Keras Facenet Model Finetuning

[Brihith Shenoy]

So im using keras-facenet(tf) for face embedding extraction(512) currently im using the pretrained one. 
So basically I give an image example saying this person A and whenever a comparission image is given i use distancing and make a conclusion. 
I have alot of pictures of that person instead of getting the embeddings
(rn i have built like a vector db with a persons embeddings so person A has multiple different Images whose embeddings are stored in the database) 
 

would i be able to fine tune the model so that i can add person classes and give more examples for that person. Can I extend it to multiple people example I retrain multiple images of the same person( different Images).
Would the accuracy be better that way?
Is it a right way to go?
Would some other modifications help me better?
Or am I missing anything.
Sorry if the question doesnt make sense