Creation of dance movements from the music is an important human art form, in this work we propose a framework of dance generation from music with beats (DGFMB) which is able to create dance sequences adhering physical plausibility at same time remaining realistic and rhythm to the input music. DGFMB consists of with several key modules including a music beat module, a music semantics module and a dance generation module. Besides, we introduce an updated public AIST++ dataset [10] with extra annotation of the beat for dance sequences. In the evaluation, the proposed framework demonstrates a remarkable improvement in the comparison with the state-of-the-art models.