If you've ever typed something into an AI image tool and gotten back... not quite what you meant, you're not alone. I can't tell you how many tokens I burned through trying to iterate my prompt directly in Nanobanana that I didn't need to! Since most AI models have vastly improved their image-generation capability (particularly ChatGPT!), the gap between what you imagined and what you got back isn't a tool problem. It's a prompt problem.

The good news is that you don't need a photography background to fix it. (But if you do, check out my other photography pro post on image prompts!)

Why your prompt matters more than the tool

AI image tools (ChatGPT, Canva's Magic Media, Nanobanana, Shutterstock AI) all work on the same principle: the clearer your instructions, the better the output. A vague prompt gets a generic image. A structured prompt gets a campaign-ready visual.

Think of it like briefing a freelancer. The more context you give them, the fewer rounds of revisions you need. (And in the AI universe, the fewer tokens – money – you’ll use!)

Copy-and-paste image prompt

Answer the bracketed questions, copy your finished prompt, and paste it into your preferred tool. Start with your first attempt, see what comes back, and refine from there. That iteration is the skill; the first image is just a starting point. Be sure to use the "thinking" and not "instant" mode to get better results. You'll burn up more tokens, but in the end, not as many as you would have with a generic prompt.

Create a [VISUAL STYLE: realistic photo / flat illustration / 3D render] of [WHO: describe the main subject, e.g., "a nurse"] who is [WHAT THEY'RE DOING, e.g., "talking with a patient”] The setting is [WHERE, e.g., "a bright modern medical practice clinic"]. The mood is [FEELING, e.g., "confident and collaborative"]. Use [YOUR BRAND COLORS, e.g., "#47737e, #482b4a, #d0a358”] along with clean neutral tones. Leave [WHERE: left side / right side / bottom / top] open for text overlay. This image is for [WHERE IT WILL LIVE, e.g., "a LinkedIn post / email header / event flyer"]. The image size should be [DIMENSIONS, or say “LinkedIn post, Facebook post, Instagram post, etc.]. Avoid: extra fingers, distorted hands, cheesy stock expressions, cluttered backgrounds, warped text on screens, low resolution. Output: high resolution, campaign-ready. Accessibility: Make sure the background color or image behind any text of any color have the appropriate contrast ratios as indicated in WCAG 2.2 AA standards. Use this Contrast Checker: https://webaim.org/resources/contrastchecker/. Further constraints: **[ADD ANY BRAND-SPECIFIC GUIDELINES HERE, such as: "do not show pets that are indoors with collars on them", "do not put stethoscopes on the veterinary technicians or veterinarians", or "be sure to use a diverse representation of people – by age, race, ability, gender"]. **

Quick reference

VISUAL STYLE: Realistic photo, cinematic still image, flat illustration, sketch, cartoon, oil painting, water color, pop art, or 3D render

WHO: Your member, your audience, your subject

WHAT THEY'RE DOING: An action tied to your message (climbing a mountain, reading a newspaper on a Kindle, giving a business presentation, etc.)

WHERE: A setting that fits your industry (ex. a nurse's station in a clean, brightly lit modern hospital)

FEELING: The emotion you want the reader to feel (happy, concerned, triumphant, serene, defiant, tender, playful, ominous, reverent, giddy, ethereal, satisfied, angry, furious, sad, anxious, content, energized, etc.)

YOUR BRAND COLORS: Name them specifically in RGB (web use) or CYMK (print use)

WHERE (text overlay): (Skip this line if no text goes over the image )

WHERE IT WILL LIVE: LinkedIn, email, flyer, website, etc.

CUSTOM BRAND CONSTRAINTS: Any further brand constraints: Do not do not show pets with collars on them if they are indoors or do not put stethoscopes on the veterinary technicians or veterinarians; be sure to use a diverse representation of people – by age, race, ability, gender, etc.

A few guardrails worth knowing

• Make sure your images are accessible. There are two places that you should review: the contrast ratio between print and background colors/images, and ALT tags.
• Keep any numbers or metrics on screens generic or fabricated.
• If an image will be published, note that it is AI-generated where your AI policy requires it.
• Always review the output yourself; the model can still miss guardrails.

This prompt has produced better results for me as I have produced iterative versions for my projects. Give it a try and drop me a line to share how it went!