Reversible data hiding in encrypted images (RDHEI) consists of embedding data in the encrypted domain. In current state-of-the-art methods, most of them use least significant bit (LSB) substitution or prediction, but fail to embed a significant amount of information. Recently, a new class of RDHEI method, based on most significant bit (MSB) substitution, has emerged. By exploiting the natural correlation between pixels in the clear domain, it is possible to have a payload close to 1 bpp with a very high image quality, without adding overhead. In particular, in the approach based on embedded prediction errors (EPE-based approach) [6], the authors propose to embed the prediction error location information in the encrypted MSB-plane. In this paper, we present a huge-capacity RDHEI (HC-RDHEI) method. In fact, we are interested in improving the proposed EPE-based RDHEI approach by using recursively other bit-planes, from MSB to LSB as long as it is possible. Indeed, depending on the image content, bit-planes can easily be predicted, and so most of them can be substituted by bits of a secret message. According to the obtained results, the payload can be much higher than 1 bpp (median equal to 1.749 bpp, on average 1.836 bpp, and 5.408 bpp in the best case), while preserving perfect reversibility.