I don't think anyone has written down a definition of 'weak' [[nlab:extranatural transformations]] for bicategories, so let's do that. There are no surprises, but all we'll be doing is defining one particular kind and showing that they compose with ordinary pseudonatural transformations, so there may well be unexpected wrinkles elsewhere. Let $T \colon K^{op} \times K \to L$ be a (pseudo)functor and $\ell$ be an object of $L$. An $ob K$-indexed family $\beta_a \colon \ell \to T(a,a)$ of morphisms of $L$ is **extranatural** if for each $f \colon a \to b$ in $K$ there is an invertible 2-cell $$ \array{ \ell & \overset{\beta_a}{\longrightarrow} & T(a,a) \\ \mathllap{\beta_b} \downarrow & \cong\,\beta_f & \downarrow \mathrlap{T(a,f)} \\ T(b,b) & \underset{T(f,b)}{\longrightarrow} & T(a,b) } $$ satisfying some fairly obvious axioms corresponding to the usual ones for pseudonatural transformations: 1. The assignment $f \mapsto \beta_f$ is natural with respect to 2-cells $m \colon f \Rightarrow g$. 1. $\beta_1$ is an identity, modulo the unitors of $T$. 1. For $a \overset{g}{\to} b \overset{h}{\to} c$, $\beta_{h g}$ is equal to a suitable pasting composite of $\beta_g$ and $\beta_h$. (It's pretty obvious if you try to draw it: paste $\beta_g$ and $\beta_h$ along $\beta_b$ and complete the square (so to speak) using the bifunctoriality isomorphisms of $T$.) Now let $S$ be another functor and $\alpha \colon S \Rightarrow T$ be an ordinary pseudonatural transformation. We want to show that the family $\alpha_{a,a} \circ \beta_a$ is again extranatural. The naturality (1) and unit (2) axioms are obvious from the diagrams; the only non-trivial part is axiom (3). To draw the diagrams here would be quite painful, so I'll just point out that $(g,h) \cong (a,h)(g,b) \cong (g,c)(b,h)$ in $K^{op} \times K$, and that this gives you the equations you need between $\alpha_{g,h}, \alpha_{g,b}, \alpha_{a,h},$ etc.