Look what Leo found this time... well, OK, he didn't find this himself, but his acquaintance, Draco had it in his stash:
http://www.unrealtexture.com/Hosted/Leo_T_C_K/BRIGHT.7z
Readme is as follows:
BRIGHT 1.83 (c) 2000 by Erik de Neve, erik@epicgames.com
"Best palette Reduction for Industrial Grade High quality Textures"
Epic Games, Inc., while specifically allowing not-for-profit public distribution,
retains all copyrights to this software.
- The BRIGHT manual -
Contents
* Introduction
* File formats and requirements
* Quick start examples
* Masking and transparency tricks
* Improving quality in the output images
* Command line parameters
* Support
* Credits
** Introduction
BRIGHT is a 32-bit windows-95 command-line utility written for
single-sweep conversion of multiple textures to 8-bit palettized
format using either unique, or a specifiable number of shared palettes,
retaining the optimal color range; there is also a special option
to provide a good color representation for mipmapped versions of
the texture.
Distribution of a limited set of optimal palettes between similar
textures is fully automatic.
The Unreal Editor recognizes textures with matching palettes when they
are first imported, and makes sure only one copy of the palette exists
in the UTX file. This can benefit hardware rendering somewhat since some
3D graphics hardware cards (notably VOODOO/3dfx) cannot cache multiple-palette
data internally and need to have it uploaded every single frame.
It may take some experimenting, but in all cases BRIGHT should be able
to provide the best possible custom palette, far better than the conversion
algorithms in most graphics packages currently on the market. Results
depend on color sampling depth (with which you can trade off quality
for speed) and the number of unique palettes allowed (see options below)
when repalettizing groups of textures together.
** File formats and requirements
Input files are either 24-bit, or automatically promoted to be 24-bit
when loading. Output files are in the (Unreal/Editor-readable)
palettized PCX(default) or BMP formats, or the Targa format.
Bright reads the following file formats:
8/16/24/32 bit TGA Targa format
24-bit truecolor BMP Windows Bitmap format
8-bit palettized BMP (*)
8-bit palettized PCX ZSoft PCX format
(*) Warning: some programs write 8-bit BMP files using an archaic
compressed format, for example Paintshop Pro; the solution is to either
promote these to 24-bits before saving, or to save them as PCX or TGA
files.
UNREAL SPECIFICS:
Note: Using shared palettes has become less of an urgent issue
since the release of Unreal 1. It is still recommended for textures
that obviously share the same hues and are used often, or each frame
( HUD textures !)
When converting texture sets for Unreal, you'll want to use
the shared-palette options (see below) to limit the number of
unique palettes, which helps to speed up rendering on hardware 3d cards
because on most hardware, palettes cannot be buffered in on-card memory
like textures, and have to be uploaded every frame.
The recommended number of palettes is less than 1/4 the number of
textures: which means, for a 40-texture set, try to see if using
a 10 palette limit gives good results.
If so, you may want to push it down to 8 or even 5 palettes per set,
depending on the number and diversity of the textures.
Alternatively, this can be done automatically with the -error option
(see below.)
** Quick start examples
Once all 24-bit textures for a texture set are collected in a single
directory, you can palettize them in one sweep.
To convert an entire directory of 24-bit BMP files to palettized
PCXes, using N shared palettes for the whole texture set:
call BRIGHT with the following parameters:
BRIGHT *.bmp -auto N -o -8 -mips -mask 0.0
To convert a single image to an 8-bit image with it's own unique palette,
do this:
BRIGHT naliqueen.bmp naliqueen.pcx
Note that for single-picture conversion, the output filename is required;
for batch conversion, the second filename is optional and will be interpreted
as either a destination directory or the first part of the output filename.
For example
BRIGHT *.bmp c:\destination\
will convert the .bmp files to .pcx files and put these in
the c:\destination directory.
BRIGHT *.bmp NEW8BIT
converts the .bmp files to .pcx files with "NEW8BIT" prepended to all
the filenames.
Source art delivered in non-BMP 24-bit format (e.g. TIFs) can be converted
to 24-bit BMPs using standard software like Photoshop.
