Changes since 8.18.6:

• tiffload: copy colormaps on page load [Ada Logics]
• buildlut: limit output lut size [Ada Logics]
• tiffload: check for undersized jp2k tiles [Ada Logics]
• hough_line: improve bounds check [Ada Logics]
• pdfiumload: check for buffer too small [Ada Logics]
• tiffload: check rgba settings between directories [Tanto Security]
• uhdrload: calculate gain map scale factor using round-to-nearest [lovell]
• jp2ksave: tag as UNTRUSTED [kleisauke]
• header: only parse EXIF metadata for blob values [Shopify]
• jp2kload: more size validation [Akokonunes]

Windows binaries here:

https://github.com/libvips/build-win64-mxe/releases/tag/v8.18.7

Changes since 8.18.5:

• openexrload: more validation [Hygge Halcyon]
• heifload: tag as UNTRUSTED for libheif before 1.23.2

Windows binaries here:

https://github.com/libvips/build-win64-mxe/releases/tag/v8.18.6

Changes since 8.18.4:

• maplut: fix possible OOB read with complex images [sage-mode-hunter]
• measure: verify image region is larger than patch count [lovell]
• matload: more specific Matlab detection
• reduceh: fix potential incorrect output with Highway < 1.1.0 [dloebl]
• jxlsave: ensure EXIF data is freed on error [kleisauke]
• icc_ac2rc: ensure ICC profile is closed on error [kleisauke]
• buildlut: guard against NaN values [kleisauke]
• tiffsave: better xyz handling [har-pi]
• more use of vips_image_decode() [har-pi]
• quantise: better conversion to rgba [har-pi]
• jp2kload: support palette images [Socialpranker]
• buildlut: better final write calculation [har-pi]

Windows binaries here:

https://github.com/libvips/build-win64-mxe/releases/tag/v8.18.5

Changes since 8.18.3:

• heifload: raise security limits on references [caseywebdev]
• tiffload: squash out read errors in openin_source_read [aleens-labs]
• jxlsave: convert CMYK images to sRGB [DarthSim]
• heifload: fix -Winteger-overflow warning on 32-bit platforms [kleisauke]
• object: guard against non-input arguments [kleisauke]
• enumtypes: avoid misuse of G_GNUC_CONST for GCC 17 compat [kleisauke]

Windows binaries here:

https://github.com/libvips/build-win64-mxe/releases/tag/v8.18.4

Changes since 8.18.2:

• invertlut: verify input mask positions [lovell]
• bandjoin_const: fix possible OOB read with RAD/LABQ images [kleisauke]
• edge: ensure parent object is built [lovell]
• object: guard against invalid interpolators [kleisauke]
• foreign: guard against NULL filenames [kleisauke]
• stdif: expand input by one column [lovell]
• dcrawload: add support for non-square pixels [jcupitt]
• image_sanity: ensure scanlines fit in INT_MAX [Niels Provos]
• rank: expand input by one column [kleisauke]
• hist_local: expand input by one column [kleisauke]
• fix possible null-pointer dereference in quantizr backend [felixbuenemann]
• jp2ksave: fix OOB read during chroma subsampling [dloebl]
• jp2ksave: prevent heap buffer overflow on non-RGB images [dloebl]
• heifload: tighten limits on memory use [provos]
• boolean: fix heap-buffer-overflow in lshift [dloebl]
• magickload: check animation height [Himanshu Anand]
• source: fix max/min confusion [Himanshu Anand]
• uhdrsave: use macro to size buffer [Himanshu Anand]
• draw_mask: fix support for packed LABQ images [lovell]
• reduceh: fix possible OOB read in Highway path [kleisauke]
• buildlut: check upper bound of x range [lovell]
• jxl: ensure status codes are propagated [lovell]
• archive: fix compatibility with MSVC [eunos-1128]
• ppmload: check image dimensions [lovell]
• thumbnail: better pyr detection [Niels Provos]
• jpegsave: fix scrgb save with uhdr disabled [jcupitt]

Windows binaries here:

https://github.com/libvips/build-win64-mxe/releases/tag/v8.18.3

Changes since 8.18.2:

