Image engine bakeoff: same prompt, two engines

Same "What costs more?" balance-scale concept (modeled on Luis's top post), generated on both engines. Both clean up his version (his had duplicated numbers and mangled text). Tell me which engine wins, and whether the double-label on nano-banana bugs you, then I generate the full set of 6 concepts in the winning style.

Nano Banana (Gemini 3 Pro Image)

2K · ~$0.18/image · truer brand purple
nano-banana version
Sharper headline, real CallSetter purple, lots of whitespace. Flaw: labels render twice (on the pan and below it). Fixable next pass.

GPT Image (OpenAI)

1024 · high · cleaner labels + icons
gpt-image version
No duplicate labels. Added a hard-hat worker icon vs a robot icon, reads instantly. Purple is slightly off-brand and it is lower resolution.