The elbow method plots the K-means inertia as a function of k. You pick the value where the curve stops dropping sharply, forming an elbow: adding more clusters then brings almost no gain. The elbow is not always clear, so you should cross-check with other criteria.
Example
inerties = [KMeans(n_clusters=k, n_init=10).fit(X_s).inertia_ for k in range(1, 10)]