Limitations

edit

The library never decodes or encodes a file approximately: a feature it does not support throws an UnsupportedImageFeatureException, and a conversion that would lose information needs an explicit setting. This page lists the main gaps. Each format page gives the details of its format.

General

No extension points. There is no public allocator, custom pixel type, codec plug-in or custom image processor.
No CMYK pixel format. CMYK JPEG and TIFF files are converted to RGB when they are loaded, with an explicit CmykConversion option; CMYK cannot be written. See Color management.
No floating-point grayscale. RgbaFloat and RgbaHalf are the floating-point formats. OpenEXR and TIFF read and write them; the other encoders need an explicit reduction to 16 bits.
No scaled or region decoding. An image is decoded at its full size; resize it afterward.
Color profiles are never applied implicitly. Colors change only when you convert them (ConvertColorProfile). See Color management.
Color management covers ICC versions 2 and 4 grayscale, RGB and CMYK profiles, and device-link and abstract profiles in sequences of profiles; no iccMAX or named-color profiles. CMYK samples are not a pixel format and no encoder writes them. Gamut mapping is per-channel clipping or a chroma reduction, opt-in.

Formats

Format	Main limitations
JPEG	Hierarchical JPEG, JPEG-LS and lossless JPEG with arithmetic coding are not read. Writing is 8-bit baseline or progressive only (no 12-bit, lossless, arithmetic or CMYK), one frame. Gain maps are only read by `AuxiliaryImageCollection` and are not written.
GIF	At most 256 colors per frame and one fully transparent color; plain-text extensions are rejected.
PNG / APNG	Sub-byte grayscale requires PngOptimization.MinimumSize or PngOptimization.Layout (still images), and APNG delta rectangles OptimizeFrames or MinimumSize; palettes are selected by MinimumSize, Layout or PngPixelLayout.Indexed. Sequential writers fix their layout and palette from the first image. Unknown ancillary chunks are not kept.
WebP	8-bit RGB(A) only; lossy output is 4:2:0; animations need a seekable destination.
QOI	8-bit RGB(A) only, one still image, no metadata.
BMP	No embedded JPEG/PNG payloads, no embedded ICC profile; one still image is written.
TGA	No premultiplied alpha; true-color and grayscale output only, no metadata.
Netpbm	Only the first image of a concatenation is read; no metadata.
TIFF	Old-style JPEG compression, YCbCr data in separate planes, Lab* and some extra-sample layouts are not read. Writing uses chunky samples (no separate planes), one page per Save (use ImageCollection for documents) and a seekable destination.
ICO / CUR	At most 256 pixels per side; 32-bit DIB or PNG representations are written.
ANI	One representation per frame; always loops; needs a seekable destination.
AVIF	Gain maps, depth maps and thumbnails are only read by `AuxiliaryImageCollection`; gain maps are not written. Sequences need a seekable destination; the encoder is single-threaded except for grid tiles.
HEIC	Bit depths above 12, multilayer and 3D extensions are not read; bursts and live-photo videos are ignored, depth maps, thumbnails and gain maps are only read by `AuxiliaryImageCollection` (Apple's gain map parameters, in maker notes, are not read). Sequences need a seekable destination.
JPEG XL	Floating-point samples, more than 16 bits, premultiplied alpha and CMYK are not read; lossy decoding is close to, not identical with, the reference decoder. Writing has no layers or blending.
JPEG 2000	Signed samples, more than 16 bits and Part 2 extensions are not read; no decoding of fewer quality layers; a region decode reads the whole compressed data of each tile it meets. Writing has no precincts and no tile-parts beyond one per tile; HTJ2K output has one quality layer and no HT refinement passes.
OpenEXR	Multi-part files, deep data, DWA/HTJ2K/ZSTD compressions and chroma (RY, BY) channels are not read; other channels than R, G, B, A and Y, and an overscan, need a layer or a window mode (OpenExrDecodingOptions). Writing uses lossless scan-line compression, but alpha premultiplication and sample conversion can change source pixels.

Processing

Convolution matrices have odd dimensions and no divisor (divide the weights).
Edge detection has no non-maximum suppression or thresholds (no Canny detector).
Histograms, equalization and auto-levels work on integer pixel formats only.

Drawing and text

Drawing is sequential: it does not use several threads.
No mesh gradients.
Text lines are broken greedily; no word lists are provided for the scripts that need a dictionary to find word boundaries.
No AAT (morx, kerx) shaping, no subpixel (LCD) rendering, no ruby in vertical text.
Hinting is opt-in and horizontal only.
Installed fonts are used only when you ask for them (Font.LoadInstalled, FontStack.WithInstalledFallback).
No SVG glyphs, Type 1 fonts or fonts without outlines.