DEBUG build
Basis Universal LDR/HDR GPU Texture Compression and Transcoding System v1.60.0 (x64)
Copyright (C) 2019-2025 Binomial LLC, All rights reserved

Usage: basisu filename [filename ...] <options>

The default processing mode is compression of one or more .PNG/.BMP/.TGA/.JPG/.QOI/.DDS/.EXR/.HDR files to a LDR or HDR .KTX2 file. Alternate modes:
 -unpack: Use transcoder to unpack a .basis/.KTX2 file to one or more .KTX/.PNG files
 -validate: Validate and display information about a .basis/.KTX2 file
 -info: Display high-level information about a .basis/.KTX2 file
 -compare: Compare two LDR PNG/BMP/TGA/JPG/QOI images specified with -file, output PSNR and SSIM statistics and RGB/A delta images
 -compare_hdr: Compare two HDR .EXR/.HDR images specified with -file, output PSNR statistics and RGB delta images
 -tonemap: Tonemap an HDR or EXR image to PNG at multiple exposures, use -file to specify filename
 -version: Print version and exit

--- Notes:

Unless an explicit mode is specified, if one or more files have the .basis or .KTX2 extension this tool defaults to unpack mode.

By default, the compressor assumes the input is in the sRGB colorspace (like photos/albedo textures).
If the input is NOT sRGB (like a normal map), be sure to specify -linear for less artifacts. Depending on the content type, some experimentation may be needed.

The TinyEXR library is used to read .EXR images. This library does not support all .EXR compression methods. For unsupported images, you can use ImageMagick to convert them to uncompressed .EXR.

For .DDS source files: Mipmapped or not mipmapped 2D textures (but not cubemaps) are supported. Only uncompressed 32-bit RGBA/BGRA, half float RGBA, or float RGBA .DDS files are supported. In -tex_array mode, if a .DDS file is specified, all source files must be in .DDS format.

Filenames prefixed with a @ symbol are read as filename listing files. Listing text files specify which actual filenames to process (one filename per line).

--- Texture Mode Options:
 -etc1s: Encode to ETC1S LDR (the default for SDR/LDR inputs). Roughly .8-2.5 bpp.
 -uastc: Encode to UASTC LDR 4x4. Roughly 5-8 bpp.
 -hdr/-hdr_4x4: Encode input as UASTC HDR 4x4 (the default if any input file has the .EXR or .HDR extension, or if any .DDS file is HDR). Roughly 5-8 bpp.
 -hdr_6x6: Encode input as RDO or highest quality UASTC HDR 6x6. Use -lambda X (try 100-20000 or higher) option to enable RDO UASTC HDR 6x6, where x controls the quality vs. size tradeoff. Roughly 1.2-3.2 bpp.
 -hdr_6x6i: Encode input as UASTC HDR 6x6 intermediate. Use -lambda X (try 100-20000 or higher) option to enable RDO UASTC HDR 6x6, where x controls the quality vs. size tradeoff. Roughly 1-3.2 bpp.

--- Options:
 -ktx2: Write .KTX2 files (the default). By default, UASTC LDR/HDR 4x4 and ASTC 6x6 files will be compressed using Zstandard unless -ktx2_no_zstandard is specified.
 -basis: Write .basis files instead of .KTX2 files (the previous default).
 -file filename.png/tga/jpg/qoi/exr/hdr: Input image filename, multiple images are OK, use -file X for each input filename (prefixing input filenames with -file is optional)
 -alpha_file filename.png/tga/jpg/qoi: Input alpha image filename, multiple images are OK, use -file X for each input filename (must be paired with -file), images converted to REC709 grayscale and used as input alpha
 -output_file filename: Output .basis/.KTX2 filename
 -output_path: Output .basis/.KTX2 files to specified directory.
 -debug or -verbose: Enable codec debug print to stdout (slightly slower).
 -debug_images: Enable codec debug images (much slower).
 -stats: Compute and display image quality metrics (slightly to much slower).
 -individual: Process input images individually and output multiple .basis/.KTX2 files (not as a texture array - this is now the default as of v1.16)

