In this paper, a self-cancelling method is proposed to embed data into multiple layers of a spatial scalable coded video. Correlation of prediction mode from multiple layers are analyzed and exploited to offset the distortion introduced at the base layer(BL) when embedding data at the enhancement layer (EL). Specifically, in the base layer, the prediction modes are divided into two groups, where one group encodes '0' while another encodes '1'. Data embedding in the enhancement layer is designed to compensate the errors introduced in the base layer. Experiment results show that the scalable coded video can effectively carry additional payload in multiple layers while maintaining the video quality and bit rate. In the best case scenario, when 104,141 bits are embedded into the BasketballDrive (BL: 1280 × 720 and EL: 1920 × 1080) video sequence, the bit rate is slightly increased while insignificant degradation in perceptual quality is observed.