As CodeTalker is trained on the dataset without blinking motions, it is impossible to produce these motions. One way to solve this is to retrain it on datasets with these blinking motions, another one I think is to convert the results into FALME parameters and add motions using the parametric method, then convert back to mesh.
As CodeTalker is trained on the dataset without blinking motions, it is impossible to produce these motions. One way to solve this is to retrain it on datasets with these blinking motions, another one I think is to convert the results into FALME parameters and add motions using the parametric method, then convert back to mesh.