** Masking and transparency tricks
To indicate holes in masked textures in Unreal, make sure to have
your 24-bit source art use an uniform full-brightness pink, orange,
or green color. When masking is enabled during palette quantization
using the -mask X switch, any such areas will be detected and mapped
exclusively to color 0 in the resulting palette.
Mask outlines may still need some manual adjustment in the final
palettized output.
The easiest way to do this is with an image editor that allows
modification of palettized images, e.g. Paint Shop Pro (available
as shareware on the web at www.jasc.com) or Adobe Photoshop.
In PSP, type shift-P to see the palette. For making final adjustments
to the mask, click on the 0th color entry of the palette and
temporarily set it to a bright color; after editing, set that color
back to black (0,0,0) before writing the final image to disk.
Most options map the special masking palette entry 0 to black because
other colors might cause artefacts in bilinear interpolation on e.g.
creature skins in the game; -pinkmask however makes sure that this color
remains pink, useful for manual editing of the texture after palettization.
For whatever purpose, when you need too make sure there is an unused black
color (palette entry 0) in the palette, you can use the -black
options as explained in the command line parameters section below.
Alternatively, just specify a smaller target number of colors.
Transparent textures usually have a black base color. This base black
can be forced to palette color 0 similar to masked textures, which
provides a speedup in Unreal software rendering (which then knows what
texels to skip.) See also the -trans option described below.
** Improving quality in the output images
Banding in pictures is a result of the limited number of colors available in a
palette. Bright tries to design its palette smartly to minimize banding, but it has
to work within the 256 (or smaller) color limit.
When the resulting picture seems to lose some color saturation, you can specify the
-RGB option to use different interal scaling mechanism, which reproduces colors more
accurately (though sometimes at the cost of slightly more banding.)
You may lose small patches of unique colors or gradations, specially if they take up
relatively small amount of space in a large texture. To force the palette to include
this color or do a better job for a certain hue, you can make a version of your texture
where you clone this part many times, thereby forcing the palettization process to assign
more importance to these colors.
Two strategies can be used for this, either painting in unused areas of a texture,
enlarging a texture and cutting out the result after palettization, or explicitly
by making the final texture use the palette of the temporaty version with the
-palette FILE command.
** Command line parameters
The command syntax is:
BRIGHT [switches] file1 [file2] [switches]
General options:
-sizebias with common-palette creation, this compensates
for image size bias. By default smaller pictures are less
important than bigger pictures when creating a common
palette, because all individual pixels have the same
weight regardless of what image they are in.
When this switch is specified, the weight is distributed
equally per-picture instead of per-pixel.
-mips predict and take into account mipmap colors.
-8 allow 8-bit .bmp image input as if it is 24-bit.
-colors N target number of colors to quantize to, N = [1..256]
default is 256.
-o, -overwrite Overwrite existing files without warning
-bmp Write output in bmp format. (default is pcx)
-tga Write output in tga format. (default is pcx)
-common Create a single common palette for all images, e.g. for
animation sequences or small texture sets that look similar
enough to need only one palette.
-dither N Dither strength: 0 = no dither, 100 = max
default is NO dither.
This is only necessary for textures with smooth ramps,
in cases where there are not enough palette colors
allocated to smoothly represent those ranges.
[Note: this is mostly broken. Bright is designed to make dithering
unnecessary, since such checkerboard artefacts always look ugly
in 3D engines.]
-bump X Output a bumpmap for use with Unreal's 'IceTextures'.
This simply maps the intensity gradient of the input image
to a horizontal displacement value in the range (-128, 127)
which is useful to create subtle glass-like refraction effects
when used as the 'GlassTexture' in an IceTexture.
X is the strength multiplier, with a useful range of about
1.0 to 8.0.
Key color masking & transparency stuff:
-pinkmask Precise masking out of all pure pink (255,0,255) colors.
The masked pixels will be mapped to palette color [0],
which will remain pink in the output file. If used on
multiple images, all output images will have color [0] set
to pink regardless of the presence of pink masked areas in
the input images.
-graymask Precise masking out of all pure gray (128,128,128) colors.
The masked pixels will be mapped to palette color [0],
which will remain gray n the output file.
