Generative fill + prompt + near-solid background = edges

I'm sure this applies to other use-cases as well, but to me it appears most evident in dark backgrounds when using a text-prompt. Generative Fill applies a relatively strong tone-difference, where you can easily see the edges of the Generative Fill, whereas Content-Aware fill does much better job. Observe 2 x Generative Fill vs. 2 x Content-Aware Fill:

Idea: include a dataset of Content-Aware background fills in Firefly training data for better handling of near-solid backgrounds, or just utilize Content-Aware Fill with Generative Fill in low-detailed areas.