This post is part of the Building a Brand Character With AI series. Next up: How to Build a Consistent Character Library from Scratch and How to Get Consistent Character Images from a Single Reference.
TL;DR: Generic AI visuals aren't a brand — they're a placeholder. I tested five illustration styles on my own face, picked a winner, and built a reusable image library from scratch. Here's the exact workflow, what broke, and the system that made Pinkie finally look like herself.
There’s a specific kind of cringe that hits when you look at your own content and think: this could be anyone’s post.
Mine hit on a Tuesday. I was scrolling through Substack, landing on newsletter after newsletter with stunning, recognizable thumbnails like The AI Break, ROBOTS ATE MY HOMEWORK, The AI Rabbit Hole. Jenny Ouyang’s 3D Pixar energy. Consistent characters with actual personality. Then I looked at my latest cover — another AI-generated “photo-realistic Pinkie” — and felt absolutely nothing.
It didn’t feel like AI Meets Girlboss. It felt like a stock photo that had a bad day.
I’d been spending more time generating individual images than building anything that could scale. Every cover was a new conversation, a new prompt, a new round of hoping it looked right. There was no system. Just vibes and hope, which is a terrible content strategy.
The turning point was Jenny Ouyang's post on consistent hero images. She wasn't sharing "the perfect prompt." She was sharing a process — one that treats your character as infrastructure, not a one-off output. That idea hit differently.
One test later, I had five versions of Pinkie, a very clear winner, and a new understanding of which AI tools actually do what.
Hey, I’m Pinkie. I help creators build recognizable Substack brands that drive consistent subscriber growth. I share the exact branding systems, prompts, and workflows I use to transform ideas into distinctive, repeatable output. If you’re new to AI Meets Girlboss, welcome! Honestly, this might be the best time to join. start here! 🩷🦩
What you’ll learn in this post:
Why testing multiple styles upfront is faster than committing to one and hoping for the best
How to extract reusable style rules from reference images using the Blind Illustrator Method
Which tool handles consistency vs. which handles rules — and why they’re not the same job
The two prompts you can copy and run today to start your own style test
The Problem With "Just Generate Something."
Most creators approach visual branding backwards. They generate an image, like it enough, use it a few times, then slowly drift into inconsistency as each new prompt produces something slightly different.
The result isn’t a brand. It’s a mood board with no memory.
The fix isn’t finding a better prompt. It’s building a system that stores your character’s rules — so any prompt you run stays consistent, not because you got lucky, but because you defined what “consistent” actually means.
I broke down that same rules-first approach for brand fundamentals in I Had My Brand Rules Written Down. They Were in Six Different Places — worth a look before you build your own character library.
Jenny Ouyang’s workflow made this concrete: build a base library of poses (front, side, standing, talking, walking), store them in one workspace, feed references to your AI tool to extract style rules, then reuse those rules every time. The prompt is almost secondary. The system is the brand.
I explored how character consistency works at the brand level in The Visual Brand System That Replaced My Morning Prompt Panic — it’s a useful read before you build your library.
That insight changed how I approached the whole experiment.
How I Tested 5 AI Illustration Styles to Find a Consistent Brand Character
Before locking a visual identity, I needed to see what Pinkie actually looked like across different aesthetics. Not in theory — in reality, on my actual face.
I went full teenage-graphic-designer mode. Shoutout to my Avril Lavigne fan-site era, which apparently taught me more about aesthetic obsession than I realized. Here’s my exact workflow:
Step 0: Pick the base.
My original profile photo. Pinkie, Founder of AI Meets Girlboss, about to get the most thorough makeover of her AI life.
Step 1: Choose five styles.
After exploring Pinterest and Google, I landed on: Comic Book, 3D Pixar Cartoon, Fashion Illustration, Caricature, and Retro Poster. Different enough to be useful. Distinct enough that there'd be a clear winner.
Step 2: Turn references into rules with the Blind Illustrator Prompt.
This is the move most people skip. Instead of describing a style in vague language (”make it look like a fashion illustration”), I fed each reference image into ChatGPT with this:
Analyze this [style name] illustration style as if describing it to a blind illustrator. Identify the exact visual rules: shapes, proportions, textures, color palette, lighting, line quality, shading, character features, poses, composition, and mood.
[Attach reference image]What came back wasn’t a vibe. It was a structured breakdown — specific, transferable, repeatable.
Step 3: Turn rules into a transformation prompt.
I then used those rules to generate a Style Transformation Prompt:
Use these rules to write a prompt beginning with: “Transform the person in the input image into a [style] illustration.” Include line work, shading, textures, colors, facial expression logic, pose, and atmosphere. Write it as a complete, polished prompt that an image model can use to recreate the style consistently.
[Insert rules here]This gave me five reusable style prompts, each built from real visual rules, not guesswork.
Step 4: Generate — and learn which tool actually does what.
Here’s where the experiment got honest. ChatGPT kept reinventing Pinkie’s face like she was in witness protection. Different nose, different hair, different proportions — every generation. Charming, but unusable as a brand system.
What ChatGPT was genuinely excellent at: analyzing styles, extracting visual rules, and producing high-quality prompts. That’s where it earned its place in this workflow.
The moment I moved to Nano Banana, consistency snapped into place. Same face, same colors, same proportions — across styles, across poses, across dozens of generations. Nano Banana is built for character continuity. For my workflow, that’s not a nice-to-have. It’s the whole point.