-colmask R G B Precise masking of pixels that have the specified color (R,G,B)
The masked pixels will be mapped to palette color [0],
which will get the specified color in the output file.
-pinkscreen Force pink colors to be masked out, with a default mask
selection range tolerance (see below) of 0.7 If you
do need a different tolerance you can combine this with
the -mask X option.
-black Force absolute black included as palette color 0, independent
of the presence of a possible other black color in the palette;
useful for manual masking.
-mask X Automatic 'magic wand' masking, tolerance X (0.0..1.0)
Similar to the magic wand selection tool in Fractal Painter,
this will interpret as 'masked' a range of colors that
is similar to the main masking-indication color.
Will only take effect on images that have a clearly
marked (pink or light-green) masked-out area.
It is recommended to use only -mask 0.0 during batch
conversion, with pre-cleaned masked textures (using -outmask)
-outmask [ Single-file usage only: ] Outputs a 24-bit bmp image with the
mask cleaned up as specified by -mask and -fatmask, but filled
in with the uniform masking color as specified in the input image.
Very useful for automatic cleaning when it has
ambiguous fuzzy pixels around mask edges; for example,
masked textures made with Fractal Painter's floaters
tend to have such edges.
-trans X [ Single-file usage only: ] Outputs a 24-bit bmp image with all
near-black colors remapped to black. Useful with some transparent
and other textures, to get rid of near-black pixels that would
be indistinguishable but potentially slow down the game
(Unreal-specific, software masking and translucency) Small values
here are the most useful; 1.0 would map any image completely
to black.
Color sampling options:
-ycc use Ycbcr color space (default) which samples colors
somewhat less accurately than RGB ( since there are conversion
steps involved ) but usually results in the best subjective
image quality, especially for difficult pictures with wildly varying
colors.
-rgb Use RGB colors space. Try this if you are converting single, limited-
hue textures or small batches of images, and want to obtain the absolute
highest quality. In most cases though, Ycc will provide the best
subjective color mappings.
-depth NNN try sampling color bit depths, range: 5 to 8
( R,G,B or Y Cr Br )
eg. -depth 888 gives the highest quality(default)
-mipmaps Take into account downscaled versions (mipmaps) when creating the palette.
Example: a sharp black-and-white checkerboard pattern may need gray mipmaps
to look good at a large distance, and thus needs gray palette entries.
-bins N force upper limit to # of distinct input colors; an internal measure of
accuracy for the palette generation. Warning: large numbers greatly
increase processing time.
Clustering palettes:
-auto N Automatic palette combining, allow a total of N
unique palettes for the total batch set.
"-auto 1" is equivalent with "-common".
-predepth NNN Color depths for pre-sampling palette combining;
this determines the accuracy at which temporary
palettes are generated to determine which textures
should share their palettes; does not affect the
final color depth (which is determined by -depth )
Default setting: 655
-error X Automatic palette combining, with max allowed error X;
this is a very subjective measure though; a useful value
is about 200, which in moderately large texture sets
(40 - 80 textures) causes the palettizer to generate
about 1 palette for every 7 textures, on average.
While this option can be very useful since it provides
an automatic limit to the error, it is usually easier
to use the "-auto N" parameter, since adjusting
the number of desired palettes directly is a bit
more intuitive than fiddling with an error value.
Special palette stuff:
-palette FILE Force use of palette from bitmap FILE;
if FILE is not an 8-bit palettized image, a
suitable palette will be created from it on the fly.
-writepal FILE Outputs all palettes generated in as sample color-ramp
pictures (numbers are prefixed to create unique names,
if necessary.)
Presets: (not recommended)
-fast faster but uses less accurate sampling
-best try more accurate sampling (can be *slow*)
-perfect force losless RGB (limited-hue pictures only!)
** Support
Please make sure you have read this manual. Epic does not provide support for
this public version. If for some reason you *absolutely* need to send us mail about
the program (no answer guaranteed), then e-mail erik@epicgames.com, with
'Bright public version' as part of your subject line.
** Credits
Color reduction algorithm inspired by the "DLQ3" color quantizer algorithm developed
by Dennis Lee. Dithering code based on Independent Jpeg Group source code.
