The Mold, Not the Clay: Prompt Engineering for Creatives

Everyone is building generative AI apps. That’s not a criticism. It’s the current reality of the field. But there’s a problem that gets very little attention: creative coherence when working with AI models. What does it take to build an AI pipeline that respects your vision, world-building, and voice? AI opens an exciting new frontier for creatives, if we approach it with intention and take control of the tool.

Design the Rules, Not the Output

The most common misconception I see in AI-generated creative work is treating AI as an omnipotent and untamable force. Because we’ve already reached a quality threshold where even a vague instruction can return impressive results, most people tend to stop there and call it a day. But as creators, we need to dig deeper. The result will be the equivalent of a stock photo: technically acceptable and creatively inert. This principle applies both to images and prose.

AI needs constraints to produce coherent creative output, not freedom. The more precisely you define its rules, its voice, its structure, and its limits, the more surprising and alive the output feels within your vision.

Here’s where the mind shift begins. We’re no longer the one with hands on the clay. We’re designing the mold. The creative flow has moved upstream, into ideating rules and systems that define our goals.

This is where advanced prompt engineering takes center stage. While it’s true that we need to embrace a double role as a creative and a technologist, prompting can be lots of fun, and a deeply satisfying creative challenge in its own terms.

A Tool, Not a Mind

No tool is without limits. An AI system can generate thousands of unique combinations, but it needs a human driving its intent. LLMs don’t understand creative outcomes any more than a piano understands Chopin. They execute instructions within a shaped space. You are the architect, which means the quality is entirely in your hands.

My Own Live Example: A Museum That Builds Itself

I’ve always loved eclectic art collections and the Victorian cabinet of curiosities. I’m also a fan of weird fiction, and always enjoy the trope of fantastic places that appear out of thin air. When I started experimenting with AI, I had the idea of a museum that dreamed of itself. After some trial and error, I released The Ravensfield Collection.

The collection produces one artifact a day. It includes an image of the artwork itself and a piece of flash fiction about its mysterious origin. It even shares quotes from imaginary experts. Every piece is AI-imagined, but the models work within a precise narrative sandbox. It took me a long time, and a deeper understanding of prompting, to build that sandbox. The narrative universe coheres, not because AI is magical, but because I engineered it to be.

Under the Hood

The pipeline behind Ravensfield runs five sequential steps, each feeding the next:

1. A Madlibs-style generator assembles a randomized creative brief from curated word lists: an art movement, a visual technique, a story subgenre, a protagonist, a theme, and a fate.
2. Anthropic’s Claude receives the brief and produces a detailed conservation-style description of a fictional artifact.
3. That description fans out in parallel: Claude writes the full story and metadata while Leonardo.ai generates the image, both working from the same source material.
4. Everything gets stored atomically in a database: either all of it or none of it. No partial entries, ever.
5. A final vision-consistency check has Claude compare the generated image against the written story to catch any contradictions before the piece goes live.

It runs across multiple AI models and produces one coherent output, every day, without intervention.

Same Questions, Different World

Ravensfield is one answer to a personal creative goal. Yours will look different: different world, different voice, different rules. As with any other project, it should scream you.

As you explore the world of creativity and AI, you will probably face a lot of the same questions I did. The articles that follow are my answers. I hope they help you find yours.

The full repository is public if you’d rather start with the code.