This paper introduces the theoretical framework allowing for the binary quantization index modulation (QIM) embedding techniques to be extended towards multiple-symbol QIM (m-QIM, where m stands for the number of symbols on which the mark is encoded prior to its embedding). The underlying detection method is optimized with respect to the minimization of the average error probability, under the hypothesis of white, additive Gaussian behavior for the attacks. This way, for prescribed transparency and robustness constraints, the data payload is increased by a factor of log2m. m-QIM is experimentally validated under the frameworks of the MEDIEVALS French national project and of the SPY ITEA2 European project, related to MPEG-4 AVC robust and semi-fragile watermarking applications, respectively. The experiments are three-folded and consider the data payload-robustness-transparency tradeoff. In the former case, the main benefit is the increase of data payload by a factor of log2m while keeping fixed robustness (variations lower than 3% of the bit error rate after additive noise, transcoding and Stirmark random bending attacks) and transparency (set to average PSNR=45 dB and 65 dB for SD and HD encoded content, respectively). The experiments consider 1 h of video content. In the semi-fragile watermarking case, the m-QIM main advantage is a relative gain factor of 0.11 of PSNR for fixed robustness (against transcoding), fragility (to content alteration) and the data payload. The experiments consider 1 h 20 min of video content.