Images: describe it, get it, fix it.

Image generation in ChatGPT stopped being a toy when it learned to follow instructions, render text, and edit the image you already have. For everyday business visuals, it's often faster than finding a stock photo — and it's yours.

01 What it's genuinely good for

• Social and marketing graphics — announcement cards, simple ads, seasonal posts, with your text rendered correctly on the image.
• Mockups — "my logo on a work van," "this menu redesigned," "what would this room look like painted sage."
• Diagrams and explainers — "a simple diagram of how water moves through this system, labeled."
• Edits to real photos — upload a photo and ask: remove the background, brighten it, extend the sky, take out the trash can.

What it's still weak at, honestly: precise brand-exact logo reproduction, fine text at small sizes, and the tenth revision of the same image staying consistent with the first. Know this going in and you'll be delighted instead of frustrated.

02 Prompt like an art director

Same principle as writing: the brief is the skill. A useful structure — subject, style, composition, mood, text:

Then iterate conversationally: "warmer," "less clip-arty," "make the badge red and move it left," "same image but for Facebook's wide format." Each revision builds on the last — you're directing, not re-rolling dice.

03 The edit workflow for real photos

04 Use, rights, and taste

You can use what you generate commercially under OpenAI's terms — but you can't trademark your way out of a generated logo looking like other generated logos. For identity (logo, core brand), use AI for exploration and a human designer for the final. For the everyday river of graphics — posts, flyers, one-off promos — this is now the fastest tool you own. And a taste rule: if an image screams AI, regenerate with "simpler, more restrained" until it doesn't.