Largest coding unit (LCU) deserves a unique Lagrangian multiplier (lambda) for its rate-distortion optimization (RDO) due to the texture and motion compensation which represents the different rate-distortion curve. Using the same lambda to whole LCUs in a picture cannot obatain optimal coding performance. How to reallocate an appropriate lambda to an individual LCU is an urgent challenge. This paper proposes a LCU-level Lagrangian multiplier adaption (LLMA) algorithm. Firstly, a motion compensation information entropy (MCIE) of 16 x 16 block is calculated by a simple binary information entropy statistics from the source. Secondly, MCIE is normalized as a standard normal distribution since the strong assumption of the video sources obey the Gaussian distribution. Thirdly, an adaption factor is assigned for a LCU which multiples on the picture-level lambda from the system calculation. Experiments show that the proposed LLMA achieves remarkable performance in low-delay and random-access configuration respectively. Compared with VTM 13.0 low-delay common test condition, the LLMA algorithm gains 1.81\\% in BD-rate.