RELEASE_NOTES.md
Rama master · réplica de lectura del repositorio original.
We're proud to announce the new feature release of darktable, 5.8.0!
The github release is here: [https://github.com/darktable-org/darktable/releases/tag/release-5.8.0](https://github.com/darktable-org/darktable/releases/tag/release-5.8.0).
To build from source, do not use the autogenerated tarball provided by GitHub on the release page, download our tar.xz file instead. If you're just building for yourself without creating a package for some distribution, then using source code cloning in git is an even more convenient way.
The checksums are:
```
$ sha256sum darktable-5.8.0.tar.xz
darktable-5.8.0.tar.xz
$ sha256sum darktable-5.8.0-x86_64.AppImage
Darktable-5.8.0-x86_64.AppImage
$ sha256sum darktable-5.8.0-aarch64.AppImage
Darktable-5.8.0-aarch64.AppImage
$ sha256sum darktable-5.8.0-x86_64.dmg
darktable-5.8.0-x86_64.dmg
$ sha256sum darktable-5.8.0-arm64.dmg
darktable-5.8.0-arm64.dmg
$ sha256sum darktable-5.8.0-win64.exe
darktable-5.8.0-win64.exe
$ sha256sum darktable-5.8.0-woa64.exe
darktable-5.8.0-woa64.exe
```
When updating from the stable 5.4 series, please bear in mind that your edits will be preserved during this process, but the new library and configuration will no longer be usable with 5.4.
You are strongly advised to take a backup first.
#### Important note: to make sure that darktable can keep on supporting the raw file format for your camera, _please_ read [this post](https://discuss.pixls.us/t/raw-samples-wanted/5420?u=lebedevri) on how/what raw samples you can contribute to ensure that we have the _full_ raw sample set for your camera under CC0 license!
Since darktable 5.6:
- ??? commits to darktable+rawspeed
- ??? pull requests handled
- ??? issues closed
_Please note that the darktable documentation is not currently complete for release 5.8
and contributions are greatly appreciated. Please see the
[project documentation](https://github.com/darktable-org/dtdocs#contributing)
for more information on how to contribute._
## The Big Ones
The following is a summary of the main features added to darktable
5.8. Please see the user manual for more details of the individual
changes (where available).
- A new tone mapper for analog film simulation based on the
[spektrafilm](https://github.com/andreavolpato/spektrafilm) project
by Andrea Volpato. The module models the physical photographic
process from film exposure and development through optical printing
and scanning, rather than applying a conventional mathematical tone
curve. It reproduces the characteristic tonal and color response
resulting from these stages, including the compression of highlights
and shadows and the gradual color changes that occur as film reaches
its exposure limits. The simulation also includes film grain,
halation, and diffusion to reproduce characteristic properties of
analog film and optical printing.
- Automatic alignment of exposure brackets in "merge HDR" function,
removing ghosting and softness caused by handheld or shaky-tripod
brackets. Enabled by default in builds with OpenCV, and tunable
under preferences > processing > HDR alignment.
- An MCP server for darktable. The new darktable-mcp binary, a
headless sibling to darktable-cli, lets an AI agent drive the raw
pipeline over the Model Context Protocol: list modules and inspect
their parameter schemas, develop an image through a given module
stack and get back a preview or per-channel statistics, and read
history, styles and metadata from the library. Parameters are
addressed by name through darktable's own introspection rather than
by fixed offsets, so a stack stays valid across module versions, and
renders run on a throwaway duplicate so the source image is never
modified.
- Printing for Windows has been introduced. This activates the print
view with feature parity for what Linux and MacOS users have had.
Printing should work on all currently supported versions of Windows,
leveraging the XPS Print API. On the first print the print settings
dialog will open so that the user can confirm settings not exposed
in the darktable UI.
- New module saturation curve: remaps saturation and brilliance via
two independent curves plotted against each pixel's own perceptually
normalized saturation (darktable UCS or JzAzBz): a scene-referred
alternative to colorzones' chroma-vs-chroma curve, with input
normalized to the local gamut boundary.
- New module contrast and texture. Scene-referred control over local
contrast and the general contrast of the image. Useful for adjusting
clarity, finer textures, and managing high dynamic range images with
control over highlights and shadows. Uses edge aware and exposure
invariant guided filters for its processing.
- A new masks panel replaces the separate drawn, parametric, raster
and combined mask modes, and the mask manager. A module's mask is
now built in one place from groups of elements: drawn shapes,
parametric channels, raster masks from other modules and AI objects
can be freely combined and nested. Each group combines its elements
with one operator and has its own opacity, inversion and refinement
(details, feathering, blur, contrast and brightness), which each
element can also have on its own. Elements and groups can be
renamed, reordered by drag and drop, disabled, soloed, and a mask
can be locked against reset, presets, styles and paste. The panel
and the canvas follow each other: hovering or selecting an element
in either shows it in the other. The panel can stay in the module's
blending section, move to a utility module, or be docked beside the
image, and built-in layout presets set up common masks in one click,
with notes on how to use them. Shortcuts assigned to the mask
manager now drive the same controls in the panel. Existing masks are
converted when an edit is opened and render as before.
- DNG files carrying an OpcodeList3 GainMap, such as those written
by DJI drones, now have it applied during demosaicing. This
corrects the lens shading the camera recorded but left in the raw
data. Existing edits are left unchanged; new edits get it by
default.
