AI design isn't bad. Your prompts are. The framework I use to write prompts that actually produce elevated visuals, plus where to grab 50 free Nano Banana Pro prompts from an ex-Warner Music designer.
AI design has a reputation for being bad. The truth is the tools are great. The prompts are bad. A generic prompt gives you a polished but obviously AI-looking image, every time. That's not the model failing, that's the prompt asking for generic.
This guide does two things. First, it walks through the framework I use to write prompts that produce elevated AI visuals (works for Nano Banana Pro, Midjourney, Sora, anything). Second, it points you to 50 specific paste-ready prompts from a designer who's worked with Warner Music and Spotify — the cheat code that compresses six months of prompting practice into an afternoon.
Every AI image tool produces the median of its training data when you give it a generic prompt. "A modern app screen." "A beautiful logo." "A startup ad." The model averages every modern app screen, every beautiful logo, every startup ad it's seen. The output is technically polished and creatively dead.
Good designers don't ask for the median. They ask for a specific thing in a specific style with specific references and specific constraints. The same exact rule applies to AI prompts. The more specific the input, the further the output gets from generic AI slop.
The litmus test: if your prompt could describe 10,000 different images, you're going to get one that looks like 10,000 different images mashed together. If your prompt could only describe ~10 specific images, the output will look like one of those ten.
Paste this into Claude before you ask any AI image tool (Nano Banana Pro, Midjourney, Sora, DALL-E) to make a visual. Claude will interview you and hand back a brief specific enough to produce a non-generic result.
Copy this prompt
You are my AI design brief writer. I'm about to use an AI image tool to make a visual. Most of the time my output looks generic. Your job is to fix that by writing me a brief specific enough to escape the AI median. Ask me these 6 questions one at a time. Wait for each answer. 1. What am I making? (one line. Logo, social post, ad, product mockup, illustration, packaging, etc.) 2. What is it FOR? (the specific use — Instagram carousel, paid ad, pitch deck, business card) 3. What 3 emotions should it create when someone looks at it? (real emotions — "premium and a little intimidating," not "professional") 4. Two specific real-world references. Two brands, photographers, designers, films, or albums whose visual style I want to feel adjacent to. Be opinionated. 5. One anti-reference. One visual style I absolutely don't want to look like. (Helps the AI know what to cut.) 6. The non-negotiables. Specific colors, copy, imagery, dimensions, or composition rules. Once I answer, write me a paste-ready visual brief in this exact format: TITLE: USE CASE: EMOTIONAL TARGET: (the 3 emotions + the specific move that creates each) COLOR PALETTE: (3 to 5 colors with hex codes + the role each plays) TYPOGRAPHY: (named typefaces or specific styles) COMPOSITION: (where the eye should go, density, hierarchy) LIGHTING / TEXTURE: (the cinematography choices) REFERENCE A: (one specific brand or work to borrow from, with what specifically to take) REFERENCE B: (another specific brand or work to borrow from) ANTI-REFERENCE: (what to actively avoid) NON-NEGOTIABLES: THE PROMPT: (4 to 6 sentences in second person, opinionated and specific, paste-ready into Nano Banana Pro, Midjourney, Sora, or any other AI image tool) Be ruthlessly specific. If I'm vague in any of my answers, push back. The whole purpose of this brief is to escape the AI median.
Why two references, not zero or ten
Zero references gives you the median. Ten references confuses the model. Two strong references give it enough constraint to land in a specific zone of the design world without averaging itself back to nothing.
Amir Mushich (ex-Warner Music, ex-Spotify) posted 50 of his top Nano Banana Pro prompts to X for free. They are paste-ready. The link is in your DM once you comment BANANA on the video, or you can grab it directly here:
https://x.com/AmirMushich/status/2019073081881293302
What's worth doing is not just copying his prompts but studying their structure. Notice the level of specificity. Notice the references he cites. Notice the constraints. Then write your own using the same shape. That's how you go from "I have 50 prompts" to "I can write any prompt I need."
If Nano Banana Pro isn't your default tool yet: it's a strong AI image tool, especially for stylized illustration and brand work. Most of Amir's prompts also work with minor tweaks in Midjourney and Sora, so don't feel locked in.
Once you've used Amir's 50 prompts as a starting point, the real upgrade is building your own paste-ready library specific to your brand and your work.
Step 1. Every time you generate an image that genuinely works, save the prompt that made it. Drop it in a Claude Project or a Notion page called "Visual Prompts That Worked."
Step 2. Tag each saved prompt by use case: social, ad, deck, illustration, product shot. Over time you build your own version of Amir's list, but pre-tuned to your brand.
Step 3. Next time you need a new visual, you start from a working prompt in your library and tweak it, instead of starting from a blank brief. Speed and consistency go up at the same time.
The shift in your taste
After a month of running specific, reference-driven prompts, you'll start to spot generic AI design in everyone else's work instantly. That's the real win. You become someone who can tell the difference between AI slop and AI used by someone with taste. The market is going to reward that distinction heavily over the next two years.
Go All In
If this guide helped, but you’re looking to go deeper, I got you!! My 30-Day Challenge takes you from saving AI tips you never use to actually building with AI, step-by-step.
I show you exactly how I automated two e-commerce brands, my social media, and most of my personal life, then hand you the agents, workflows & systems to do the same. I’m teaching you every single thing I know with one lesson and one build a day.
Join the Challenge →Doors close June 21. We start June 22.
© 2026 Mariah Brunner. All rights reserved.