At killallhumans.be, we believe that achieving perfection with GPTs isn’t about getting it right on the first try—it’s about collaboration, iteration, and learning. Our methodology for refining FLUX text-to-image prompts transforms basic ideas into vivid, compelling imagery through a dynamic, back-and-forth dialogue with the AI.

Step 1: Set the Intention

Before diving into prompt refinement, it’s vital to instruct the GPT about our intentions. For example, here’s our initial input:

“let’s make the ultimate FLUX prompt generator. idea is that i have an AI assistant that can craft a perfect text to image prompt based on a basic idea. Modus operandi is you ask for the idea, I tell you and you create a first prompt. From there we’ll pingpong the output and you enhance the image either by analysing or based on my comments.”

This clear instruction establishes the framework for a collaborative process, setting expectations right from the start.

The Methodology

1. Initial Vision Statement:
   Begin with a basic description of the image you want, setting the overall direction.
2. Generate & Analyze:
   Input your initial prompt into the FLUX text-to-image model. Then, bring the output back to ChatGPT for a detailed analysis. This dialogue helps identify what’s working and what needs adjustment.
3. Iterative Refinement:
   Based on feedback, adjust your prompt. This “ping-pong” of input and feedback sharpens the prompt until it perfectly aligns with your creative vision.
4. Extracting the Blueprint:
   Once the image meets your standards, review the conversation to extract key insights. Compile these into a set of instructions that encapsulate the successful strategies.
5. Deploy Custom GPT:
   Use the refined instruction set as the system prompt for your custom GPT. Now, all you need is a basic idea—the model handles the rest, consistently generating high-quality, refined prompts.

Food for Thought

Most users expect instant, perfect output on the first try. But the real magic happens when you treat GPTs as sparring partners rather than as instant answer machines. Embracing this iterative, feedback-driven process not only yields superior results but also deepens your understanding of AI’s capabilities. It’s a shift from seeking perfection in one go to achieving it through continuous learning and collaboration.

Case Study: Transforming Fairytales into 1900’s Realistic Horror

As a practical example, I applied this methodology to create a custom GPT that transforms fairytale and disney stories into 1900’s realistic horror imagery—shot as if captured by a vintage camera. Now, by simply typing in a fairytale title, the custom GPT produces a series of eerie, atmospheric visuals.

Below are a couple of images generated using this method:

This approach demonstrates how iterative feedback can evolve a simple idea into a refined, creative output, unlocking new avenues for innovation in AI-generated art.

---

By embracing this collaborative, iterative process, you’re not just refining a prompt—you’re mastering the art of communication with AI. Let’s keep pushing boundaries and learning together.