Step 5: Pick the style that feels like you.
This part needs human judgment. Emotion plays a role in brand identity. Here are the five final styles I created with NanoBanana:
⬆️ I’m curious — really! Before you scroll further, please do tell me which style is your favorite! ⬆️
The Style That Won (And Why It Had to Be Her.)
When Fashion Illustration Pinkie appeared, I knew immediately. She had that feminine confidence — the kind that walks into a room and doesn’t explain herself. She felt like AI Meets Girlboss: stylized, direct, a little extra, zero apologies.
Visual style only works when it's downstream of an actual personality. If you haven't locked that part yet, Turn Your Personality Into Your Brand Voice And a Claude Skill in 45 Minutes.
I’m fully aware I chose a couture-forward visual identity in a space where the top-performing newsletters skew masculine and minimal. That’s exactly why I chose her. If Pinkie’s showing up, she’s showing up in the right outfit.
The early versions were close but not quite finished. I used ChatGPT — not for generation, but for rule refinement.
I tightened the style rules:
consistent proportions,
controlled texture,
clear distinction between what could stay loose (background washes)
and what had to stay locked (face, posture, palette).
Palette lock matters just as much as face lock. If your colors are still doing whatever they want, Picking Your Brand Colors for Your Substack Is the Most Overlooked Thing walks through the same logic for your palette
Then I returned to Nano Banana, and she finally looked like herself. Then I built the library: standing, walking, talking, presenting, thinking, close-ups, seasonal outfits. Pinkie’s visual identity system — consistent, stylized, repeatable, and mutation-ready — was done.
That library-building process has its own post. If you’re ready to go beyond style-testing and actually build your character system, How to Build a Consistent Character Library from Scratch is the next step.
I even rebranded one of my older post covers to test the identity in the wild:
What I Learned From 50+ Images.
1. ChatGPT and Nano Banana are not competitors. They’re partners.
ChatGPT is exceptional at analyzing styles, extracting visual rules, and producing high-quality prompts. Nano Banana is unmatched at character consistency. The workflow that works: ChatGPT for rules, Nano Banana for generation. Stop expecting one tool to do both jobs.
2. Systems outlast prompts.
A prompt is a one-time instruction. A system is infrastructure that remembers. Jenny’s biggest lesson applies here: once you have your style rules locked, any prompt becomes consistent. The rules are the real asset — not the images they produce.
I saw exactly this play out once the system (not just the prompts) was locked — Visual Brand System With AI: From 64 to 200 Substack Subscribers In 2 Weeks is what happened next.
3. What works for someone else might not work for you.
Jenny gets incredible consistency with ChatGPT. I did not. Her workflow is brilliant — it’s just not mine. AI isn’t a recipe you copy. You learn the tools, find what holds your character together, and build from there. The destination is the same. The route is yours
Try Both Prompts Today.
Run this on one style first. Get one clean output. Then repeat for each style you want to test.
Prompt 1 — Blind Illustrator Analysis (paste into ChatGPT with your reference image attached):
Analyze this [style name] illustration style as if describing it
to a blind illustrator. Identify the exact visual rules: shapes,
proportions, textures, color palette, lighting, line quality,
shading, character features, poses, composition, and mood.Prompt 2 — Style Transformation (paste your rules output back into ChatGPT):
Use these rules to write a prompt beginning with: "Transform the
person in the input image into a [style] illustration." Include
line work, shading, textures, colors, facial expression logic,
pose, and atmosphere. Write it as a complete, polished prompt
that an image model can use to recreate the style consistently.Once you have your winner, the next move is building a library of poses so you stop regenerating from scratch every time. This post walks you through exactly how to do that.
And if you want to skip ahead to generating consistent versions of your character from a single reference image — without building a full library first — this one is for you.
In the comments…
Which illustration style fits your brand — and have you ever tested it on your own face? Drop your style below: fashion illustration, 3D cartoon, retro poster, something else entirely. If you’ve already run the prompts and generated something, share it. I want to see which version wins for you.
A huge thank you to Jenny Ouyang for sharing her method and sparking this whole experiment. Go read her post — it’s worth it.
Talk later, Girlbosses.
Pinkie 🩷
If today’s post made you think about your own visual brand, I am launching The Branding Flock, a 12-month visual brand program for Substack creators who are done guessing.
Full breakdown posts, real prompts, real outputs. Your color palette, your logo, your brand character, your content system — built step by step alongside the flock.
33% off closes after the first 100 seats. The discount lives inside your quiz result.
Growth comes from practice
I’m committed to actually using these systems — not just testing them. So here are a few well-earned shoutouts to the Substackers who helped shape this edition:
Thank you to Claudia Faith for the One Prompt System Outline, which helped shape this newsletter.
And a thank you to Mia Kiraki 🎭 whose Make AI Argue With Itself Prompt helped me choose today’s title.
Love this, I'm on the journey too, and you're ahead of me sorting it all out. Yay!
So next post is more on this?
So I used ChatGPT 5.1—not for generation, but for refining the style rules:
consistent proportions
controlled texture
clarity on what can stay “unfinished” (background washes)
clarity on what must stay consistent (face, posture, palette)
This was a great post and inspired me to create my own brand images. I love this and thank you for sharing.