## UI/UX Improvements
- Added back and forward navigation in collections, with user-assignable
shortcuts. Navigation remembers collection rules and thumbnail scroll
positions for up to 32 visits during the current session.
- The code has received a large set of changes in preparation of the
Gtk4 migration.
- The code for handling popover menus has been changed.
- Checkboxes are now Bauhaus widgets and are reset to default values
when e.g. a tab or a module is reset.
- Mask shapes on the canvas now have a contrasting edge, so they stay
visible over dark and bright areas alike, and the hovered or
selected shape is drawn heavier.
- Escape now cancels a mask shape that is being drawn, as a right-click
does.
- Mask nodes close or outside the image boundaries are now editable.
When "edit shapes on canvas" is enabled, the node editing area is
extended beyond the image borders to allow editing of these nodes.
- Raster masks now show up on the canvas. When "edit shapes on canvas"
is enabled, hovering a raster mask in the masks panel tints the area
of the image that it covers.
- In color assessment mode, when the image is zoomed in show a
dashed line on the sides of the white border where the image
extends beyond the current view (horizontally or vertically).
- Added support for XMP drag&drop directly in the darkroom. Single
XMPs can be used to create duplicates or replace the current
edit. Multiple XMPs create one duplicate per file.
- Lighttable footer star ratings are now individual CSS-targetable
buttons (instead of a single drawing area), so spacing and sizing
can be themed. The rating filter reuses the same star paoint helper.
- In the borders (framing) module, disable frame line controls when
the border settings would prevent the frame line from being visible.
- Added an interactive editing mode allowing users to correct a color
by selecting it directly on the photo with the mouse wheel, just as
was already possible with the "tone equalizer" module.
- Grey theme: The filmstrip is now slightly darker to improve visual
distinction between unselected/selected images.
- Neural restore now takes a full output path per task instead of an
output folder, so the output file name and its suffix can be chosen.
The extension still comes from the task.
- Some safety checks when switching to the quick access panel were
added.
- On the color equalizer module a new collapside section with the
color sliders has been added.
- The list of color checker charts in the color calibration module now
has one entry for the Datacolor SpyderCheckr 24 and one for the
SpyderCheckr 48. As far as we know, there never was a separate 2018
version of these charts.
- Added a "copy to clipboard" export storage, which places the
exported images on the system clipboard.
- A workspace button in the top toolbar, between the keyboard shortcuts
button and the preferences button, opens a menu to switch workspaces
or create a new workspace (blank by default + optional copy). Workspace
button only appears when "allow for multiple workspaces" is enabled.
Switching or creating relaunches darktable into the chosen workspace.
- The lens correction module now warns when it applies Lensfun
vignetting correction to a Panasonic RW2 or Olympus ORF file shot
with the camera's shading compensation on, which can overcorrect the
corners.
- In the "select parts to copy" and "select parts to paste" history
dialogs, the module order is now a check box below the item list
instead of a last entry in the list itself.
- In the AI object mask tool, every object you click on is kept in the
selection, including one that doesn't touch the others. This needs a
model that returns separate objects, such as SAM; SegNext selects
one region at a time.
- In tethering mode, the tethering info bar (located above the live
camera view) now displays the camera model.
- Pressing ESC or clicking on the dialog X icon (top-right) now exits
Darktable instead of launching the previous workspace.
- The settings of the path mask "shrink or grow" (the step per scroll
tick, the unit, the tracing resolution and the curve smoothing) are
now in a popover opened by a gear on the slider, which no longer has
a unit toggle. The options for the scroll direction of mask
parameters and for the mask indicator in module headers moved from
the preferences dialog to the masks panel's options popover, and the
masking section of the preferences dialog is gone.
- Color calibration has a new illuminant option, "from the white
balance module", which derives the illuminant from the settings of
the white balance module. With it, the color picker and the
temperature, tint, and channel coefficient sliders in the white
balance module set the white balance that color calibration uses.
The new option is the default for new scene-referred edits of color
raw files. It needs the white balance module's new "prepare data for
color calibration" checkbox to be checked, as it is by default in
these edits. The checkbox is available in all modes except "camera
reference", which does not need it. The old "as shot to reference"
mode becomes "as shot" with the checkbox checked. While color
calibration adapts with any other illuminant option, such as "as shot
in camera" or "custom", the white balance module disables these
controls and its camera presets, because the white balance is then
controlled in color calibration.
- The recent collections module now handles single or double click
according to the preference setting for collections.
## Performance Improvements
- New edits convert the input color profile directly to the working RGB
profile, avoiding an unnecessary round trip through Lab. Existing edits
keep their previous color processing.
- The lighttable timeline no longer slows down quadratically with the
number of collected images: with a 100,000-image collection it took
about 15 minutes to draw, now a few seconds.
- Replaced quadratic XMP history writes with a linear algorithm.
- Don't invalidate the pixelpipe cache on every commit
when a raster mask is used.
- Don't calculate unused rastermasks while exporting.
- Speed up editing an image that uses a detail mask. Every history
change used to discard the cached output of every module from
demosaic onwards, so adjusting a mask or toggling the mask overlay
recomputed most of the pipeline each time.
