This is really subjective. AI art generators -- just like real artists -- are usually only good at doing certain styles, and even then, sometimes only in certain contexts.
Because they are only amalgamating information from images they were trained on, it's impossible for AI image generators to do much beyond the scope of mimicking their data set. So, their proficiency depends - at least in large part - on the images in the data set.
If you were trying to generate a photorealistic image, for example, Midjourney would likely do a MUCH better job of this than Adobe Firefly. Even if you just tried different art styles, you'd probably get wildly different results.
So declaring a "winner" here is pretty pointless.