Stop making generic images and start creating art that speaks to you
Most AI art tools today function like glorified dice. You roll the input, and you get a pretty picture. But what if the process was different? What if the AI didn’t just generate a visual, but actually asked you questions to understand your mood, your history, and your specific taste before creating anything?
That is the core promise of Mythograph Atelier, a new project born from a recent trip to Zaragoza, Spain, and a few key shifts in how we view software.
A museum visit that changed the rules
During a recent trip to Zaragoza, I visited the IAACC Pablo Serrano, a contemporary art museum housing sculptures and paintings. While the abstract works there were visually stunning, one piece stopped me in my tracks.
It looked like a child’s random scribble. Shapes and strokes with no obvious narrative. My first reaction was skepticism: “Can anyone just make random strokes and have it exhibited in a museum?”
I knew there had to be context, intention, or technique behind it, but I also felt a sense of unfairness. Art often speaks only to those who know how to read it. If a piece requires an art degree to understand, what happens to ordinary people?
Then I realised the beauty of it. Maybe not everyone needs to find meaning in every piece. Maybe a painting does not have to explain itself to everyone. And when you do find that connection, it feels like discovering a small treasure.
This sparked the first inspiration for the project: What if we could help people find abstract art that feels personally meaningful to them?
Imagine a painting where the visual elements carry the feeling of Marcus Aurelius’ quote:
You have power over your mind, not outside events. Realize this, and you will find strength.
Not as text on the image. Not as an obvious illustration. But as an abstract composition that somehow carries that feeling. Because the meaning can be the connection between the piece and the person looking at it.
From static menus to fluid conversations
My second inspiration came from thinking about the future of applications. When ChatGPT launched in November 2022, it felt like a powerful search engine. But after a year, I had an “a-ha” moment watching Satya Nadella talk about agentic AI.
Nadella described AI as a bridge between the user and the database, replacing conventional SaaS workflows and interfaces.
This changed how I imagine apps. Instead of fixed pages, fixed menus, and fixed buttons, we should be building dynamic interfaces. Consider an e-commerce site. Today, most still use fixed categories and product listing pages. They might recommend products based on cookies or past behaviour, but the layout remains largely the same for everyone.
AI can change this completely. You could start with a simple text box and write:
I’d like to buy a budget-friendly pair of shoes.
Or:
I just want to look at flower pots and get inspired.
The app could then shape itself around your intention. The categories, filters, explanations, and visual layout could be created dynamically for that specific session.
The power of a curious AI
The third inspiration came from a colleague introducing me to an AI agent skill called “grill me,” created by Matt Pocock. It is a simple set of instructions designed to interview a user relentlessly about a plan until a shared understanding is reached.
The skill instructs the AI to walk down each branch of the design tree, resolving dependencies one-by-one. It asks questions one at a time. It explores the codebase to find answers before proceeding.
Before starting a coding task, this skill is incredibly useful. It does not jump directly into implementation. It asks questions. It explores. It tries to understand the plan deeply before acting.
Obviously, we cannot exhaust users by asking question after question forever. A good app should not feel like an interrogation. But the core idea is powerful: An AI should be curious. It should make an effort to understand the user before producing the final result.
For Mythograph Atelier, this means the AI should not immediately generate an image after one prompt. It should ask a few thoughtful questions. It should understand the user’s taste, emotions, references, and the meaning they want to express. But it must also keep the experience smooth. Curious, but not tiring. Personal, but not complicated.
Early results and the path forward
The current version is still early, but the direction is becoming clearer. The user starts with an idea, a feeling, a quote, or a loose intention. The AI asks follow-up questions to understand the user better. Then it creates a prompt for an image generation model, with the goal of producing an abstract painting that is visually interesting and personally meaningful.
I want the final image to be something the user can explain to a friend. Not:
“I don’t know, AI made it.”
But more like:
“This mountain represents patience. The door is change. The empty space is uncertainty. The colors are calm because I wanted the painting to feel hopeful.”
During the development process, I used Codex as my coding agent to build faster. After a few back-and-forth iterations, I was able to generate an image with FLUX based on the description:
A mountain is not decoration; it is patient ambition. A door gives it a second force: a threshold into change. That is the idea behind the painting: I want something about trying to build meaning inside confusion.
This was one of the first moments where the project started to feel real. It was not perfect, but it had something.
Further trials produced another image with the description:
A calm, geometric dance where lines suggest nature’s flow and empty spaces let you choose your path.
These images are still not exactly where I want the project to be. They are early, imperfect, and there is a lot to improve. But they gave me the feeling that the direction is right, and that Mythograph Atelier can become something interesting with more iteration.
There are still many things I want to improve. I want the conversation flow to feel smoother. I want the AI to ask better questions. I want the final prompt to be richer, more visual, and more connected to the user’s answers. I also want the generated paintings to feel less random and more intentional.
My goal is to keep improving Mythograph Atelier step by step and see whether this small idea can become something emotionally interesting: An AI atelier where abstract art is not only beautiful, but personal.
Key takeaways
- Mythograph Atelier uses a conversational AI agent to ask questions about your emotions and references before generating art, rather than relying on a single prompt.
- The project aims to create abstract paintings where the visual elements connect to personal meanings, allowing users to explain the work to others.
- By combining the “grill me” methodology with dynamic interface design, the tool adapts the user experience to each individual session.
Stay ahead of AI. Get the most important stories delivered to your inbox — no spam, no noise.