- Reduced memory requirements of the mask feathering guided filter
- Separated the demosaicer's crop-and-scale step into a new hidden
module to improve cache efficiency and UI responsiveness.
- Add a OpenCL code path to tone equalizer which gives around 6-20x
faster processing times in comparison to CPU code path, depending on
CPU & GPU.
- Contrast & texture now runs on OpenCL, so it no longer moves the
image back to the CPU in the middle of an otherwise GPU pipeline.
- Highlight reconstruction's inpaint opposed method now runs on OpenCL
for linear raws, such as sRAW and linear DNG files.
- Enlarge canvas now runs on OpenCL, so it no longer moves the image
back to the CPU in the middle of an otherwise GPU pipeline.
- Chromatic aberrations RGB now runs on OpenCL.
- Grain now runs on OpenCL.
- Surface blur now runs on OpenCL. It was by far the slowest module
left on the CPU in an OpenCL pipeline, and moved the image back to
the CPU for it.
- Rotate pixels, which straightens the sensor layout of Fujifilm
SuperCCD raws, now runs on OpenCL.
- Scale pixels, which sets the pixel aspect ratio for non-square-pixel
sensors and anamorphic desqueeze, now runs on OpenCL.
- Hovering over and dragging nodes of long drawn brush strokes and
paths is much faster, as the mask overlay is no longer redrawn on
every mouse move and long outlines are drawn with far fewer points.
- Drawn masks render faster, most of all masks with nested groups or
shapes reused by several modules: each group no longer clears and
combines a full-size buffer just to copy in its first shape, and
folds each shape only over the part of the image it covers.
- Lens correction with lensfun is faster on the OpenCL path: lensfun
is now asked for its corrections on a coarse grid instead of for
every pixel, so the CPU no longer computes and uploads hundreds of
megabytes on each run. The correction is also more precise on x86,
where it could be up to 0.3 px off, or 0.6 px when distorting.
- The raw denoise module can now run on the OpenCL device for Bayer
sensors.
- The retouch module's heal tool now runs its solver on the OpenCL
device instead of copying the image back to the CPU for each
healed shape.
- The hot pixels module can now run on the OpenCL device, so a raw
no longer goes back to the CPU and up again around it.
- Module color pickers now compute their statistics on the OpenCL
device instead of copying the picked area back to the CPU.
- darktable-cli uses less memory when exporting many images: it no
longer keeps the decoded raw of every exported image in memory
until its cache is full.
- darktable keeps fewer decoded raws in memory: the full-size image
cache now holds one image per worker thread plus one, instead of
twice the number of worker threads.
- Modules showing a histogram of their input (levels, rgb levels, rgb
curve, tone curve and color zones) now count it on the OpenCL device
instead of copying the image back to the CPU.
- The raw chromatic aberrations module can now run on the OpenCL
device.
- Haze removal no longer copies the whole image back from the OpenCL
device to estimate the ambient light.
- The dither module's random mode can now run on the OpenCL device.
- The AMaZE demosaicer, plain and dual, can now run on the OpenCL
device instead of sending the image back to the CPU for the
demosaic.
- Copying images to the OpenCL device is up to four times faster on
NVIDIA and macOS drivers, which speeds up every step where the
pipeline moves the image from the CPU to the GPU.
- Softproofing, gamut check and LUT-based ICC profiles in input and
output color profile can now run on the OpenCL device, through a 3D
LUT sampled from LittleCMS, instead of sending the image back to the
CPU. The "always use LittleCMS 2" export preference still applies the
exact transform on the CPU.
- The LMMSE demosaicer can now run on the OpenCL device instead of
sending the image back to the CPU for the demosaic.
## Other Changes
- AgX and sigmoid now adapt to HDR output, so one edit exports to
SDR, HLG and PQ alike. HLG output peaks at 1000 nits, and PQ output
at the new "PQ peak" setting of the export module, 1000 nits by
default, which export presets store and darktable-cli sets with
`--pq-peak`. With AgX, shadows and midtones keep their SDR brightness
and highlights extend to the peak, following Blender's AgX HDR, with
a new "HDR purity" slider for the highlights' saturation. Sigmoid
scales its target white to the peak, up to 1600% (3248 nits). SDR
output is unchanged. Existing edits render as before, and a "follow
output peak" toggle in each module opts them in.
- PNG exports using a PQ output color profile now include content
light level metadata (MaxCLL and MaxFALL) for HDR tone mapping.
- A new `.dtdata` sidecar next to the XMP holds per-pixel edit data
such as raster masks. It is only created for images that use such
data, and only when sidecar writing is enabled; most images never
get one. It follows the image when it is moved, renamed, duplicated
or deleted.
- HEIF exports using a PQ output color profile now include HDR10
content light level metadata (MaxCLL and MaxFALL) when built with
libheif 1.15.0 or newer.
- AVIF exports using a PQ output color profile now include HDR10
content light level metadata (MaxCLL and MaxFALL) when built with
libavif 1.0.0 or newer.
- JPEG XL exports using a PQ output color profile now record the image's
measured peak luminance, so viewers no longer tone map them as if they
reached 10000 nits.
- The external raster masks module imports its mask into the image's
`.dtdata` sidecar instead of referencing a file in a root folder.
Each instance holds one imported mask. Existing edits that point at
a folder keep working, and the folder chooser remains available when
sidecar writing is set to "never".
