The Texas Two-Step Behind My 15-Minute AI Image Workflow

Chrysalis Part 3: The tech stack and two-layer prompt framework that turns inconsistent AI images into a fast, repeatable visual brand

This post shares the exact AI image prompt framework I use to create a consistent visual brand in under 15 minutes, including a free download.

Editors Note: This is Part 3 of the Chrysalis series on building a personal brand with AI. Part 1 covered the chaos and the breakdown. Part 2 revealed the brand identity and the style guide. This is the operating manual. How I actually produce every image you see in this publication, and the prompt framework you can use for free. (Part 1: When AI Told Me to Go Touch Grass / Part 2: Stop Chasing Pretty. Build Sticky.)

---

The very best way to learn about AI is to pick a project and just do it. It will be messy. It will be frustrating. It might make you want to throw your computer across the room.

Today is Part 3 where I share the workflow and the system I built that took me from days of prompt wrangling to under 15 minutes per picture.

These three posts tell the story about branding with AI, but the bigger story is how the lesson applies to everything you build when you adopt AI.

Foundations first. Tools second.

This is the phoenix part of the story.

---

Tool Time and “RTB”

First, the tools that make up my tech stack. I know you want to skip to the prompts, but wait! There’s more. The tools matter because understanding why each one earns it spot is what keeps you from chasing the next shiny object and shouting “squirrel!”

I’m not a developer and I don’t code. I’m a former Navy helicopter pilot who also spent 15 years in tech strategy and operations at places like AWS, Intuit and Qualcomm. My “tech stack” (just the tools you use…nothing fancy) is built for someone who needs to create without losing half her day doing it. Getting a solid system down makes everything else easier.

Having a system provides a home base to return to after experimenting and exploring. As a pilot, we’d call out “RTB” to fellow fliers. Return to Base.

This is my RTB:

Claude (Sonnet and Claude Opus) - my command center. (soon to include Claude Code). I use this tool for brain storming, prompt engineering, analysis and critical review of my work. It’s not perfect at everything, but I invested a lot of time to have it act as a coach/editor/critic. The investment really matters because training up a tool and then starting over with a new chatbot every week is like firing your best employee because someone else has a shinier resume. I use other chatbots as “external” consultants to double-check my work. They get personas, too. I never just drop something in and ask for a review. I always give them a character prompt like, “You’re a critic who has never read my work. Where does the reader get bored?”

Gemini, ChatGPT and Perplexity - my research consultants. I use Perplexity heavily for fact finding and research. They all provide second opinions and get called in for specific jobs, but not the whole workflow.

Kittl.ai with FluxPro - my art studio. I tested over fifteen tools before I landed here. Kittl.ai gives me access to FluxPro2 HD, which is the only model that I’ve found that can capture the chemical soul of 19th centruy cyanotype photography instead of just slapping a blue filter on everything. And for the record, I have no affiliation with FluxPro2, Kittl, Claude or any other tool I mention. Just sharing what worked for me.

CarouselBot - does one thing really well. Turns content into carousels for Substack and LinkedIn. Clean, reliable, and no-drama. Made by one of my favorite developers, Karen Spinner.

Microsoft Word - yes, Word! I draft everything here. Old school? Sure. But I’ve been writing for years and I’m not disrupting a workflow that works because someone told me I should be using something trendier. It doesn’t connect to anything else and everything in it stays with me. I can use it offline and think without being distracted.

NotebookLM - my research summary consultant. I used this in combination with all of my research results. It’s great to be able to drop everything into one place and search just that work, as well as summarize and quiz myself. I heard Claude Code does something similar, so will report back.

Notion - the newest addition to my tech stack that offers one place for everything. This is where I’ll be centralizing everything including my brand documents, prompt libraries, content calendars and research notes. I’m still building it out and learning it. There’s some cross-over capability with Claude Code, but not as intimidating as CLI (command line interface) and dark mode. Which Notion, of course, can be used in dark mode as well.

Hardware and OS - Mac.

Eight tools. That’s the whole stack. No tool-of-the-week habit for my home base. I’d rather go deep with these tools that are working than get distracted with all of the recommendations I see every day.

I have security in background and am quite cautious including tools that connect to everything. I experiment and play with new tools when I can. But, every hour I spend learning a new tool is an hour I’m not creating. This is the same philosophy that I write about. Foundations over fads. My stack is just my brand philosophy applied to software.

---

Prompt Rot Blooper Reel

Before I get into what is working, let’s chat a bit about what didn’t.

None of the tools could pick up my Anna Atkins cyanotype meets AI Flight Plan vision quite as I had it in my head. I love that there are actually many variations of the blue, depending how long it was exposed to light. This gave me a lot to work with.

Anna Atkins circa 1842-50

I think I was 70 pages (put in Word) of back and forth prompt conversations with Claude before we hit what I called, “prompt rot.”

Prompt rot: when a model accumulates so much contradictory context that outputs become unpredictable and unusable.

This is the point where the model has so much contradictory context built up that the outputs start getting really strange and nothing looks like what you asked for. How did that guy at the computer in the top right get rendered from my prompt?! I don’t know if I laughed out loud more or used profanity more during this process.

