Recently, a fast error diffusion halftoning algorithm using look-up tables (LUT) was proposed to speed up the multiplication of error filter coefficients. In this letter, we propose another LUTbased error diffusion halftoning algorithm which is more flexible in terms of the size of the LUT that can be used and thus allows for a more optimal tradeoff between halftone quality, processing speed, hardware complexity and parallelizability. Furthermore, the aggregate error in the proposed algorithm can be computed with different bitdepths for the different errors. As an example, we present a variant of the Floyd-Steinberg error diffusion algorithm which consists of two 256K bytes LUTs and the calculation of the modified input requires 1 addition and 2 table look-up operations per pixel processed.