- AI object masks are now stored as pixels in the image's `.dtdata`
sidecar, exactly as selected, instead of being traced into paths.
This replaces the shift+right-click PNG export that fed the external
raster masks module, which is removed; the module itself is
unchanged. Click an object's icon to refine it. "apply as paths" in
the mask manager keeps the traced output; it is always on when
sidecar writing is "never". When an object's stored mask is missing,
the module it masks applies nowhere and the loss is reported;
clicking the object's icon regenerates it when sidecar writing is
enabled.
- Added a new collection filter for the original image dimensions.
- Support for Canon's Highlight Tone Priority.
- AI models can now be installed from additional repositories. Models
installed this way remember where they came from and are included in
the update check.
- Added a preference to disable automatic AI model update checks,
with a manual check button for when it is off.
- Added a compression option for the TIFF files written by neural
restore denoise and upscale, matching the TIFF export module.
- Model packages can now declare their ONNX Runtime graph optimization
level, so models that misbehave under aggressive optimization no
longer need workarounds in feature code.
- Added support for DNG files providing a forward matrix (FM) and
'ProfileHueSatMapData'.
Two new input profiles in colorin can be selected if any of these
tags are available.
- 'DNG forward matrix' is available if a FM is provided
- 'DNG embedded look' is available if there is no FM but
'ProfileHueSatMapData' is provided.
If one of these new input profiles is selected, the pipe gets its
data corrected according to DNG specs matching the intended look.
- The highlights module will be disabled automatically when a raw
image with a clean history is opended in darkroom the first time
and the opposed algorithm did not find clipped photosites.
- Panasonic RW2 files now offer embedded lens distortion correction,
covering built-in and Lumix-branded lenses that lensfun has no
profile for.
- Printer (paper) profiled print jobs will remain 16 bit through
the full print pipeline.
- When installing development versions of darktable (snapshots, self
compiled etc.) the Windows installer now allows to setup a custom
configuration-directory and custom shortcut-name, useful for
multiple parallel darktable installations.
- Added a control in the bottom panel of the lighttable to reset the
star rating of images.
- The pixelpipe dump files requested via cli switches are now written
in ppm or pgm format.
- The OpenCL configs had a bump to v7 and were simplified for user edits.
For a device "XXXX" we now have 3 configs:
cldevice_v7_XXXX=events:on asyncmode:off device:on unifraction: 0.250
For events, asyncmode and device it can be on/off.
(enable a disabled device (marked as device:off) by editing to device:on)
The unifraction is a float and can be chosen by the user in the 0.02-0.5 range
cldevice_v7_XXXX_id0=headroom: 600
You can modify the headroom if "tunehead" is available
cldevice_v7_XXXX_nocl=
List of modules that will not process it's OpenCL code.
OpenCL fast/default kernels are cached at different locations.
- The aspect ratio chosen on the camera is now applied as a crop when
the raw was left uncropped, so a frame shot at 1:1 or 16:9 opens
framed as intended while the full sensor area stays available to
reframe within. Read from Canon and Olympus raws.
- Export filename patterns gained `$(CATEGORY_EACH[n,category])`,
which yields one path per matching tag instead of a single
comma-joined value. Referencing several levels of the same category
walks each tag's path, so an image tagged for several people can be
exported once into each person's folder rather than into one
"John,Jane" folder.
- Import filename patterns gained `$(CONFLICT[n])`, a counter that is
empty unless the file name is already taken, so burst shots named by
capture time no longer clash.
- Added `--print-paths` and `--print-paths-as-flags` to darktable
and darktable-cli. These print the resolved configdir, cachedir,
tmpdir, datadir, moduledir, localedir and library paths for the
current install (honoring path overrides), then exit. The
`-as-flags` form emits a single line of `--flag value` pairs
ready to splice into another invocation.
- Lightroom import now applies a crop with the crop module, and its
straightening angle with the rotate and perspective module, instead
of the deprecated crop and rotate module.
- Lightroom import now sets the working profile to linear ProPhoto
RGB, the space Lightroom works in, instead of linear Rec709 RGB.
- In the darkroom in the left panel the navigation can now be toggled
through the panel's right click menu in addition to the strg+shift+n
shortcut.
- Lightroom import adds its exposure and blacks adjustments to
darktable's default exposure instead of replacing it. In the
scene-referred workflow, images with a Lightroom exposure adjustment
were imported without the default exposure boost and exposure bias
compensation, so a small positive adjustment rendered darker than
none at all.
- Lightroom import creates tone curves in RGB mode with color
preservation, as darktable's tone curve module now defaults to, and
no longer enables the module for a curve that changes nothing.
Clarity is imported with the local laplacian filter of the local
contrast module instead of the older bilateral grid.
- Lightroom import applies spot removals with the retouch module,
cloning or healing as each spot did in Lightroom, instead of the
deprecated spot removal module.
- History items created by the Lightroom import are labeled "lr import"
instead of having an empty label.
- Raw files that RawSpeed can't decode are now loaded with LibRaw if
LibRaw supports their compression mode: Fujifilm lossy compressed
RAFs, and with LibRaw 0.23 (snapshot 202611) or newer also Nikon high
efficiency (HE/HE*) NEFs from the Z6 III, Z8, Z9 and Z f and Sony
ARW 6.0 compressed and compressed (HQ) files. Bayer DNGs RawSpeed can't
decode are loaded with LibRaw as well, e.g. lossless JPEG with
unsupported predictors or uncompressed files with inconsistent strip
sizes as written by some DJI drones. All other files are still loaded with RawSpeed.
