Dall-e 2 would give pretty realistic looking photos, even though it was terrible with text and accuracy. v3 is was better with that stuff, but every photo looks super dramatic and like a "digital art" prompt. E.g."photo of a beautiful snowy mountain"
https://www.bing.com/images vs https://labs.openai.com/
I know the other models might do it better, but I already pay for gpt4 so have dall-e 3 for free