• invertlut: verify input mask positions [lovell]
• bandjoin_const: fix possible OOB read with RAD/LABQ images [kleisauke]
• edge: ensure parent object is built [lovell]
• object: guard against invalid interpolators [kleisauke]
• foreign: guard against NULL filenames [kleisauke]
• stdif: expand input by one column [lovell]
• dcrawload: add support for non-square pixels [jcupitt]
• image_sanity: ensure scanlines fit in INT_MAX [Niels Provos]
• rank: expand input by one column [kleisauke]
• hist_local: expand input by one column [kleisauke]
• fix possible null-pointer dereference in quantizr backend [felixbuenemann]
• jp2ksave: fix OOB read during chroma subsampling [dloebl]
• jp2ksave: prevent heap buffer overflow on non-RGB images [dloebl]
• heifload: tighten limits on memory use [provos]
• boolean: fix heap-buffer-overflow in lshift [dloebl]
• magickload: check animation height [Himanshu Anand]
• radload: deprecate old-style scanline read [Himanshu Anand]
• source: fix max/min confusion [Himanshu Anand]
• uhdrsave: use macro to size buffer [Himanshu Anand]
• draw_mask: fix support for packed LABQ images [lovell]
• reduceh: fix possible OOB read in Highway path [kleisauke]
• buildlut: check upper bound of x range [lovell]
• jxl: ensure status codes are propagated [lovell]
• archive: fix compatibility with MSVC [eunos-1128]
• ppmload: check image dimensions [lovell]
• thumbnail: better pyr detection [Niels Provos]

Changes since 8.18.1:

• convolution: avoid using unsigned accumulators [nakrovati]
• exif: check ifdN range [trailofbits]
• tiffload: check jpeg and jp2k components [wooseokdotkim]
• uhdrsave: set Q for gainmap recompress, don't chroma subsample RGB gainmaps
• uhdrsave: prevent early unref of image with alpha channel [lovell]
• rad2float: limit band count when converting from radiance [lovell]
• jxlload: check EXIF buffer size before reading header [lovell]

Homebrew has updated, win binaries are here:

https://github.com/libvips/build-win64-mxe/releases/tag/v8.18.2

This release fixes a few reported bugs, plus a lot of bugs found by recent improvements to fuzzing (thanks @dloebl!).

• vector: mask supported Highway targets by builtin targets [kleisauke]
• fix Highway paths on big-endian targets [kleisauke]
• fix build with MSVC [star-hengxing]
• fix compatibility with glibc 2.43 [mtasaka]
• pngload: avoid an expensive check during header read [kleisauke]
• improve vips7 JPEG load compatibility [kleisauke]
• fix saving 3-band MATRIX images [kleisauke]
• uhdrload: expose original gainmap scale factor [lovell]
• uhdrsave: control generated gainmap scale factor [lovell]
• source: guard against length truncation [Niebelungen-D]
• shrinkh: fix possible OOB read in Highway path [dloebl] [kleisauke]
• string_to_array_double: ensure delimiter list is consistent [dloebl] [lovell]
• fix loading/saving of non-8-bit JXL images [DarthSim]
• invertlut: check height before interpolating values [dloebl] [lovell]
• extract: check bounds using unsigned arithmetic [Niebelungen-D] [lovell]
• matrixload: guard against empty and very large inputs [Niebelungen-D] [lovell]
• unpremultiply: check alpha_band is in range [Niebelungen-D] [lovell]
• maplut: ensure lookup table index is unsigned [dloebl] [lovell]
• fwfft: fix heap-buffer-overflow for single-row images [dloebl]
• draw_flood: reject out-of-bounds start point [dloebl]
• openexrload: fix a leak with invalid images [kleisauke]
• canny: fix UB in atan2 lookup table init [dloebl] [kleisauke]
• conva: guard against divide by zero [kleisauke]
• csvload: check whitespace and separator are ASCII [Niebelungen-D] [lovell]
• bandrank: check index is in range [Niebelungen-D] [lovell]
• remainder_const: guard against divide by zero [kleisauke]
• vips_window_take: prevent int underflow for small mapped images [jcupitt]
• composite: fix UB (invalid-enum-value) in ->build() [kleisauke]
• add: prevent possible int overflow [kleisauke]
• bandmean: prevent possible int overflow [kleisauke]
• fastcor: prevent possible int overflow [kleisauke]
• multiply: prevent possible int overflow [kleisauke]
• subtract: prevent possible int overflow [kleisauke]
• hist_cum: prevent possible int overflow [kleisauke]
• convasep: prevent possible int overflow [kleisauke]
• conva: prevent possible int overflow [kleisauke]
• convi: prevent possible int overflow [kleisauke]
• draw_circle: prevent possible int overflow [kleisauke]
• csvload: guard against negative index access [kleisauke]
• convi: guard against invalid shift [kleisauke]
• nary: guard against empty input [kleisauke]
• maplut: add missing overflow checks [kleisauke]
• jpegsave: fix assert fail when saving 2-band image [kleisauke]
• LabS2LabQ: fix UB (undefined-shift) [kleisauke]
• convasep: use unsigned accumulator where appropriate [kleisauke]
• min: fix possible OOB read with complex images [kleisauke]
• max: fix possible OOB read with complex images [kleisauke]
• guard against dimension overflow [ElhananHaenel] [kleisauke]
• gifload: prevent int overflow on 32-bit platforms [ElhananHaenel] [kleisauke]
• sign: use ready image instead of unary->in [kleisauke]
• statistic: use ready image instead of statistic->in [kleisauke]
• hist_plot: handle near-zero ranges for float images [kleisauke]
• affine: check bounds using unsigned arithmetic [kleisauke]

