PixelProse is a comprehensive dataset of 16M synthetically generated detailed and accurate image captions developed to power future vision-language research.