How to generate images with the Visual Prompt Library

Choose a style

Open the gallery. Use search (for example neon, poster, food), filter by category, tag or colour, or press Ctrl+K for quick search. Save the styles you like.

See the examples

On a style page, the example tabs show several images with the same style but different subjects. This shows what stays fixed (the visual system) and what changes (subject, text). Click an image to view it full screen.

Fill in the prompt builder

• Start from an example: pick one as a base, then adapt it. The Randomise button mixes details from the other examples.
• Subject: who or what is in the image. Be specific (for example “a street vendor at a morning food stall”).
• Main text: the headline to show on the image. Type it exactly as you want it and keep it short.
• Ratio: 16:9 (YouTube/web), 9:16 (Reels/TikTok/Stories), 1:1 or 4:5 (Instagram).
• Fields left blank are chosen by the AI to suit the style.

Copy and paste

Press Copy prompt, open your image tool, paste and send.

• ChatGPT: the “Open in ChatGPT” button pre-fills the prompt. Ask it to “generate an image”.
• Gemini (Nano Banana): copy first, open Gemini, then paste. Gemini links cannot carry a prompt.
• Midjourney: paste after /imagine and add --ar 16:9 (or another ratio) at the end.
• Claude: does not generate images, but can help write or translate text before you paste it into an image tool.

Refine

• If the text on the image is wrong, retry and be explicit: “Render the headline exactly: ...”.
• If the result is too close to the example, change the subject and location more noticeably.
• If the style is lost, paste the full style.json to the AI and ask it to use that style as the “locked style”.
• Do not mention real logos, brands or people.