My approach to crafting prompts has changed over the years, becoming more subtle on one hand but more descriptive on the other. When I started experimenting with prompts for AI image generators, I often underestimated how much detail was required to get high-quality results. Over time, I’ve realised that specificity is key—especially if you’re aiming for photorealism or want the AI to capture complex concepts.
If you want to create something truly photorealistic, you need to think like a photographer. Specify details like camera speed, focal length, lighting conditions, and even environmental factors like haze or noise. For instance, asking for “A street at dusk” might yield generic results, but adding “captured with a 50mm lens, bokeh lights in the background, subtle lens flare” makes a world of difference. These details help the AI understand not just the scene but also the style and mood you’re trying to evoke.
On the flip side, when testing the capabilities of a model, simplicity often works best. Stripping the prompt back to its essentials can reveal a lot about the strengths and weaknesses of the AI. A minimalist prompt like “A red apple on a wooden table” allows you to see how well the model handles textures, colours, and lighting without overwhelming it with too many variables.
Over the years, I’ve experimented with a number of personal test prompts, each reflecting whatever had captured my imagination at the time. For a long while, my go-to prompt was “Cat on the Moon in a spacesuit with Earthrise in the background.” It was whimsical yet technically challenging, a great way to test how well models could juggle multiple elements. Unfortunately, too many models struggled to get it right—some missed the spacesuit, others misrepresented Earthrise, and a few inexplicably turned the cat into a dog.
Lately, I’ve gravitated toward something simpler yet equally interpretative: “Dog wearing sunglasses on a train.” This prompt offers plenty of room for creative variation. Will the dog be anthropomorphic, sitting upright in a suit, or simply lounging by a window? What kind of train will it be? A sleek modern bullet train or a nostalgic steam locomotive? The open-ended nature of this prompt makes it ideal for testing video models as well as image generators, as it allows you to assess how the AI interprets storytelling elements alongside visual coherence.
The journey of refining prompts is as much about understanding AI as it is about understanding creativity. Whether you’re aiming for high fidelity or exploring a model’s artistic tendencies, the prompt is your most powerful tool. And as AI models grow more sophisticated, so too will the art of prompting.
No responses yet