Welcome new contributor. I am just writing my comment as an answer, and expanding on the observation of Prof. Arapura. For a smooth, projective scheme $X$ over a field $k$, the space of first order deformations of $X$ as a $k$-scheme is naturally isomorphic to the $k$-vector space $H^1(X,T_X)$.

For a locally free sheaf $E$ on $X$ of rank $r>0$, the "Atiyah extension" is an element in $\text{Ext}^1_{\mathcal{O}_X}(E,E\otimes_{\mathcal{O}_X}\Omega^1_{X/k})$ whose "characteristic polynomial" has coefficients in $\text{Ext}^r_{\mathcal{O}_X}(\mathcal{O}_X,\Omega^r_{X/k}) = H^r(X,\Omega^r_{X/k})$ that are equal to the images of the degree $r$ Chern classes of $E$ under the cycle class map from $\text{CH}^r(X)$ to the "de Rham cohomology groups" $H^r(X,\Omega^r_{X/k})$. In particular, the "pairing" of the Atiyah extension with an element of $H^1(X,T_X)=\text{Ext}^1_{\mathcal{O}_X}(\Omega^1_{X/k},\mathcal{O}_X)$ gives an element in $\text{Ext}^2_{\mathcal{O}_X}(E,E)$. As proved in references on deformation theory (e.g., Illusie's book on the cotangent comples or Grothendieck's earlier LNM on the good 2-term truncation of the cotangent complex), this element is the obstruction to lifting $E$ to a locally free sheaf on the corresponding first-order deformation of $X$. Altogether, this defines a $k$-linear map,
$$
H^1(X,T_X) \to \text{Hom}_k(\text{Ext}^1_{\mathcal{O}_X}(E,E\otimes_{\mathcal{O}_X}\Omega^1_{X/k}), \text{Ext}^2_{\mathcal{O}_X}(E,E)).
$$

Now consider the case when $E$ is an invertible sheaf. The $k$-linear map above reduces to the form,
$$
H^1(X,T_X) \to \text{Hom}_k(H^1(X,\Omega^1_{X/k}),H^2(X,\mathcal{O}_X)).
$$
For a polarized K3 surface $X$ over $\mathbb{C}$, using the trivialization of $\Omega^2_{X/\mathbb{C}}$, this map is "equivalent to" the usual cup-product pairing,
$$
H^1(X,\Omega^1_{X/\mathbb{C}}) \to \text{Hom}_{\mathbb{C}}(H^1(X,\Omega^1_{X/\mathbb{C}}),H^2(X,\Omega^2_{X/\mathbb{C}}).
$$
The cup-product pairing is a perfect pairing. Thus, this $\mathbb{C}$-linear map is an isomorphism of $\mathbb{C}$-vector spaces.

Finally, since $H^1(X,\mathcal{O}_X)$ vanishes for a K3 surface, the cycle class map from the Picard group to $H^1(X,\Omega^1_{X/\mathbb{C}})$ is injective (in characteristic $p$, the kernel of this cycle class map obviously contains the $p$-power image of the Picard group, so this is one place characteristic $0$ is crucial). Thus the induced $\mathbb{C}$-linear map,
$$
H^1(X,T_X) \to \text{Hom}_{\mathbb{Z}}(\text{Pic}(X),H^2(X,\mathcal{O}_X)),
$$
is surjective. In particular, if the Picard rank is at least $1$, if we fix a saturated rank $1$ sublattice, there are first order deformations such that the corresponding "obstruction map" on $\text{Pic}(X)$ has kernel precisely equal to this saturated rank $1$ sublattice. Since also $H^2(X,T_X)$ vanishes, these first order deformations extend. Therefore, there are deformations of the K3 surface over a $1$-dimensional base such that the Picard lattice of a very general member of the family equals the saturated rank $1$ sublattice, i.e., the Picard rank is $1$. In applications, typically we choose the saturated rank $1$ sublattice to be generated by an ample divisor class, so that the deformation is a family of projective K3 surfaces (rather than just Kaehler K3 surfaces that are not projective).

Some part of this is included in the proof of Claim 3.5 of my note about Artin's axioms.

Artin's axioms, composition and moduli spaces
http://www.math.stonybrook.edu/~jstarr/papers/moduli4.pdf