- Last modified: April 10 '00
-/-
http://www.unrealtexture.com/Hosted/Leo_T_C_K/BRIGHT.7z
Readme is as follows:
BRIGHT 1.83 (c) 2000 by Erik de Neve, erik@epicgames.com
"Best palette Reduction for Industrial Grade High quality Textures"
Epic Games, Inc., while specifically allowing not-for-profit public distribution,
retains all copyrights to this software.
- The BRIGHT manual -
Contents
* Introduction
* File formats and requirements
* Quick start examples
* Masking and transparency tricks
* Improving quality in the output images
* Command line parameters
* Support
* Credits
** Introduction
BRIGHT is a 32-bit windows-95 command-line utility written for
single-sweep conversion of multiple textures to 8-bit palettized
format using either unique, or a specifiable number of shared palettes,
retaining the optimal color range; there is also a special option
to provide a good color representation for mipmapped versions of
the texture.
Distribution of a limited set of optimal palettes between similar
textures is fully automatic.
The Unreal Editor recognizes textures with matching palettes when they
are first imported, and makes sure only one copy of the palette exists
in the UTX file. This can benefit hardware rendering somewhat since some
3D graphics hardware cards (notably VOODOO/3dfx) cannot cache multiple-palette
data internally and need to have it uploaded every single frame.
It may take some experimenting, but in all cases BRIGHT should be able
to provide the best possible custom palette, far better than the conversion
algorithms in most graphics packages currently on the market. Results
depend on color sampling depth (with which you can trade off quality
for speed) and the number of unique palettes allowed (see options below)
when repalettizing groups of textures together.
** File formats and requirements
Input files are either 24-bit, or automatically promoted to be 24-bit
when loading. Output files are in the (Unreal/Editor-readable)
palettized PCX(default) or BMP formats, or the Targa format.
Bright reads the following file formats:
8/16/24/32 bit TGA Targa format
24-bit truecolor BMP Windows Bitmap format
8-bit palettized BMP (*)
8-bit palettized PCX ZSoft PCX format
(*) Warning: some programs write 8-bit BMP files using an archaic
compressed format, for example Paintshop Pro; the solution is to either
promote these to 24-bits before saving, or to save them as PCX or TGA
files.
UNREAL SPECIFICS:
Note: Using shared palettes has become less of an urgent issue
since the release of Unreal 1. It is still recommended for textures
that obviously share the same hues and are used often, or each frame
( HUD textures !)
When converting texture sets for Unreal, you'll want to use
the shared-palette options (see below) to limit the number of
unique palettes, which helps to speed up rendering on hardware 3d cards
because on most hardware, palettes cannot be buffered in on-card memory
like textures, and have to be uploaded every frame.
The recommended number of palettes is less than 1/4 the number of
textures: which means, for a 40-texture set, try to see if using
a 10 palette limit gives good results.
If so, you may want to push it down to 8 or even 5 palettes per set,
depending on the number and diversity of the textures.
Alternatively, this can be done automatically with the -error option
(see below.)
** Quick start examples
Once all 24-bit textures for a texture set are collected in a single
directory, you can palettize them in one sweep.
To convert an entire directory of 24-bit BMP files to palettized
PCXes, using N shared palettes for the whole texture set:
call BRIGHT with the following parameters:
BRIGHT *.bmp -auto N -o -8 -mips -mask 0.0
To convert a single image to an 8-bit image with it's own unique palette,
do this:
BRIGHT naliqueen.bmp naliqueen.pcx
Note that for single-picture conversion, the output filename is required;
for batch conversion, the second filename is optional and will be interpreted
as either a destination directory or the first part of the output filename.
For example
BRIGHT *.bmp c:\destination\
will convert the .bmp files to .pcx files and put these in
the c:\destination directory.
BRIGHT *.bmp NEW8BIT
converts the .bmp files to .pcx files with "NEW8BIT" prepended to all
the filenames.
Source art delivered in non-BMP 24-bit format (e.g. TIFs) can be converted
to 24-bit BMPs using standard software like Photoshop.
