If you’ve spent any time with AI image generators, you’ve run into this wall.
You have a clear picture in your mind. The lighting. The composition. The vibe. You can see it. But when you sit down to describe it in a prompt, you type something vague — “modern interior, professional feel, warm tones” — and the output looks like nothing you imagined.
The images aren’t bad, exactly. They’re just not what you wanted. And you can’t quite figure out why.
The problem isn’t the tool. It’s vocabulary.
Why Image Prompts Fail
AI image models are trained on enormous datasets of photography, art, design, and illustration — all tagged with specific professional terminology. When a model generates “editorial-style architectural photography with golden hour light and shallow depth of field,” it draws on patterns from thousands of images described exactly that way.
When you describe your vision in plain language — “modern and warm and professional” — you’re speaking a different language than what the model knows best. The output is technically coherent but not quite right. It’s the difference between telling a chef “something tasty” versus “pan-seared, finishing with a butter baste.”
The gap is vocabulary. You can picture the result. You just can’t find the words the model responds to.
The Pinterest Reverse-Prompt Trick
Here’s the workaround I use, and I’ve been teaching it to everyone who asks about image generation.
Step 1: Find your reference on Pinterest.
Browse Pinterest for images with the aesthetic you’re after. The specific industry or subject doesn’t need to match — you’re looking for the visual vocabulary. If you want architectural photography that looks like a design magazine, find a design magazine photo. If you want a product shot with a specific lighting style, find a product shot you love.
Step 2: Paste the image into Claude or ChatGPT.
Drop the image into a text-and-vision AI model. Both Claude and ChatGPT handle this well.
Step 3: Ask for the description and the prompt.
The exact ask I use: “Describe this image in detail, including the photography style, lighting, composition, and any relevant technical details. Then write a prompt that an AI image generator could use to recreate this style.”
Step 4: Use the output in your image generator.
Take what the AI gives you — not your original vague description, the vocabulary it surfaces — and paste it into your image generator.
What Comes Back
The first time I did this properly, for work with Padel Society, I needed photorealistic interior images of a padel club — courts, common areas, locker rooms — that would look like actual editorial photography rather than AI-generated renders.
When I pasted my reference image and asked for prompt language, here’s the kind of vocabulary that came back:
- “Architectural photography, interior shot”
- “Wide-angle lens, slight perspective correction”
- “Natural light from floor-to-ceiling windows, late afternoon golden hour”
- “High-end editorial style, similar to Architectural Digest”
- “Shallow depth of field on midground elements”
None of those phrases would have occurred to me. I would have typed “modern sports club interior, professional photography.” These prompts produced images that actually looked like the reference.
The same process works for any visual style: product photography, event imagery, brand mockups, social media visuals. Whatever the aesthetic, Pinterest has thousands of examples, and any text AI can translate them into prompt vocabulary.
The Meta-Insight: Using AI to Talk to Other AI
There’s a bigger principle here worth naming.
I teach this as one of two advanced AI techniques I call “meta prompting” and “reverse prompting.”
Meta prompting is having the AI write prompts for you. Instead of figuring out how to phrase a request to get what you want, you describe the desired outcome and ask the AI to generate the prompt. You’re using one AI to communicate with another AI more effectively.
Reverse prompting is what the Pinterest trick is. You already have the output you want — you just don’t have the prompt. You give the AI the result and ask it to reverse engineer the approach.
Most people never do either. They spend time trying to get better at prompting by intuition — typing and retrying until something works. That’s fine, but it’s slow. The faster path is recognizing when you can use an AI to solve the prompting problem for you.
When you see a great image someone else generated, ask the AI to write the prompt that produced it. When you have a reference with the style you want, use the AI to translate it. When you don’t know how to describe what you want, describe what you’re going for in plain language and ask the AI to produce a better version of that description.
You stop struggling with vocabulary. You let the translation happen.
When to Use This
The Pinterest reverse-prompt trick is most useful when:
- You know what you want visually but can’t find the technical language
- Your current prompts produce outputs that are “close but not right”
- You’re working in an unfamiliar visual style — product photography, architectural imagery, illustration styles you haven’t produced before
- You need to match a specific reference or brand aesthetic
It’s less useful when you’re happy with how your prompts already work or when you’re doing something highly abstract where visual references don’t directly translate.
For anyone doing serious work with AI image generation — business development visuals, product photography, content imagery, event materials — this is one of the first techniques worth building into your workflow.
P.S. Advanced prompting techniques for image generation and visual AI workflows are part of what we cover in the Two Hour Workday program.
Recommended for you
Want the full system? 25X is the flagship productivity system we teach.
Explore 25X →