- The LibRaw loader now supports X-Trans sensors. Previously, RAFs
routed to LibRaw via the libraw_extensions setting came out red.
## Bug Fixes
- Fixed faint seams on a 112 pixel grid in images demosaiced with
LMMSE, where each tile read leftover data from the previous tile.
- Upscaled exports no longer show shifted bands where they were
processed in tiles, and the upscale now runs on OpenCL when it fits
the device.
- Fixed Lightroom develop import which was not done. Only the tags,
color labels, stars where imported.
- Fixed missing Active D-Lighting exposure compensation for Nikon
images shot in sRGB or Adobe RGB.
- Fixed drawn masks appearing too early in the history stack of an
image that got a style or a pasted history before it was first
opened: going back in the history showed masks from before the step
that added them.
- Fixed darktable hanging on exit, with its window already gone, after a
raw file failed to load during the session. The lock taken while
reading the file was not released when the read failed, so every later
raw open waited for it forever.
- Fixed AgX rendering everything above the pivot black when pivot
target output was 100% and shoulder power was 7 or more.
- Fixed shift-click on the selected entry of the history stack, which
did nothing instead of revealing that module in the right-hand panel.
- Fixed overly dark PQ and HLG image imports and overly bright exports,
including HDR PNG and JPEG XL profile detection.
- Fixed a possible crash when opening an image in the darkroom after
leaving it, caused by white balance data still pointing at the
previous image's modules.
- Fixed the feather of paths created by vectorizing a raster mask or an
AI object, which could not be changed from the mask manager.
- Fixed a dark fringe along the right and bottom edges of images
pixelated by the censorize module, whose last column and row were
left unpainted.
- Fixed color reconstruction giving different colors with OpenCL than
without it: the OpenCL grid only counted one pixel of each run of
neighbors, and zooming in past 100% changed its spatial extent.
- Fixed HEIF export compilation with libheif versions older than 1.15.0,
and with versions older than 1.16.0 when libsharpyuv is available.
- Do not convert the pipe input in place for blending, which may result
in a corrputed buffer.
- Fixed an OpenCL bug causing Lab colorspace overlays showing as solid
colors.
- Fixed a trashing error dialog when deleting a virgin duplicate of an
image while sidecar creation is set to "after edit".
- Fixed color harmonizer's auto-detect harmony applying the hue
histogram of the previously viewed image after switching images in the
darkroom.
- Entering the darkroom no longer expands one instance of a module
while a different instance of it holds the focus. The instance that
gets the focus is now the one that is expanded.
- Clarified multi-image rating toasts for un-reject and for mixed
upgrade/downgrade results across the selection.
- Fix blank module instance names when going back in darkroom action
history.
- Do not override library edit history from XMP sidecar file when
using darktable-cli with a library file.
- Fixed collection module not updated properly on preset load.
- Fixed a crash at the end of neural restore's raw denoise on certain
sensor sizes, where blending the tile seams wrote past the edge of
the image.
- Fixed an empty Media Type dropdown in the Print module for Canon
printers whose drivers use the `CNIJMediaType` option instead of the
standard `MediaType` option.
- Fixed lighttable bottom-panel star rating and color label toggles
not responding on repeated clicks.
- Fixed the feather on dense drawn path masks rendering stripes,
phantom arcs and crossing lines.
- Fixed the feather dipping inside the shape on paths with narrow
spikes.
- Fixed AI segmentation masks being offset by up to several pixels
from the object they were generated for.
- Fixed the AI raw denoise preview showing identical before and after
images on 4BAYER (CYGM/RGBE) raws, which now shows a warning.
- Fixed auto-applied denoise presets running a second time on images
produced by AI raw denoise.
- Fixed paths and nodes occasionally snapping to the top left
corner during move operations.
- Fixed crash when navigating images in the darkroom with an
image pinned in the 2nd window.
- Fixed darkroom watermark issues related to tag & metadata changes.
- Fixed refinements to forwarded raster masks being ignored along
the OpenCL path.
- Windows installer: Added an option to enable/disable including
darktable in the explorers context menu.
- Compute JzAzBz's Az and Bz without cancellation so that results
computed by the CPU and OpenCL paths do not diverge.
- Fixed mask-only edits not causing thumbnail refreshes and XMP
updates (when enabled).
- Fixed zoom position moving around when switching between images.
- A path that grows past the border of the image is no longer clipped.
- Fixed broken dragging in sliders' precise-entry mode
in GTK3.
- Fixed the feather of a drawn path being lost when the shape was
resized with the scroll wheel.
- Fixed resizing a path mask with the scroll wheel or the "shrink or
grow" slider bending its curves into straight chords and giving it a
different set of nodes at every step. Its nodes now follow the
outline as it grows or shrinks.
- Fixed numeric error in the framing module (borders) causing the
frame line to be off-center for certain border sizes.
- Fixed a crash or hang on Windows when checking a faulty custom ONNX
Runtime library.
