In the nlab, it is stated that a compactly generated space could equivalently be defined as a space $X$ such that there is a family of compact Hausdorff spaces $K_\alpha$ such that for any space $Y$, a set function $f:X \to Y$ is continuous iff for every continuous $u:K_\alpha \to X$ we have that $fu$ is continuous.
It is clear that this implies the usual definition, on the other way I was told the following argument:
For each non-open subset of $X$, choose a $C$ and a continuous $t: C \to X$, such that the preimage is not open. (If such a $C$ exists). Since $X$ has a set worth of subsets, this forms a set. Now assume $f: X \to Y$ isn't continuous. Then there exists a continuous function $t:C\to X$ and an open set $U$ in $Y$, such that $t^{-1}(f^{-1}(U))$ is not open. In particular $f^{-1}(U)$ satisfies our above assumptions, so we have a chosen $t$ and $C$ such that $t^{-1}(f^{-1}(U))$ is not open.
The problem with this argument is that the choice of $C$ is taken from a proper class of spaces. Is there anyway to bound this class (say, by using the cardinality of $X$) so that the only axiom of choice is needed, instead of global choice? Better yet, can this be proven without using the axiom of choice?