A Python script for converting images to AWBMv2 (Award BIOS Bitmap V2).

img2awbmv2.py

import os
import argparse
import numpy as np
from PIL import Image

# Format description from: http://fileformats.archiveteam.org/wiki/Award_BIOS_logo
MAGIC = b'AWBM'
PALMAGIC = b'RGB '

def parse():
    parser = argparse.ArgumentParser(description='img2awbmv2')
    parser.add_argument('file')
    parser.add_argument('--format', '-f',
        default='auto', choices=['auto', '4', '8'],
        help='use 4 or 8-bit variant. default: auto (deduce from the input image)')
    parser.add_argument('--save', '-s', default='')
    args = parser.parse_args()
    return args

def main(args):
    assert os.path.exists(args.file), "File not found!"
    if len(args.save) == 0:
        args.save = os.path.basename(
            os.path.splitext(args.file)[0]+'.awbm',
        )
    if os.path.dirname(args.save) != '':
        os.makedirs(os.path.dirname(args.save), exist_ok=True)

    # read image
    im = Image.open(args.file)
    nim = np.asarray(im)
    if nim.ndim == 2:
        nim = nim[:, :, None]
    unique_colors = np.unique(nim.reshape(-1, nim.shape[-1]), axis=0)
    
    # detect colors
    if args.format == 'auto':
        if unique_colors.shape[0] <= 16:
            colors = 16
        else:
            colors = 256
        print ('Total {} unique colors found, using {} colors.'.format(
            unique_colors.shape[0], colors
        ))
    elif args.format == '4':
        colors = 16
    elif args.format == '8':
        colors = 256

    sw = np.uint16(im.width).tobytes()
    sh = np.uint16(im.height).tobytes()

    pim = im.convert('P', palette=Image.ADAPTIVE, colors=colors)

    # RGB palette
    palette = np.asarray(pim.getpalette()[:colors*3], dtype=np.uint8) // 4

    if colors == 16:    # 4bit
        # convert to BGR
        palette = palette.reshape(16, 3)[:, ::-1].reshape(-1)
        # preparing data
        npim = np.asarray(pim)
        assert npim.ndim == 2
        cmp = 0x1
        fs = []
        for _ in range(4):
            f = np.packbits(npim & cmp, axis=-1)
            fs.append(f)
            cmp = cmp << 1
        data = np.concatenate(fs, axis=1).tobytes()
    else:               # 8bit
        data = pim.tobytes()
        assert len(data) == im.width * im.height

    # assemble
    paldata = palette.tobytes()
    with open(args.save, mode='wb') as f:
        f.write(MAGIC)
        f.write(sw)
        f.write(sh)
        f.write(data)
        f.write(PALMAGIC)
        f.write(paldata)
    print ('Saved to {}.'.format(args.save))
        

if __name__ == '__main__':
    args = parse()
    main(args)

Some quirks and tips:

1. Since AWBMv2 only supports 6bit intensity (0..63), there is an 8bit to 6bit quantization (L56).
2. The ``ADAPTIVE'' RGB to Palette conversion in PIL is usable but not the best. The index mode conversion in Photoshop has better visual quality IMO.
3. One can convert AWBM back to conventional formats (PNG) to validate the result. deark is pretty handy.