Existing methods of 3D dense face alignment mainly concentrate on accuracy,thus limiting the scope of their practical applications. In this paper, wepropose a novel regression framework named 3DDFA-V2 which makes a balance amongspeed, accuracy and stability. Firstly, on the basis of a lightweight backbone,we propose a meta-joint optimization strategy to dynamically regress a smallset of 3DMM parameters, which greatly enhances speed and accuracysimultaneously. To further improve the stability on videos, we present avirtual synthesis method to transform one still image to a short-video whichincorporates in-plane and out-of-plane face moving. On the premise of highaccuracy and stability, 3DDFA-V2 runs at over 50fps on a single CPU core andoutperforms other state-of-the-art heavy models simultaneously. Experiments onseveral challenging datasets validate the efficiency of our method. Pre-trainedmodels and code are available at https://github.com/cleardusk/3DDFA_V2.