← Back to all free guides
A free guide by Mariah Brunner ↗ INSTAGRAM ↗ TIKTOK

Why Your AI Designs
Always Look Like AI
(And How To Fix It)

AI design isn’t bad. Your prompts are. The framework I use to write prompts that produce elevated visuals, plus 50 free Nano Banana Pro prompts from a designer who did branding for Warner Music and Spotify.

The image tools are great now. A generic prompt still gives you a polished image that screams AI, because you asked for the average.

Below is the brief prompt I paste before making any visual, then where to grab 50 proven prompts and how to build your own library from them.

Why AI Design Looks Like AI

Give an image model "a modern app screen" or "a beautiful logo" and it lands on the middle of everything it has seen. Technically clean, creatively dead.

Good designers never brief that way. They ask for a specific thing in a specific style, with references and constraints. Do the same with AI.

The Litmus Test

If your prompt could describe 10,000 different images, you’ll get all of them mashed together. Write one that could only describe about ten.

The Brief To Paste First

Paste this into Claude or ChatGPT before you open any image tool. It interviews you, then hands back a prompt specific enough to escape the average.

Visual Brief Writer
You are my AI design brief writer. I'm about to use an AI image tool (Nano Banana in Gemini, ChatGPT Images, Midjourney or similar) to make a visual. My results usually look generic. Your job is to write a brief specific enough to escape the AI average. If I attach reference images, my logo or brand colors, study them first and use what you see so you don't ask me twice. Ask me these 6 questions ONE AT A TIME. Wait for each answer. 1. What am I making? (logo, social post, ad, product mockup, illustration, packaging) 2. Where will it live? (Instagram carousel, paid ad, pitch deck, website hero) and what size or aspect ratio. 3. What 3 feelings should it create? Push me past words like "professional" to something like "premium and a little intimidating." 4. Two real references: brands, photographers, designers, films or album covers whose look I want to sit next to. 5. One anti-reference: a style I absolutely don't want. 6. Non-negotiables: exact copy, colors, products, people or layout rules. If an answer is vague, push back once and ask for something more specific. Then write the brief in exactly this format: USE CASE AND SIZE: FEELINGS: (each of the 3, plus the visual choice that creates it) COLOR PALETTE: (3 to 5 colors with hex codes and the role of each) TYPOGRAPHY: (named typefaces or a precise style, plus the exact words that must appear) COMPOSITION: (where the eye lands first, how dense, what's in front and behind) LIGHTING AND TEXTURE: (camera, lens, light source, materials, finish) BORROW FROM REFERENCE A: BORROW FROM REFERENCE B: AVOID: NON-NEGOTIABLES: THE PROMPT: one 4 to 6 sentence paragraph I can paste straight into the image tool. Describe the scene concretely (subject, setting, camera, light, style, text) instead of listing adjectives. Rules: - Don't name living artists in THE PROMPT. Describe the qualities of their style instead. - Keep any text that must appear in the image short and in quotation marks. - Give me one alternate version of THE PROMPT with a bolder creative direction.

Why Two References

Zero references gets you the average. Ten confuses the model. Two strong ones put it in a specific corner of the design world.

Grab The 50 Prompts

Amir Mushich is a designer and creative director who says he did branding and ads for Warner Music, PepsiCo and Spotify. In February 2026 he posted 50 of his design prompts for Nano Banana Pro, free, as an article on X. Comment BANANA on my video and I’ll send it, or open it here.

Don’t only copy them. Study how they’re built: how specific each one gets, which references they lean on and what constraints they set. Then write your own in the same shape.

They’re written for Nano Banana Pro, but they work as a strong starting point in other tools too. In the Gemini app, Nano Banana 2 is now the default image model. (Thurrott) On a Google AI Pro or Ultra plan you can still switch an image to Pro by regenerating it from the three-dot menu. In ChatGPT Images or Midjourney, expect to tweak a few words.

Build Your Own Library

01

Save every prompt that works

When an image comes out right, save the exact prompt in a Claude Project or a doc called "Visual Prompts That Worked."

02

Tag it by use

Social, ad, deck, illustration, product shot. Over time it becomes your own version of Amir’s list, already tuned to your brand.

03

Start from a winner

Next time you need a visual, open a prompt that already worked and tweak it instead of starting blank. You get speed and a consistent look.

After a month of this, you’ll spot generic AI design in other people’s work instantly. That eye for the difference is the real skill.