- The "condensed panels' controls" preference now also condenses the
sliders and comboboxes themselves, instead of only shrinking their
font.
- Fixed a small memory leak each time a history stack was pasted onto
the image open in darkroom.
- Fixed Windows paths losing their backslashes in export and import
patterns, which sent files to the wrong location.
- Fixed wrong output or a crash from an export pattern containing an
unclosed variable substitution, such as "$(FILE_NAME/foo".
- Fixed corrupted output or a crash when an AI model returns more data
than darktable reserved for it, affecting object masks and Lua
models.
- Fixed snapshots being applied onto the original image instead of the
current image.
- Fixed a crash when importing a style whose module order is
empty. The malformed order is now ignored and the style keeps the
default one.
- Fixed automatic exposure rendering a black image when its raw
histogram was unavailable, and black or inverted images when the
black level was set too high for the exposure applied.
- Fixed exposure's area mapping blowing out the image when its target
lightness was set to zero. Such a target cannot be reached, so the
correction is now left alone instead.
- Fixed a possible crash when loading an image edited with a newer
version of Darktable. This scenario is not supported but we should
not crash. The offending modules are now reset to use the default
parameters and a message is displayed to the user.
- Fixed auto calculation of capture radius and jumping of capture
radius sliders in UI.
- Fixed small memory leaks when expanding variables, which grew with
the number of images exported or imported in one run.
- Fixed the splash screen text jumping sideways while startup messages
change, most visibly on macOS.
- Fixed camera mutex locks before the mutexes are initialized.
- Fixed tether mode leaving its event and live-view threads running after
leaving the view or tearing down a camera, and used full gphoto teardown
when a mounted camera disappeared so the device is released cleanly.
- Fixed tether mode treating transient PTP Device Busy errors as a lost
camera connection (common when enabling live view), and stopped
echoing camera-driven property updates back to the camera.
- Fixed metadata tag Xmp.xmpMM.PreservedFileName not written on camera
import.
- Fixed keyboard shortcuts in the darkroom doing nothing until the
center view was clicked, after a snapshot had been taken while the
snapshots module was collapsed.
- Fixed a crash on using tag floating window after editing metadata.
- Fixed rotate and perspective sometimes not applying its automatic
crop when the rotation was changed (e.g. with a shortcut) before the
module had been enabled, leaving empty corners in the image.
- Corrected the reference colors of the Datacolor SpyderCheckr Photo
in the color calibration module. Existing calibrations are not
updated. Profile the chart again for a more accurate calibration.
- Fixed the color calibration module swapping the white and middle
gray reference patches on the Datacolor SpyderCheckr Photo.
- Fixed the color calibration module not using the gray patch closest to
middle gray on the Datacolor SpyderCheckr 48 and Photo.
- Fixed a thin line of pixels along the outer edge of a gradient mask
that the module's effect was applied to in reverse, most visible on
gradients with a sharp transition.
- Fixed various interpolating issues at top & lest image border.
- Fixed a bug in OpenCL demosaicer leading to slighly increased chroma
differences compared with CPU code.
- Fixed XMP sidecar files not being written for images whose creator,
description or copyright Exif fields (e.g. Artist, Canon OwnerName)
contain non-ASCII characters such as umlauts stored by the camera in
a legacy encoding. These fields are now converted to UTF-8 on
import. Use "refresh EXIF" on already imported images to fix them.
- Fixed rotate and perspective occasionally computing its automatic
crop from a mismatched preview size, which could leave the crop box
slightly wrong until the module was adjusted again.
- Fixed a crash on Wayland when disconnecting a monitor.
- Fixed raw denoise on X-Trans images giving slightly different
results depending on the number of CPU threads, and occasionally
wrong colors near the left or right image edge.
- Copying a whole history stack no longer carries over the module order
that was selected in an earlier "copy parts of the history" dialog.
- Fixed the lens correction module reporting distortion correction as
applied from embedded metadata when the file only carries chromatic
aberration correction data.
- Fixed raw denoise on X-Trans images replacing the green pixels in
the rightmost column with the value of a neighboring pixel, often a
red one, which slightly changed the colors along the right edge of
the image.
- Fixed raw denoise using pixels of the wrong color near the top
corners of some X-Trans images.
- Fixed raw denoise on X-Trans images giving different results
depending on white balance's "prepare data for color calibration"
setting: with highlight reconstruction in inpaint opposed or
segmentation mode, clipped highlights were rebuilt with a white
balance correction meant for white-balanced data, although raw
denoise runs without white balance.
- Fixed possible crashes after removing an image that had snapshots
from the library, while either a snapshot was shown for comparison
or all ten snapshot slots were in use.
- Fixed the workspace selection dialog growing past the screen when
many workspaces are present; the list is now capped and scrollable.
- Fixed a snapshot showing edits from another snapshot, even one of a
different image. Removing an image that had snapshots from the
library, then taking a new snapshot, could cause the mix-up.
- Fixed small memory leaks in the lighttable that grew with each rating,
color label, tag or duplicate applied, and each time the thumbnails or
the range filters were refreshed.
- Fixed edits made just before quitting darktable, such as a rating,
being missing from the image's XMP sidecar while the library kept
them.
- Fixed tethered additional properties: user-added entries can be removed
or replaced (same label), the O property picker works more than once,
and unloadable saved properties show as deletable unavailable stubs.
