As one of the representative works, RAFT-Stereo designs an iterative strategy based on GRU and realizes the state-of-the-art (SOTA) performance. However, this iterative strategy is difficult to effectively distinguish between long-time state features and short-time state features, which leads to the coupling problem. To solve the feature coupling problem in the iterative process, we design the Mogrifier LSTM module for iterative disparity updating based on RAFT-Stereo. Specifically, we design the Mogrifier LSTM to contain the standard LSTM framework, as well as an interaction header for long and short time state sequences. We used this design for the replacement of RAFT-Stereo, and designed a refine network for the output of RAFT-Stereo to correct the disparity results.