As an isolated image, I prefer the Dall-E 2 sample (of the basketball player) to all the others on that page, aesthetically. Due perhaps to having used a more fine-art-heavy training corpus, or a less specific correspondence to prompts?
I appreciate your preference (I like things heaver on impressionism too), but I don't think it's due to the corpus but rather the model capability. DALL-E 2 is just behind in capability. Of course we won't know until October but I suspect you could prompt v3 to get a style closer to v2 if you wanted.
This is actually an interesting issue the Midjourney team has thought a lot about. As each version has gotten "better", ie more realistic, there's been some loss of the "artistic" side. There are a lot of users who still use the old V2 model (compared to the most recent V5) specifically because of how "bad" it was. The grimy and less coherent parts are what they're actually looking for, instead of a more precise or perfect looking result. This has led to there being flags for adding in more stylisation or "weirdness" or being able to choose between more realistic or more artistic versions of models.
it looks less like an 'oil painting' though. Looks to me like one of those stencil, spray-painted images you see people selling at tourist attractions.
Perhaps the Dall-E 2 unintentionally got that better.