** Masking and transparency tricks
To indicate holes in masked textures in Unreal, make sure to have
your 24-bit source art use an uniform full-brightness pink, orange,
or green color. When masking is enabled during palette quantization
using the -mask X switch, any such areas will be detected and mapped
exclusively to color 0 in the resulting palette.
Mask outlines may still need some manual adjustment in the final
palettized output.
The easiest way to do this is with an image editor that allows
modification of palettized images, e.g. Paint Shop Pro (available
as shareware on the web at www.jasc.com) or Adobe Photoshop.
In PSP, type shift-P to see the palette. For making final adjustments
to the mask, click on the 0th color entry of the palette and
temporarily set it to a bright color; after editing, set that color
back to black (0,0,0) before writing the final image to disk.
Most options map the special masking palette entry 0 to black because
other colors might cause artefacts in bilinear interpolation on e.g.
creature skins in the game; -pinkmask however makes sure that this color
remains pink, useful for manual editing of the texture after palettization.
For whatever purpose, when you need too make sure there is an unused black
color (palette entry 0) in the palette, you can use the -black
options as explained in the command line parameters section below.
Alternatively, just specify a smaller target number of colors.
Transparent textures usually have a black base color. This base black
can be forced to palette color 0 similar to masked textures, which
provides a speedup in Unreal software rendering (which then knows what
texels to skip.) See also the -trans option described below.
** Improving quality in the output images
Banding in pictures is a result of the limited number of colors available in a
palette. Bright tries to design its palette smartly to minimize banding, but it has
to work within the 256 (or smaller) color limit.
When the resulting picture seems to lose some color saturation, you can specify the
-RGB option to use different interal scaling mechanism, which reproduces colors more
accurately (though sometimes at the cost of slightly more banding.)
You may lose small patches of unique colors or gradations, specially if they take up
relatively small amount of space in a large texture. To force the palette to include
this color or do a better job for a certain hue, you can make a version of your texture
where you clone this part many times, thereby forcing the palettization process to assign
more importance to these colors.
Two strategies can be used for this, either painting in unused areas of a texture,
enlarging a texture and cutting out the result after palettization, or explicitly
by making the final texture use the palette of the temporaty version with the
-palette FILE command.
** Command line parameters
The command syntax is:
BRIGHT [switches] file1 [file2] [switches]
General options:
-sizebias with common-palette creation, this compensates
for image size bias. By default smaller pictures are less
important than bigger pictures when creating a common
palette, because all individual pixels have the same
weight regardless of what image they are in.
When this switch is specified, the weight is distributed
equally per-picture instead of per-pixel.
-mips predict and take into account mipmap colors.
-8 allow 8-bit .bmp image input as if it is 24-bit.
-colors N target number of colors to quantize to, N = [1..256]
default is 256.
-o, -overwrite Overwrite existing files without warning
-bmp Write output in bmp format. (default is pcx)
-tga Write output in tga format. (default is pcx)
-common Create a single common palette for all images, e.g. for
animation sequences or small texture sets that look similar
enough to need only one palette.
-dither N Dither strength: 0 = no dither, 100 = max
default is NO dither.
This is only necessary for textures with smooth ramps,
in cases where there are not enough palette colors
allocated to smoothly represent those ranges.
[Note: this is mostly broken. Bright is designed to make dithering
unnecessary, since such checkerboard artefacts always look ugly
in 3D engines.]
-bump X Output a bumpmap for use with Unreal's 'IceTextures'.
This simply maps the intensity gradient of the input image
to a horizontal displacement value in the range (-128, 127)
which is useful to create subtle glass-like refraction effects
when used as the 'GlassTexture' in an IceTexture.
X is the strength multiplier, with a useful range of about
1.0 to 8.0.
Key color masking & transparency stuff:
-pinkmask Precise masking out of all pure pink (255,0,255) colors.
The masked pixels will be mapped to palette color [0],
which will remain pink in the output file. If used on
multiple images, all output images will have color [0] set
to pink regardless of the presence of pink masked areas in
the input images.
-graymask Precise masking out of all pure gray (128,128,128) colors.
The masked pixels will be mapped to palette color [0],
which will remain gray n the output file.
