PNG file processing tool to use for texture in game.

3D graphics API (OpenGL and DirectX) provides any efficiency texture formats like RGBA5551, RGB565, RGBA444 and so on. They can save graphics' memory and improve performance.

And current gaming API supports PNG file for texture. But PNG doesn't support any 16 bit per pixel formats.

This program generates a color reducted PNG file with dithering and it uses 16 bit color space only. It generates high quality PNG files by using Floyd–Steinberg dithering algorithm. So you can easy to create 16 bit texture from this file like it:

// if it has no alpha channel
png_byte  src[height][width * 3];
uint16_t  texture[height * width];
for (size_t y = 0; y < height; ++y)
{
    for (size_t x = 0; x < width; ++x)
    {
        int r = src[y][x * 3];
        int g = src[y][x * 3];
        int b = src[y][x * 3];
        texture[y * width + x] = static_cast<uint16_t>((r << 11) | (g << 5) | (b));
    }
}

PNG is east to handle for developers and designers. And the result PNG file become smaller than the original PNG file. It will help many mobile game developers.

ngCore already supports RGBA4444 and RGB565 format. You can create 16bit GL2.Texture from this PNG file directly.

• 16 Bit Conversion:

  Cut lower bits.

• Dithering:

  Standard Floyd-Steinberg dithering.

• Quantize(1):

  It uses libimagequant (this is a library version of pngquant).

• Quantize(2):

  Cuts lower bits and applies Median-Cut algorithm.

• Clean RGB:

  Sets R, G, B pixel color values to 0 if the alpha value is 0.

• Clean Palette:

  Removes unused palette.

• PNG zlib parameter

  It tries 4 strategies (Z_DEFAULT_STRATEGY, Z_FILTERED, Z_RLE, Z_HUFFMAN_ONLY, not Z_FIXED) x 6 filter options and decides the smallest option.

• Zopfli

  There are two options. --optimize 2 runs zlib otpion trial (Z_DEFAULT_STRATEGY and all filter options) and then tries Zopfli compression with the smallest filter. --optimize 3 is a Brute-force method. It tries all zlib options and all Zopfli options.

It needs the following library. You should download and decompress into the root folder of lightpng's folder:

• boost_1_53_0

It uses Zopfli, libimagequant as submodule. At first you should run the following command

$ git submodule init
$ git submodule update

It uses the following libraries, but these source codes are bundled:

• libpng (modified version to use Zopfli)
• libz
• jpeglib
• pthreads-win32 (for windows cross compile)

This repository contains a build setting file for SCons. I tested on Mac OS X with MacPorts environment, but maybe it will work on any scons/gcc environment:

$ sudo port install scons
$ scons
$ ./lightpng input.png -16m output.png

If you have MinGW environment, you can build Win32 binary in your environment

$ sudo port install i386-mingw32-gcc scons
$ soncs --mingw32
$ wine lightpng.exe input.png -16m output.png

Use this command like this:

$ lightpng [opt] input_image [output_options]

input_image can accept .png and .jpg file.

• -o level, --optimize level:

  Set optimize level:

  • 0 - No optimize (fastest)
  • 1 - PNG zlib option optimize + index color optimize(default)
  • 2 - Use zopfli with one fiilter + 1
  • 3 - Use zopfli with all filters + 1

  level 2 and 3 gives the same results in almost all cases. Preview modes always use 0.

• -b, --benchmark:

  Display time to process.

• -v, --verbose:

  Display compression result.

• -h, --help: Show this message.

• -16m PATH:

  16 bit PNG with 1 bit alpha (RGBA 5551). If source image doesn't have alpha, it generates RGB 565 PNG.

• -16a PATH:

  16 bit PNG with 4 bit alpha (RGBA 4444). If source image doesn't have alpha, it generates RGB 565 PNG.

• -16 PATH:

  It is as same as -16a .

• -16i PATH:

  16 bit PNG with 4 bit alpha (RGBA 4444) and 256 colors palette. If source image doesn't have alpha, it generates RGB 565 PNG.

• -32 PATH:

  24/32 bit PNG. It tries several compression option and reduce file size.

• -32i PATH:

  24/32 bit PNG with 256 colors palette. It tries several compression option and reduce file size.

