Midjourney paints, ChatGPT pixels — and my client could tell

Both can make images now, so I ran a fair test: a client moodboard, six asset types, one set from Midjourney and one from ChatGPT's image generation. Same briefs, same style references. The client picked without knowing which was which, and the picks weren't close.

For anything atmospheric — hero backgrounds, conceptual illustrations, a moody product-in-a-scene shot — Midjourney took every round. The compositions have intent. The light does things. ChatGPT's versions were competent and flat, the visual equivalent of clip art that went to art school.

But for the utilitarian half, ChatGPT quietly won. Revising an image by conversation ('same scene, move the mug to the left, make it morning light') is something Midjourney still makes awkward. And for simple layouts with actual readable text — a save-the-date card, a labeled diagram — ChatGPT's text rendering finally crossed the line from joke to usable. Midjourney still draws letters like it's never met one personally.

The workflow cost matters too. Midjourney means parameters, seeds, rerolls, a Discord-shaped learning curve that I've stopped resenting and started respecting. ChatGPT is conversation. For a quick 'make me something like this' from a non-designer, the easier tool is the right tool.

Verdict: brands and portfolios are Midjourney territory, and the client could feel it without knowing why. Process images, internal decks, anything iterative or text-bearing goes to ChatGPT. If your job is one or the other exclusively, congrats, you need one subscription. Most of us need both, annoyingly.