- Fixed an unexpected image change in AgX when disabling 'keep the pivot
on the diagonal' after using 'read exposure from metadata and exposure
module'.
- Fixed the filmstrip cursor triangle ignoring CSS. Its color again
follows `#thumb-cursor`; the default is unchanged (`@bg_color`).
- Fixed modules with mask feathering guided by the input image falling
back to the CPU when processed with OpenCL, whenever the module
needs a larger input area than it outputs, for example liquify in a
zoomed-in darkroom view.
- Fixed the darkroom filmstrip losing the active-image triangle after
that thumbnail is scrolled off screen and back.
- Improved internal stability (more reproducible wavelet denoising and
haze removal).
- Fixed green tints and color shifts along image edges when using raw
denoise on X-Trans images, including some downscaled exports.
- Fixed shortcuts firing when pressed together with modifiers they
were not assigned with: ctrl+K, for example, no longer runs the
action bound to K, unless a fallback gives the modifier a meaning
for it.
- Fixed small memory leaks when hovering over unselected images in the
lighttable, and when importing images with a new camera maker, model
or lens.
- Fixed darktable-cli crashing when `--threads` was set below the
number of CPUs.
- Fixed memory leaks when processing images.
- Fixed memory leaks when exporting images or the tag list.
- Fixed month collection filter layout issues.
- Fixed capture time based collection filters using the whole time range
of the library, instead of the range of the current active collection.
- Fixed GPU memory leaks while displaying the highlight reconstruction
mask of filmic rgb, and possible crashes in basic adjustments, rgb
levels and base curve when OpenCL memory allocation fails.
- Fixed small memory leaks that grew while a camera was connected, and
when filtering collections by inverse values (such as exposure time),
using the date range popup, importing localized shortcuts, or
creating and editing styles.
- Fixed a possible crash while checking database snapshots for removal
when the folder of the data database could not be determined.
- Fixed a GPU memory leak in the OpenCL path of the clipping indicator.
- Fixed an issue where the Windows installer wouldn't run without user
input when requested.
- Fixed the right-hand panel jumping when switching tabs in modules
taller than the panel, such as AgX or spektrafilm: clicking a tab
scrolled the panel until the tab row was at the top.
- Fixed retouch shapes and drawn masks landing in the wrong place when
the scale pixels module was enabled: they were shrunk toward the
top-left corner of the image instead of being stretched with it.
- Fixed black output from filmic rgb's v1 and v3 color science, when
no work profile is used for the luminance, and wrong alpha from its
gaussian and poissonian noise, on macOS OpenCL: the kernels read
vector lanes they had not set.
- Fixed the vignetting module using a linear falloff instead of
its smooth one on the CPU when dithering is off, so that CPU and
OpenCL now give the same result.
- Fixed Lightroom imports of parametric tone curves, which washed out
the highlights, of split toning, which used the wrong hues and toned
the highlights across nearly the whole image but never the shadows,
and of post-crop vignetting, which was far weaker than in Lightroom
and only started near the frame's edge.
- Fixed the shape of post-crop vignetting imported from Lightroom on
portrait images, straightened crops and wide frames. Portrait images
got a landscape vignette that darkened their top and bottom but not
their sides.
- Fixed color calibration's saturation sliders having a different
effect with OpenCL than on the CPU when the "version 1 (2020)"
saturation algorithm is selected.
- Fixed the macOS app loading GTK input method modules, and with them
a second copy of GTK, from Homebrew when Homebrew's GTK is installed,
instead of the copies bundled in the app. Without Homebrew, the
modules were not found at all.
- Fixed the wrong raster mask being applied when three or more external
raster masks were used in one edit.
- Fixed memory leaks of libgphoto2 event data, CameraFile objects, and
strdup'd filenames during tether polling and download.
- Fixed tether mode treating non-fatal PTP errors as a lost camera
connection; only clear I/O / claim / missing-device failures now force
remount and leave tethering.
- Fixed diffuse or sharpen giving a different result on the CPU than
with OpenCL whenever an anisotropy was set: the CPU path computed the
diffusion weights with a coarse approximation of the exponential.
Edits using anisotropy may now look slightly different when exported
without OpenCL.
- Fixed workspace discovery truncating names at the first '.' in the
database filename (for example library-client.v2.db was shown as
"client"), which could open or delete the wrong database. Workspace
names that contain path-unsafe characters are now rejected on create.
- Fixed the non-local means modes of denoise (profiled) and astrophoto
denoise giving a different result with OpenCL than on the CPU: the
OpenCL path gave the horizontal neighbors twice their weight. Edits
using them may look slightly different when exported with OpenCL.
- Fixed tethered bracket capture progress using the wrong image total.
- The non-local means modes of denoise (profiled) and astrophoto
denoise now treat the image edges the same way on the CPU and with
OpenCL, by mirroring the image. Before, the two gave different
results in a band along the edges. Edits using them may look
slightly different near the edges.
- Fixed PPG demosaic with an edge threshold giving different pixels
along the image edges with OpenCL than on the CPU.
- Fixed the RGB curve and tone curve modules giving brighter or darker
tones with OpenCL than on the CPU above a curve's last node, when that
node is left of the right edge.
- Fixed occasional magenta pixels on the CPU in red areas blended with
the chromaticity, hue, color or HSV blend modes in display RGB, where
OpenCL gave the right color.
