e

RawLab v0.9.0

Version 0.9.0, build 15. Based on development at
b118e217d1ea6bb1f84a146d9d0b21ee3d8d1cc0, compared with the published v0.8.2
tag. This release reorganizes the editor on every client, adds Lightroom-style
copy and paste, presets, new color and tone tools, photo collections, desktop AI
denoising and automatic desktop updates.

Changes

  • Reorganized editor. macOS, iOS, Android and Windows share six adjustment
    groups with change markers and per-group reset, single-line numeric inputs,
    side-by-side and split comparison, and press-and-hold to view the photo
    before adjustments. iOS and Android add a tap for a clean photo preview;
    Android adds and iOS upgrades to a 256-bin RGB histogram.
  • Copy, paste and history. Copy a photo's adjustments, choose the items to
    paste onto other photos, and undo or redo within the editing session.
  • User presets. Save named adjustment presets, separate from built-in
    looks and imported LUT/DCP profiles.
  • New adjustments. Black point, white point and vibrance; shadow, midtone
    and highlight color grading; a multi-point tone curve that keeps the previous
    S-curve control; clarity and dehaze in both directions. Agent recipes and
    exports support all of them. Clarity and dehaze previews refresh
    immediately; the fit-to-window view approximates them at screen size, while
    100% view and export use the full image.
  • HSL color picking. Pick a color on the photo and drag to adjust the
    matching HSL range. The target mode stays active when the adjustment panel is
    collapsed, with an on-canvas indicator and exit control.
  • Photo collections. Sync adjustments across photos, fine-tune one photo,
    and export each photo with its own frozen settings after confirmation.
  • Processing version 2. New edits use VNG demosaic for ordinary Bayer RAW
    files and an updated denoiser. Existing edits keep processing version 1.
    macOS and Windows offer a processing-version choice.
  • Desktop AI denoising. macOS on Apple Silicon and Windows x64 with a
    hardware DirectML GPU can denoise RAW files with the bundled DualDn and
    Restormer models. The first calculation runs at full resolution; exposure and
    look changes reuse the cached result.
  • Automatic desktop updates. macOS and Windows check, download, verify and
    install updates from GitHub Releases, back up editing data first and offer
    recovery if an installation fails.
  • MCP server. tools/mcp exposes five tools over the existing Agent CLI to
    local MCP clients on macOS and Windows.
  • Large images on mobile. Android and iOS process and export large JPEG,
    PNG and RAW files at full resolution with bounded temporary memory.

Upgrade And Compatibility

  • No downgrade. Android 0.8.2 cannot read the new edit library. Saving
    photos after reverting macOS, iOS or Windows discards the new adjustment
    fields and processing-version markers.
  • HSL rendering. Active HSL adjustments now keep out-of-range color detail
    after contrast and saturation instead of clamping it. Existing edits that use
    HSL can look different; there is no switch to restore the old behavior.
  • Windows single instance. Opening another file through Open With, or
    starting a second copy, asks the user to quit the other RawLab instance; the
    new process then exits.
  • First installation from 0.8.2. Install 0.9.0 manually once; later
    versions update automatically. Windows automatic updates require the
    current-user Setup installation; the ZIP package only supports manual
    updates. The Mac app must be in a location the current user can write to.
  • Signing. RawLab is free and does not use Apple notarization or Windows
    Authenticode. macOS uses a fixed self-signed certificate plus Sparkle EdDSA;
    Windows trusts only a signed release manifest that binds the package size and
    SHA-256. The first launch may need System Settings → Privacy & Security →
    Open Anyway
    on macOS or More info → Run anyway in Windows SmartScreen.
    The Mac signing identity changes from 0.8.2, so macOS may ask for folder
    access again.
  • AI requirements. Core ML on Apple Silicon with macOS 14 or later, or a
    hardware DirectML/D3D12 GPU on Windows x64. There is no CPU AI mode, and Auto
    does not fall back to CPU neural inference. Intel Macs, iOS and Android do not
    include AI denoising in this release. The models increase download size.
  • Model licenses. DualDn weights and converted models remain under
    CC BY-NC 4.0; the
    Restormer generic RGB model keeps its upstream Academic Public License. Both
    are for noncommercial use only and are not relicensed under RawLab's GPL.
    See licensing.
  • Platform requirements remain Windows 10/11 x64, macOS 14+, Android 8+ and
    iOS 18+. iOS distribution still requires valid signing and provisioning.

Downloads

Download packages and checksums from the
v0.9.0 release.
Final file names, sizes and checksums are included in SHA256SUMS.txt and BUILD-PROVENANCE.json.

File Requirements and signing
RawLab-Mac-0.9.0-arm64.zip Apple Silicon, macOS 14+; self-signed and EdDSA-signed, not notarized; includes AI denoising
RawLab-Mac-0.9.0-x86_64.zip Intel, macOS 14+; self-signed and EdDSA-signed, not notarized; no AI denoising or models
RawLab.Desktop-stable-Setup.exe Windows 10/11 x64 current-user installer; required for automatic updates
RawLab-Windows-0.9.0-x64.zip Windows 10/11 x64 portable package; manual updates only
RawLab.Desktop-0.9.0-stable-full.nupkg Update package referenced by the signed Windows manifest
appcast-arm64.xml / appcast-x86_64.xml / windows-stable.json Signed desktop update metadata
RawLab-Android-0.9.0.apk Android 8+ ARM64/x86_64; existing release signing identity
RawLab-iOS-0.9.0-unsigned.ipa Unsigned device-arm64 build; valid user signing/provisioning required
Source archives, SHA256SUMS.txt / BUILD-PROVENANCE.json Tagged source, patched dependencies, asset hashes and build identities

See installation and the
desktop release procedure.

Verification And Packaging

The release includes the two Mac updater fixes: normal quit retains failed-update recovery data, and an EdDSA-valid package with the wrong code certificate is refused before starting the installer. Real isolated upgrade/restart/recovery acceptance passed on Apple Silicon and Windows; the Intel package was exercised under Rosetta. Native update controls and recovery/reopen regressions passed.

Known coverage limits remain: physical Intel/minimum-macOS testing, fresh-quarantine Gatekeeper/App Management and TCC continuity; current-build Windows hardware GPU/native-input coverage; Android API 26/physical-device memory and iOS device signing/provisioning; remaining accessibility/interaction cases. These are unverified areas, not successful tests. German UI issues at the largest accessibility text sizes are a known non-blocking limitation on all platforms.

macOS package identity verification downloads the update archive a second time. iOS IPA is unsigned and requires valid user signing/provisioning; it is not an App Store or TestFlight distribution.

Build identities, CI results, signed-package verification and actual desktop update/recovery evidence are recorded in BUILD-PROVENANCE.json.

The concise Chinese announcement is in
the Xiaohongshu update notes.