5 types: Cyberpunk version of "The Colossus of Lodoss"

These are re-generated versions of previously created works for viewing purposes. From left to right: Z-Image 3:4, Flux 2 Pro 3:4 1K, GPT Image 2 1024 × 1536 low, Byte Plus Seedream 4 3:4 2K, Nano Banana 3:4. See the differences in the style produced by the image generation AI. Even when specifying the same thing—a cyberpunk version of the "Colossus of Lodoss" resembling a skyscraper—the results differ significantly in both presentation and size. Looking at the prompts, it seems quite difficult to place something that doesn't exist in reality into a port city setting. Which do you prefer? Personally, I think the Flux.2 body is beautiful, but the Z-Image layout is cool. If you consider it a recreation of an ancient colossus, Seedream's design is also hard to dismiss. Generally speaking, Nano Banana-type images are often rather uninspired, while GPT-type images tend to be overly elaborate. Z-Image is rough. Flux.2 is uncooperative. That's my impression.