• -p16m PATH:

  Preview mode of -16m.

• -p16a PATH:

  Preview mode of -16a.

• -p16 PATH:

  Preview mode of -16.

• -p16i PATH:

  Preview mode of -16i.

If you want to create final output, you should use texture mode. But this file truncates unused bit, so it is not good for preview.

For example, RGBA 5551 format has only 1 bit for alpha channel. The solid color pixel's alpha value become transparent.

Before Convert

After Convert

Preview mode completes the lack of this information. You can see final image on your favorite viewer.

All samples are uploaded on Flicker.

• Lenna: Original File (512x512): 525,521 bytes (converted from tiff file by using Preview.app)

• Smooth UI Kit (400x300): 103,410 bytes (converted from PSD file by using Photoshop CS5)

  This file is created by Matt Gentile. Released under Creative Commons Attribution 3.0

• 473,226 bytes (lightpng lenna.png -o 2 -32 lenna_32.png)

• 75,570 bytes (lightpng uikit.png -o 2 -32 uikit_32.png)

• 274,089 bytes (lightpng lenna.png -o 2 -16 lenna_16.png)

• 59,532 bytes (lightpng uikit.png -o 2 -16 uikit_16.png)

• 178,214 bytes (lightpng lenna.png -o 2 -32i lenna_32i.png)

• 56,506 bytes (lightpng uikit.png -o 2 -32i uikit_32i.png)

• 153,976 bytes (lightpng lenna.png -o 2 -16i lenna_16i.png)

• 34,289 bytes (lightpng uikit.png -o 2 -16i uikit_16i.png)

This program experimentally supports compressed texture generation (pvr/atc). It needs PVRTexLib and/or AdrenoSDK. They are available here:

• https://www.imgtec.com/powervr/insider/powervr-utilities.asp
• https://developer.qualcomm.com/develop/

AdrenoSDK is proveded in Windows execution. So you need wine to install and extract needed libraries/headers.

To create this version, set following flags to scons command:

• --no-opensource:

  Enable Texture Compression format support

• --PVRTexLib=DIR:

  Enable PVRTC Texture Compression convert/preview feature. Default is "./PVRTexLib".

• --AdrenoSDK=DIR:

  Enable ATITC Texture Compression convert/preview feature. Default is "~/.wine/drive_c/AdrenoSDK".

It adds following output options:

• -pvr PATH:

  4 bpp PVRTC compressed texture file (PVR version 3). PVRTC should be square powered by two sizes. lightpng adds extra spaces to match this rule.

• -lpvr PATH:

  4 bpp PVRTC compressed texture file with legacy format (PVR version 2).

  It is a same format as Apple's texturetool command.

• -ppvr PATH:

  Preview mode of PVRTC.

• -atc PATH:

  8 bpp ATITC compressed texture file. It is a just raw bit data. It doesn't contain picture size information.

• -fatc PATH:

  8 bpp ATITC compressed texture file with header information.

• -patc PATH:

  Preview mode of PVRTC

If you use this option, use this program for internal use. You can't distribute your modified source code and/or binary in public. If you want to reuse my source code to your product, please sent me Amazon.com or Amazon.co.jp e-mail gift to my address (yoshiki at shibu.jp).

I am reading the license of Ardeno SDK, but it maybe prohibit linking with open source tool (legal English is very difficult). If I find any good way to link with open source code, I will merge this code into main.

• (lightpng lenna.png -ppvr lenna_pvr.png)

• (lightpng uikit.png -ppvr uikit_pvr.png)

• (lightpng lenna.png -patc lenna_atc.png)

• (lightpng uikit.png -patc uikit_atc.png)

This source code is released under MIT License.

This source code is created by other many people's help. Including my family, friends of Python/Sphinx communitiy, coworkers (DeNA, ngmoco) and so on.

I learned good dithering algorithm from Sc4Freak's blog (https://adtsai.blogspot.com/2011/03/high-quality-dithering-for-windows.html) and his source code. My implementation is an improvement/portable version of his code.

And thanks from Greg Roelofs (libpng author) and Jean-loup Gailly and Mark Adler (zlib authors), I can implement it in a day.