 -fastest: Set UASTC LDR 4x4 and HDR 4x4/6x6 to fastest but lowest quality encoding mode (same as -uastc_level 0 or -hdr_6x6_level 0)
 -slower: Set UASTC LDR 4x4 and HDR 4x4/6x6 to slower but a higher quality encoding mode (same as -uastc_level 3 or -hdr_6x6_level 5)
 -parallel: Compress multiple textures simumtanously (one per thread), instead of one at a time. Compatible with OpenCL mode. This is much faster, but in OpenCL mode the driver is pushed harder, and the CLI output will be jumbled.
 -linear: Use linear colorspace metrics (instead of the default sRGB or scaled RGB for HDR), and by default linear (not sRGB) mipmap filtering.
 -tex_type <2d, 2darray, 3d, video, cubemap>: Set Basis file header's texture type field. Cubemap arrays require multiples of 6 images, in X+, X-, Y+, Y-, Z+, Z- order, each image must be the same resolutions.
  2d=arbitrary 2D images, 2darray=2D array, 3D=volume texture slices, video=video frames, cubemap=array of faces. For 2darray/3d/cubemaps/video, each source image's dimensions and # of mipmap levels must be the same.
  For video, the .basis file will be written with the first frame being an I-Frame, and subsequent frames being P-Frames (using conditional replenishment). Playback must always occur in order from first to last image.
 -cubemap: same as -tex_type cubemap
 -tex_array: Process input images as a single texture array and write a single .basis/.KTX2 file (the former default before v1.16)
 -fuzz_testing: Use with -validate: Disables CRC16 validation of file contents before transcoding
 -multifile_printf: printf() format strint to use to compose multiple filenames
 -multifile_first: The index of the first file to process, default is 0 (must specify -multifile_printf and -multifile_num)
 -multifile_num: The total number of files to process.
 -opencl: Enable OpenCL usage (currently only accelerates ETC1S encoding)
 -opencl_serialize: Serialize all calls to the OpenCL driver (to work around buggy drivers, only useful with -parallel)

--- ETC1S specific options (-etc1s - the LDR/SDR default):
 -q X: Set ETC1S quality level, 1-255, default is 128, lower=better compression/lower quality/faster, higher=less compression/higher quality/slower, default is 128. For even higher quality, use -max_endpoints/-max_selectors.
 -comp_level X: Set ETC1S encoding speed vs. quality tradeoff. Range is 0-6, default is 1. Higher values=MUCH slower, but slightly higher quality. Higher levels intended for videos. Use -q first!
 -max_endpoints X: ETC1S: Manually set the max number of color endpoint clusters from 1-16128, use instead of -q
 -max_selectors X: ETC1S: Manually set the max number of color selector clusters from 1-16128, use instead of -q

--- UASTC LDR/HDR 4x4 specific options (-uastc):
 -uastc: Enable UASTC LDR 4x4 texture mode, instead of the default ETC1S mode. Significantly higher texture quality, but much larger (~8bpp) files. (Note that UASTC .basis files must be losslessly compressed by the user.)
 -uastc_level: Set UASTC LDR/HDR 4x4 encoding level. LDR Range is [0,4], default is 2, higher=slower but higher quality. 0=fastest/lowest quality, 3=slowest practical option, 4=impractically slow/highest achievable quality
				UASTC HDR 4x4 range is [0,4] - higher=slower but higher quality. HDR 4x4 default level=1.
 -uastc_rdo_l X: Enable UASTC LDR 4x4 RDO post-processing and set UASTC LDR 4x4 RDO quality scalar (lambda) to X. Lower values=higher quality/larger LZ
                 compressed files, higher values=lower quality/smaller LZ compressed files. Good range to try is [.25-10].
                 Note: Previous versons used the -uastc_rdo_q option, which was removed because the RDO algorithm was changed.
 -uastc_rdo_d X: Set UASTC LDR 4x4 RDO dictionary size in bytes. Default is 4096, max is 65536. Lower values=faster, but less compression.
 -uastc_rdo_b X: Set UASTC LDR 4x4 RDO max smooth block error scale. Range is [1,300]. Default is 10.0, 1.0=disabled. Larger values suppress more artifacts (and allocate more bits) on smooth blocks.
 -uastc_rdo_s X: Set UASTC LDR 4x4 RDO max smooth block standard deviation. Range is [.01,65536]. Default is 18.0. Larger values expand the range of blocks considered smooth.
 -uastc_rdo_f: Don't favor simpler UASTC LDR 4x4 modes in RDO mode.
 -uastc_rdo_m: Disable RDO multithreading (slightly higher compression, deterministic).