When Prompts Go Wrong

I struggled with three things specifically:

Facial Features. <expletive representing my true feelings removed> Every AI model want to show you a face. A beautiful, symmetrical, algorithmically perfect face. Ugh. The more I tried to describe my pilot’s face, the worse it got. The models would either ignore my description or produce something that definitely was not in my head. And when it would get close, it couldn’t repeat it. Ever. Which brings me to my next big challenge.

Consistency. Each new generation felt like starting over. The hair would change. The flight suit would shift colors and random patches. The cyanotype would drift from photography to water color or sketches or something like it was designed by a committee of robots.

Cost. Testing fifteen different tools across four days adds up fast. Every credit spent on a failed experiment is a lesson, but that tuition gets expensive.

The breakthrough came from a decision that felt like giving up but turned out to be my path to success.

I made my pilot faceless.

No face means no facial features to fight over. The pilot is always seen as a secondary character to the message my article is trying to convey. The moment I made that call, my image generation success rate went from frustrating to reliable. Sometimes the constraint IS the solution.

---

The Texas Two-Step

Being from Texas originally, I’m just going to work that in whenever I can. And now some of you know why I write “y’all” in my Notes without giving it a second thought.

This is my Two-Layer System that simplified everything.

“One shot” prompting with just one prompt and expecting an image to match your prompt the first time is a gutsy move, Mav. Having a consistent visual brand requires architecture. It doesn’t have to be exotic to work. A simple system that locks what stays the same and frees what can change.

Layer 1: The Core Foundation

This is a very detailed, locked prompt that goes into every single image I generate and it doesn’t change. Here is a summary of the five things you can see in all my images:

1. The art medium. Victorian-era cyanotype photographic process made with paper coated with light-sensitive iron salts.

2. The pilot character. Female pilot with long brown hair in a ponytail, warm, natural skin tones on any visible hands, olive green flight suit.

3. The flora. White botanical silhouettes rendered exactly like Anna Atkin’s cyanotypes. Sharp photographic images showing full structural detail.

4. The light source. Every image uses one contained light source drawn from the color palette.

5. The color constants. Prussian blue as primary field. Seven specific colors in the color palette, each with a defined role. Hex codes in the prompt, not descriptions.

That’s the foundation that lives in a document and I pasted it into every prompt. Then I add in the variation layer.

Layer 2: Variation on a Theme

This is where each image becomes its own thing based on the message from my article. I have six variation templates, each designed for a different mood or post type.

A darker exposure for authority and strategy posts with a deep Prussian blue, dramatic contrast. A light exposure for possibility and hope with luminous cyan and airy tones. I use warm amber tinting for craft and heritage, while using olive undertones for foundation building. And so on.

Each variation specifies the exposure level, the accent colors, the lighting quality, the scene and what the pilot is doing. Because every variation sits on top of the same Core Foundation layer, every image is unmistakably part of the same brand, while none of them look identical.

The Workflow

Write the post. Find the theme that emerges. Open the Core Foundation doc and choose the right Variation template for the mood. Add a brief scene description to that post. Paste everything into FluxPro2 HD on kittl. Generate. Evaluate. Adjust the Variation only. Never the Core Foundation. Repeat until it works or until I’ve hit my self-imposed attempt limit and accept that “good enough” is actually good enough.

The last part is real. I put in an attempt limit for myself because without one, I will just do it all day long. The image for Part 2 of this series took ten attempts. Ten. And I have a rule that once I hit my limit, I ship the best one and move on. Turns out that perfection is just procrastination with better branding.

Here are some of those ten attempts. I found the huge rabbit hole that it insisted on rendering hilarious since I truly was in a pretty big hole trying to figure all of this out.

Failed attempts creating Part 2's Image

I really struggled with both clear prompts and the tools to execute my vision. If it wasn’t for Dallas Payne casually mentioning in her branding post that she ended up using FluxPro2 on Kittl.ai, I might never have returned from the depths of that rabbit hole. Only FluxPro2 HD on Kittl.ai was able to give me what I need. I still struggle a bit with consistency and not being able to build memory into using what I’ve already done, but it gets the job done.

---

Download Prompt Guide Here

Just because I ended up with a detailed prompt guide doesn’t mean you need something quite so in-depth. You could just as easily use the Two Step Prompt guide with foundational elements + variation in just a couple of sentences, depending on how detailed you want the image. I was able to create the image for this post in under ten minutes!

The actual “core foundation” and “variation on a theme” prompts are quite detailed and specific to my thumbnail art. Below I’m sharing the prompt guide and framework for you to use to create your own. If you need inspiration or further branding tips, check out AI Meets Girlboss, Dallas Payne, Mia Kiraki 🎭 , or Daria Cupareanu.

Two Step Prompt Guide

192KB ∙ PDF file

Download

Free download to experiment building your own prompt system.

Download

---

For Princess Bride Fans

In the immortal words of Inigo Montoya, “No, there is too much. Let me sum up.” For my personal system and workflow, this is my foundation:

• The tools that matter: Claude for brainstorming and prompt development. FluxPro2 HD on kittl.ai for image generation. Everything else is supporting cast.

• The decision that changed everything: Make the pilot faceless

• The system that makes it sustainable: Core Foundation locked. Variation flexible. Attempt limit enforced.

• The philosophy underneath the brand: Know who you’re writing for and who you’re NOT writing for. Foundation over fads. A consistent brand isn’t made from better prompts. It’s made from a better system. Get yours today!

The Chrysalis, The Rabbit Hole and The Phoenix

About Me

LinkedIn