Yeah, the understanding of biology to go from DNA to full body reconstruction is very far away. What they might be thinking of doing is creating a database of face and matching DNA. Then when unknown DNA is queried it probably finds the most similar sequences and then blends some of the most similar faces together.
I was imagining that they were going to acquire a large dataset of DNA sequences paired with photos of the patient and then train a neural network to predict one from the other. I don't think that would be particularly successful either (you would need a tremendous amount of data and DNA sequencing is expensive), but it would be interesting to see the results.