--- UASTC HDR 4x4 specific options (-hdr or -hdr_4x4 - the HDR default):
 -uastc_level X: Sets the UASTC HDR 4x4 compressor's level. Valid range is [0,4] - higher=slower but higher quality. HDR default=1.
                 Level 0=fastest/lowest quality, 3=highest practical setting, 4=exhaustive
 -hdr_uber_mode: Allow the UASTC HDR 4x4 encoder to try varying the CEM 11 selectors more for slightly higher quality (slower). This may negatively impact BC6H quality, however.
 -hdr_ultra_quant: UASTC HDR 4x4: Try to find better quantized CEM 7/11 endpoint values (slower).
 -hdr_favor_astc: UASTC HDR 4x4: By default the UASTC HDR 4x4 encoder tries to strike a balance or even slightly favor BC6H quality. If this option is specified, ASTC HDR 4x4 quality is favored instead.

--- UASTC HDR 6x6 specific options (-hdr_6x6 or -hdr_6x6i):
 -lambda X: Enables rate distortion optimization (RDO). The higher this value, the lower the quality, but the smaller the file size. Try 100-20000, or higher values on some images.
 -hdr_6x6_level X: Sets the codec to 6x6 HDR mode (same as -hdr_6x6) and controls encoder performance vs. max quality tradeoff. X may range from [0,12]. Default level is 2. Higher values result in better quality but slower encoding. Values above 10 are extremely slow.
 -hdr_6x6i_level X: Sets the codec to 6x6 HDR intermediate mode (same as -hdr_6x6i) and controls encoder performance vs. max quality tradeoff. X may range from [0,12]. Default level is 2.
 -rec_2020: The input image's gamut is Rec. 2020 vs. the default Rec. 709 - for accurate colorspace error calculations.
 -hdr_6x6_jnd X, -hdr_6x6_extra_pats, -hdr_6x6_brute_force_pats, -hdr_6x6_comp_levels X Y or -hdr_6x6i_comp_levels X Y: Low-level control over the encoder's configuration.

--- SDR/LDR->HDR upconversion options (only used when encoding to HDR formats from an LDR/SDR source image):
 -hdr_ldr_no_srgb_to_linear: If specified, LDR images will NOT be converted to normalized linear light (via a sRGB->Linear conversion) during SDR->HDR upconversion before compressing as HDR.
 -hdr_ldr_upconversion_nit_multiplier X: Specify how many nits (candelas per sq. meter) LDR/SDR images are converted to after converting to linear light. Default is 100 nits. Note: Previous builds used 1 nit.

--- More options:
 -test: Run an automated LDR ETC1S/UASTC encoding and transcoding test. Returns EXIT_FAILURE if any failures
 -test_hdr_4x4/-test_hdr_6x6/-test_hdr_6x6i: Run automated UASTC HDR encoding and transcoding tests. Returns EXIT_FAILURE if any failures
 -test_dir: Optional directory of test files. Defaults to "../test_files".
 -y_flip: Flip input images vertically before compression
 -normal_map: Tunes codec parameters for better quality on normal maps (linear colorspace metrics, linear mipmap filtering, no selector RDO, no sRGB)
 -no_alpha: Always output non-alpha basis files, even if one or more inputs has alpha
 -force_alpha: Always output alpha basis files, even if no inputs has alpha
 -separate_rg_to_color_alpha: Separate input R and G channels to RGB and A (for tangent space XY normal maps)
 -swizzle rgba: Specify swizzle for the 4 input color channels using r, g, b and a (the -separate_rg_to_color_alpha flag is equivalent to rrrg)
 -renorm: Renormalize each input image before any further processing/compression
 -no_multithreading: Disable multithreading
 -max_threads X: Use at most X threads total when multithreading is enabled (this includes the main thread)
 -no_ktx: Disable KTX writing when unpacking (faster, less output files)
 -ktx_only: Only write KTX files when unpacking (faster, less output files)
 -write_out: Write 3dfx OUT files when unpacking FXT1 textures
 -format_only: Only unpack the specified format, by its numeric code.
 -etc1_only: Only unpack to ETC1, skipping the other texture formats during -unpack
 -disable_hierarchical_endpoint_codebooks: Disable hierarchical endpoint codebook usage, slower but higher quality on some compression levels
 -compare_ssim: Compute and display SSIM of image comparison (slow)
 -compare_plot: Display histogram plots in -compare mode
 -bench: UASTC benchmark mode, for development only
 -resample X Y: Resample all input textures to XxY pixels using a box filter
 -resample_factor X: Resample all input textures by scale factor X using a box filter
 -no_sse: Forbid all SSE instruction set usage
 -validate_etc1s: Validate internal ETC1S compressor's data structures during compression (slower, intended for development).
 -ktx2_animdata_duration X: Set KTX2animData duration field to integer value X (only valid/useful for -tex_type video, default is 1)
 -ktx2_animdata_timescale X: Set KTX2animData timescale field to integer value X (only valid/useful for -tex_type video, default is 15)
 -ktx2_animdata_loopcount X: Set KTX2animData loopcount field to integer value X (only valid/useful for -tex_type video, default is 0)
 -framerate X: Set framerate in .basis header to X/frames sec.
 -ktx2_no_zstandard: Don't compress UASTC texture data using Zstandard -- store it uncompressed instead.
 -ktx2_zstandard_level X: Set ZStandard compression level to X (see Zstandard documentation, default level is 6)
 -tonemap_dither: Dither tonemapper's 8-bit/component output by adding a small amount of white noise, only used with -tonemap mode