Windows binaries here:

https://github.com/libvips/build-win64-mxe/releases/tag/v8.18.1

The libvips 8.18 release notes are here:

https://www.libvips.org/2025/12/04/What's-new-in-8.18.html

Notable changes since 8.17:

• add dcrawload, dcrawload_source, dcrawload_buffer: load raw camera files using libraw [lxsameer]
• add magickload_source: load from a source with imagemagick
• add vips__worker_exit(): enables fast threadpool shutdown
• larger mmap windows on 64-bit machines improve random access mode performance for many file formats
• pdfload: control region to be rendered via page_box [lovell]
• system: add "cache" argument
• add vips_image_get_tile_width(), vips_image_get_tile_height(): get tile cache geometry hints [jbaiter]
• add vips_uhdrload(), vips_uhdrload_buffer(), vips_uhdrload_source(): load an UltraHDR image
• add vips_uhdrsave(), vips_uhdrsave_buffer(), vips_uhdrsave_target(): save an UltraHDR image
• fix cpp binding leak [VivitionDeveloper]
• add "gainmap" to VipsKeep
• add vips_image_get_gainmap()
• thumbnail updates gainmaps
• add "bitdepth" to jxlsave
• add "path" option to vipsthumbnail, deprecate "output" option [zjturner]
• add "exact" to webpsave
• add "uhdr2scRGB"
• dzsave has gainmap support
• cpp: add .gainmap(), .set(VImage)
• add support for Oklab and Oklch colourspaces
• add vips_Oklab2XYZ(), vips_XYZ2Oklab()
• add vips_Oklab2Oklch(), vips_OKlch2Oklab()
• add vips_interpretation_bands()
• heifsave: add "tune" parameter
• require C++14 as a minimum standard [kleisauke]
• prefer libpng over spng

Windows binaries here:

https://github.com/libvips/build-win64-mxe/releases/tag/v8.18.0

Third release candidate for 8.18. The release notes are here:

https://www.libvips.org/2025/12/04/What's-new-in-8.18.html

Notable changes since 8.17:

• add dcrawload, dcrawload_source, dcrawload_buffer: load raw camera files
  using libraw [lxsameer]
• add magickload_source: load from a source with imagemagick
• add vips__worker_exit(): enables fast threadpool shutdown
• larger mmap windows on 64-bit machines improve random access mode for many
  file formats
• pdfload: control region to be rendered via page_box [lovell]
• system: add "cache" argument
• add vips_image_get_tile_width(), vips_image_get_tile_height(): get tile
  cache geometry hints [jbaiter]
• add vips_uhdrload(), vips_uhdrload_buffer(), vips_uhdrload_source(): load an
  UltraHDR image
• add vips_uhdrsave(), vips_uhdrsave_buffer(), vips_uhdrsave_target(): save an
  UltraHDR image
• fix cpp binding leak [VivitionDeveloper]
• add "gainmap" to VipsKeep
• add vips_image_get_gainmap()
• thumbnail updates gainmaps
• add "bitdepth" to jxlsave
• add "path" option to vipsthumbnail, deprecate "output" option [zjturner]
• add "exact" to webpsave
• add "uhdr2scRGB"
• dzsave has gainmap support
• cpp: add .gainmap(), .set(VImage)
• add support for Oklab and Oklch colourspaces
• add vips_Oklab2XYZ(), vips_XYZ2Oklab()
• add vips_Oklab2Oklch(), vips_OKlch2Oklab()
• add vips_interpretation_bands()
• heifsave: add "tune" parameter
• require C++14 as a minimum standard [kleisauke]
• prefer libpng over spng

Test windows binaries (for rc2) here:

https://github.com/libvips/build-win64-mxe/releases/tag/v8.18.0-rc2