- Fixed inpaint opposed highlight reconstruction on linear raws, such
as sRAW and linear DNG, taking the clipped areas of every channel
from the red channel alone.
- Fixed floating-point raw files with embedded DNG gain maps coming out
wrong with OpenCL. The gain maps of other raw files are now also
applied the same way with OpenCL as on the CPU.
- Fixed the contrast equalizer with an edge sharpness set, and the
soft clip of negadoctor, giving a different result on the CPU than
with OpenCL: the CPU path used a coarse approximation of the
exponential. Edits using them may now look slightly different when
exported without OpenCL.
- Fixed rotate pixels, which straightens the sensor layout of Fujifilm
SuperCCD raws, using the scaling interpolator instead of the warp
interpolator set in the preferences, as the other rotating and
distorting modules do.
- Fixed random dither giving a slightly different noise pattern
depending on the number of CPU threads. Its noise pattern changed,
so exports using it differ in the noise but not in its strength.
- Fixed stale color temperature and illuminant color in color calibration
after white balance changes when using "as shot in camera".
- Fixed input color profile and highlight reconstruction (opposed and
segmentation modes) still applying part of the white balance
correction when the white balance module, set to "as shot to
reference", was switched off on a raw image. Edits made that way
render with different colors.
- Fixed a color cast when white balance in "as shot to reference" mode
was pasted from another image, or applied through a style or preset:
white balance applied the pasted multipliers, but input color profile
completed the correction with the target image's own as-shot
multipliers.
- Fixed scale pixels output shifted by a fraction of a pixel in each
tile when the image was processed in tiles.
## Lua
### API Version
- API version is now 9.8.0
### New Features
- Extended dt_lua_snapshot_t and added a function to remove the snapshot.
- Extended Lua entry widgets with case-insensitive substring completion
and callbacks for text changes and activation.
- Added per-image storage for Lua scripts. `darktable.metadata.register(script, key)`
defines a metadata field named after the script, and `image:get_metadata(script, key)`
and `image:set_metadata(script, key, value)` read and write it. Values are kept in the
library database, removed together with the image, and left out of exported files
unless the script asks otherwise.
### Bug Fixes
- Fixed a crash when a script took more snapshots than the snapshots
module has slots. Once all slots are in use, `take_snapshot()` now
takes none and says so, as the take snapshot button already refused.
### New Scripts
- N/A
### Other Lua changes
- Button images are now decoded at the display's resolution and sized
to the interface font, so script icons are sharp on HiDPI screens
and no longer stuck at the image's own pixel size.
## Notes
- Masks are converted to the new masks panel's format when an edit is
opened, and written back in that format. The conversion is one way:
edits saved by this version cannot be read by older versions of
darktable with their masks intact. Take a backup before opening
edits you may still want to use with an older version.
- When exporting to AVIF, EXR, JPEG XL, or XCF, selecting specific
metadata (e.g. geo-tag or creator) is not currently possible. For
AVIF, EXR, JPEG XL, and XCF formats, darktable will not include any
metadata fields unless the user selects all of the checkboxes in the
export module's preference options.
- Starting with release 5.6, macOS versions older than 15.0 are no
longer supported on Apple Silicon Macs. Intel Macs are no longer supported.
- AI object masks and the new masks panel raise the mask format version
to 7. An image whose masks this release has written is not readable by
darktable 5.6 or earlier: it reports a mask version mismatch and drops
every mask on that image.
## Changed Dependencies
### Mandatory
- libarchive, previously only needed for AI model downloads, is now
required for the `.dtdata` sidecar
### Optional
- OpenCV 4 or 5 for HDR exposure-bracket auto-alignment (only the
core, imgproc and flann modules plus features2d/calib3d on OpenCV 4,
or features/geometry on OpenCV 5)
## Camera support, compared to 5.6
### Base Support
- Canon EOS R50 V (requires LibRaw 0.23 or newer)
- Canon EOS R6 Mark III (requires LibRaw 0.23 or newer)
- Canon PowerShot V1 (requires LibRaw 0.23 or newer)
### White Balance Presets
- N/A
### Noise Profiles
- N/A
### Missing Compression Mode Support
- Apple ProRAW DNGs
- CinemaDNG lossless (Blackmagic, some DJI, etc.) and lossy (Blackmagic)
- DNG 1.7 using JPEG XL (Adobe enhanced, Samsung Expert RAW)
- Nikon high efficiency NEFs, except from the Z6 III, Z8, Z9 and Z f
with LibRaw 0.23 or newer
- Phase One other than IIQ L
- Sony ARW 4.0/5.0 downsized lossless ("M" for full-frame, "S" for full-frame & APS-C)
- Sony ARW 6.0 compressed and compressed (HQ) with LibRaw older than 0.23
### Suspended Support
Support for the following cameras is suspended because no samples are available on https://raw.pixls.us:
- Creo/Leaf Aptus 22(LF3779)/Hasselblad H1
- Fujifilm IS-1
- Kodak EasyShare Z980
- Leaf Aptus-II 5(LI300059)/Mamiya 645 AFD
- Leaf Credo 60
- Leaf Credo 80
- Olympus SP320
- Phase One IQ250
- ST Micro STV680
## Translations
- ???