As you note, because $A$ is nilpotent, the characteristic polynomial is $f_A(x)=x^n$ with $n=\dim(V)$ (assuming your characteristic polynomials are monic; otherwise, it is $(-1)^nx^n$). THe minimal polynomial is $m_A(x) = x^k$, with $1\leq k\leq n$.
Express $f_A(x) = m_A(x)x^{n-k}$. Then $\mathrm{Im}(m_A(A))=\mathrm{ker}(A^{n-k})$. But $m_A(A)$ is the zero linear transformation, so the image is trivial. Thus, $\mathrm{ker}(A^{n-k})$ is trivial, so $A^{n-k}$ is one-to-one. This is impossible if $n-k\gt 0$ (since $A$, being nilpotent, has nontrivial kernel), thus $n-k=0$ and hence $m_A(x)=f_A(x)$.
Conversely, suppose that $m_A(x)=f_A(x)$. Let $\beta=\{v_1,ldots,v_n\}$ be a Jordan canonical basis for $A$; that is, $A(v_m) = v_{m-1}$ if $m=2,\ldots,n$, and $A(v_1)=\mathbf{0}$. Then it is easy to verify that $\mathrm{Im}(A^r) = \mathrm{span}(v_1,\ldots,v_{n-r})$, and $\mathrm{ker}(A^s) = \mathrm{span}(v_1,\ldots,v_s)$. Thus, if $f_A(x) = x^rx^{n-r}$, we have
$$\mathrm{Im}(A^r) = \mathrm{span}(v_1,\ldots,v_{n-r}) = \mathrm{ker}(A^{n-r}),$$
as required.
Related, this question characterizing the (finite dimensional) linear operators for which the minimal and characteristic polynomials agree.