--- Mipmap generation options:
 -mipmap: Generate mipmaps for each source image
 -mip_srgb: Convert image to linear before filtering, then back to sRGB
 -mip_linear: Keep image in linear light during mipmap filtering (i.e. do not convert to/from sRGB for filtering purposes)
 -mip_scale X: Set mipmap filter kernel's scale, lower=sharper, higher=more blurry, default is 1.0
 -mip_filter X: Set mipmap filter kernel, default is kaiser, filters: box, tent, bell, blackman, catmullrom, mitchell, etc.
 -mip_renorm: Renormalize normal map to unit length vectors after filtering
 -mip_clamp: Use clamp addressing on borders, instead of wrapping
 -mip_fast: Use faster mipmap generation (resample from previous mip, not always first/largest mip level). The default (as of 1/2021)
 -mip_slow: Always resample each mipmap level starting from the largest mipmap. Higher quality, but slower. Opposite of -mip_fast. Was the prior default before 1/2021.
 -mip_smallest X: Set smallest pixel dimension for generated mipmaps, default is 1 pixel
  By default, textures will be converted from sRGB to linear light before mipmap filtering, then back to sRGB (for the RGB color channels) unless -linear is specified.
  You can override this behavior with -mip_srgb/-mip_linear.

--- ETC1S backend endpoint/selector RDO codec options:
 -no_selector_rdo: Disable backend's selector rate distortion optimizations (slightly faster, less noisy output, but lower quality per output bit)
 -selector_rdo_thresh X: Set selector RDO quality threshold, default is 1.25, lower is higher quality but less quality per output bit (try 1.0-3.0)
 -no_endpoint_rdo: Disable backend's endpoint rate distortion optimizations (slightly faster, less noisy output, but lower quality per output bit)
 -endpoint_rdo_thresh X: Set endpoint RDO quality threshold, default is 1.5, lower is higher quality but less quality per output bit (try 1.0-3.0)

--- Set various fields in the Basis file header:
 -userdata0 X: Set 32-bit userdata0 field in Basis file header to X (X is a signed 32-bit int)
 -userdata1 X: Set 32-bit userdata1 field in Basis file header to X (X is a signed 32-bit int)

--- Example LDR ETC1S/UASTC LDR 4x4 command lines:
 basisu x.png : Compress sRGB image x.png to x.ktx2 using default settings (multiple filenames OK, use -tex_array if you want a tex array vs. multiple output files)
 basisu -basis x.qoi : Compress sRGB image x.qoi to x.basis (supports 24-bit or 32-bit .QOI files)
 basisu x.ktx2 : Unpack x.basis to PNG/KTX files (multiple filenames OK)
 basisu x.basis : Unpack x.basis to PNG/KTX files (multiple filenames OK)
 basisu -uastc x.png -uastc_rdo_l 2.0 -ktx2 -stats : Compress to a UASTC .KTX2 file with RDO (rate distortion optimization) to reduce .KTX2 compressed file size
 basisu -file x.png -mipmap -y_flip : Compress a mipmapped x.ktx2 file from an sRGB image named x.png, Y flip each source image
 basisu -validate -file x.basis : Validate x.basis (check header, check file CRC's, attempt to transcode all slices)
 basisu -unpack -file x.basis : Validates, transcodes and unpacks x.basis to mipmapped .KTX and RGB/A .PNG files (transcodes to all supported GPU texture formats)
 basisu -q 255 -file x.png -mipmap -debug -stats : Compress sRGB x.png to x.ktx2 at quality level 255 with compressor debug output/statistics
 basisu -linear -max_endpoints 16128 -max_selectors 16128 -file x.png : Compress non-sRGB x.png to x.ktx2 using the largest supported manually specified codebook sizes
 basisu -basis -comp_level 2 -max_selectors 8192 -max_endpoints 8192 -tex_type video -framerate 20 -multifile_printf "x%02u.png" -multifile_first 1 -multifile_num 20 : Compress a 20 sRGB source image video sequence (x01.png, x02.png, x03.png, etc.) to x01.basis

