AI for Beginners
✦ Getting Good · May 2026

Make it look good

6 min read · images, logos, and social tiles

Words aren't the only thing these tools make. You can ask for images too — a logo sketch, a graphic for a post, an illustration, a quick mockup. For a small business with no designer on call, that's a genuine unlock.

🎨 What it can actually make

Type a description, get a picture. Social media graphics, simple logos and icons, illustrations, product mockups, background images, "what could this look like" concepts. Which tool you're in matters more here than usual: ChatGPT and Gemini make images right inside the chat (Gemini's image model goes by the deeply serious name of Nano Banana). Claude — the one I use for words — doesn't do photos or artwork at all, though it's handy for diagrams, charts and code-based mockups. So for pictures, I hop across to one of the others. Either way, you're describing — not drawing.

🗣️ How to ask for a good one

Vague in, vague out. Don't say "a logo for my cafe." Say: "a simple, modern logo for a coastal cafe called Salt — warm sand and ocean-blue tones, a small wave motif, clean and friendly, on a white background." The more you describe — style, colours, mood, what it's for — the closer it lands.

Describe it like you're briefing a designer who can't see inside your head. Because that's exactly what you're doing.

🔁 Treat the first one as a sketch

You'll rarely nail it first go, and that's fine. "Make it warmer." "Less busy." "Try it without the text." "Same idea, but flat and minimal." Iterate. Trying again is basically free — which is the whole point. You can explore ten directions before lunch.

⚠️ The honest limits

Two things to know. Text inside images used to come out reliably garbled. The newest models (ChatGPT's Images 2.0, Gemini's latest Nano Banana) are much better at it — but "much better" isn't "always right", so proofread every word, or add real text yourself afterwards. And for anything you'll use commercially — a final logo, a brand identity — treat the output as a starting point and check the tool's usage rights. Brilliant for ideas, drafts and "good enough" social graphics; for your actual brand mark, it's a brainstorming partner, not the final word.

🧰 Where this pays off

The social tiles you'd otherwise skip. A quick graphic for a flyer. Concepts to show a real designer so they get you faster. Placeholders while you build. Mood and direction before you spend a cent. It doesn't replace a designer for the big stuff — it just means the small stuff actually gets done.

I still believe taste and judgement are the human's job; knowing what good looks like is the whole game. But for getting from blank page to "oh — something like that," it's hard to beat.

Sources & further reading: OpenAI — Introducing ChatGPT Images 2.0 · TechCrunch — Images 2.0 and text rendering · Google — Gemini image generation (Nano Banana) · Claude Help Center — Can Claude produce images?
Facts checked July 2026.
Previous How to sound less like AI Next Claude vs Cowork