-colmask R G B Precise masking of pixels that have the specified color (R,G,B)
The masked pixels will be mapped to palette color [0],
which will get the specified color in the output file.
-pinkscreen Force pink colors to be masked out, with a default mask
selection range tolerance (see below) of 0.7 If you
do need a different tolerance you can combine this with
the -mask X option.
-black Force absolute black included as palette color 0, independent
of the presence of a possible other black color in the palette;
useful for manual masking.
-mask X Automatic 'magic wand' masking, tolerance X (0.0..1.0)
Similar to the magic wand selection tool in Fractal Painter,
this will interpret as 'masked' a range of colors that
is similar to the main masking-indication color.
Will only take effect on images that have a clearly
marked (pink or light-green) masked-out area.
It is recommended to use only -mask 0.0 during batch
conversion, with pre-cleaned masked textures (using -outmask)
-outmask [ Single-file usage only: ] Outputs a 24-bit bmp image with the
mask cleaned up as specified by -mask and -fatmask, but filled
in with the uniform masking color as specified in the input image.
Very useful for automatic cleaning when it has
ambiguous fuzzy pixels around mask edges; for example,
masked textures made with Fractal Painter's floaters
tend to have such edges.
-trans X [ Single-file usage only: ] Outputs a 24-bit bmp image with all
near-black colors remapped to black. Useful with some transparent
and other textures, to get rid of near-black pixels that would
be indistinguishable but potentially slow down the game
(Unreal-specific, software masking and translucency) Small values
here are the most useful; 1.0 would map any image completely
to black.
Color sampling options:
-ycc use Ycbcr color space (default) which samples colors
somewhat less accurately than RGB ( since there are conversion
steps involved ) but usually results in the best subjective
image quality, especially for difficult pictures with wildly varying
colors.
-rgb Use RGB colors space. Try this if you are converting single, limited-
hue textures or small batches of images, and want to obtain the absolute
highest quality. In most cases though, Ycc will provide the best
subjective color mappings.
-depth NNN try sampling color bit depths, range: 5 to 8
( R,G,B or Y Cr Br )
eg. -depth 888 gives the highest quality(default)
-mipmaps Take into account downscaled versions (mipmaps) when creating the palette.
Example: a sharp black-and-white checkerboard pattern may need gray mipmaps
to look good at a large distance, and thus needs gray palette entries.
-bins N force upper limit to # of distinct input colors; an internal measure of
accuracy for the palette generation. Warning: large numbers greatly
increase processing time.
Clustering palettes:
-auto N Automatic palette combining, allow a total of N
unique palettes for the total batch set.
"-auto 1" is equivalent with "-common".
-predepth NNN Color depths for pre-sampling palette combining;
this determines the accuracy at which temporary
palettes are generated to determine which textures
should share their palettes; does not affect the
final color depth (which is determined by -depth )
Default setting: 655
-error X Automatic palette combining, with max allowed error X;
this is a very subjective measure though; a useful value
is about 200, which in moderately large texture sets
(40 - 80 textures) causes the palettizer to generate
about 1 palette for every 7 textures, on average.
While this option can be very useful since it provides
an automatic limit to the error, it is usually easier
to use the "-auto N" parameter, since adjusting
the number of desired palettes directly is a bit
more intuitive than fiddling with an error value.
Special palette stuff:
-palette FILE Force use of palette from bitmap FILE;
if FILE is not an 8-bit palettized image, a
suitable palette will be created from it on the fly.
-writepal FILE Outputs all palettes generated in as sample color-ramp
pictures (numbers are prefixed to create unique names,
if necessary.)
Presets: (not recommended)
-fast faster but uses less accurate sampling
-best try more accurate sampling (can be *slow*)
-perfect force losless RGB (limited-hue pictures only!)
** Support
Please make sure you have read this manual. Epic does not provide support for
this public version. If for some reason you *absolutely* need to send us mail about
the program (no answer guaranteed), then e-mail erik@epicgames.com, with
'Bright public version' as part of your subject line.
** Credits
Color reduction algorithm inspired by the "DLQ3" color quantizer algorithm developed
by Dennis Lee. Dithering code based on Independent Jpeg Group source code.
- Last modified: April 10 '00
-/-