--- Example UASTC HDR 4x4 command lines:
 basisu x.exr : Compress a HDR .EXR (or .HDR) image to a UASTC HDR 4x4 .KTX2 file. LDR/SDR images will be upconverted to linear light HDR before compression. See HDR upconversion options, above.
 basisu -hdr_4x4 x.exr : Compress a HDR .EXR image to a UASTC HDR 4x4 .KTX2 file.
 basisu x.hdr -uastc_level 0 : Compress a HDR .hdr image to a UASTC HDR 4x4 .KTX2 file, fastest encoding but lowest quality
 basisu -hdr x.png : Compress a LDR .PNG image to UASTC HDR 4x4 (image is converted from sRGB to linear light first, use -hdr_ldr_no_srgb_to_linear to disable)
 basisu x.hdr -uastc_level 3 : Compress a HDR .hdr image to UASTC HDR 4x4 at higher quality (-uastc_level 4 is highest quality, but very slow encoding)
 basisu x.hdr -uastc_level 3 -mipmap -basis -stats -debug -debug_images : Compress a HDR .hdr image to UASTC HDR 4x4, .basis output file, at higher quality, generate mipmaps, output statistics and debug information, and write tone mapped debug images
 basisu x.hdr -stats -hdr_favor_astc -hdr_uber_mode -uastc_level 4 : Highest achievable ASTC HDR 4x4 quality (very slow encoding, BC6H quality is traded off)

--- Example UASTC HDR 6x6 command lines:
 basisu -hdr_6x6 x.exr : Compress a HDR .EXR (or .HDR) image to a UASTC HDR 6x6 .KTX2 file. LDR/SDR images will be upconverted to linear light HDR before compression. See HDR upconversion options, above.
 basisu -lambda 1000 -hdr_6x6 x.exr : Compress a HDR .EXR (or .HDR) image to a UASTC HDR 6x6 .KTX2 file with rate-distortion optimization (RDO), at lambda level 1000.
 basisu -hdr_6x6i x.exr : Compress a HDR .EXR image to a compressed intermediate format UASTC HDR 6x6 .KTX2 file.
 basisu -lambda 1000 -hdr_6x6i x.exr : Compress a HDR .EXR image to a compressed intermediate format UASTC HDR 6x6 .KTX2 file with rate-distortion optimization (RDO), at lambda level 1000.

--- Video notes: For video use, it's recommended to encode on a machine with many cores. Use -comp_level 2 or higher for better codebook
generation, specify very large codebooks using -max_endpoints and -max_selectors, and reduce the default endpoint RDO threshold
(-endpoint_rdo_thresh) to around 1.25. Videos may have mipmaps and alpha channels. Videos must always be played back by the transcoder
in first to last image order.
Video files currently use I-Frames on the first image, and P-Frames using conditional replenishment on subsequent frames.

ETC1S Compression level (-comp_level X) details. This controls the ETC1S speed vs. quality trandeoff. (Use -q to control the quality vs. compressed size tradeoff.):
 Level 0: Fastest, but has marginal quality and can be brittle on complex images. Avg. Y dB: 35.45
 Level 1: Hierarchical codebook searching, faster ETC1S encoding. 36.87 dB, ~1.4x slower vs. level 0. (This is the default setting.)
 Level 2: Use this or higher for video. Hierarchical codebook searching. 36.87 dB, ~1.4x slower vs. level 0. (This is the v1.12's default setting.)
 Level 3: Full codebook searching. 37.13 dB, ~1.8x slower vs. level 0. (Equivalent the the initial release's default settings.)
 Level 4: Hierarchical codebook searching, codebook k-means iterations. 37.15 dB, ~4x slower vs. level 0
 Level 5: Full codebook searching, codebook k-means iterations. 37.41 dB, ~5.5x slower vs. level 0.
 Level 6: Full codebook searching, twice as many codebook k-means iterations, best ETC1 endpoint opt. 37.43 dB, ~12x slower vs. level 0