Prompt to Image: Book Reflections in Generative AI
- gawonlee
- Aug 26
- 4 min read
Updated: Dec 5
Chasing Images in the Midnight Glow
In 2022, I began using Midjourney for the first time and started generating images as a way to review the books I read. Diffusion models at that time had distinct personalities, and the same prompt produced completely different results depending on the model. I preferred Midjourney over DALL·E and Stable Diffusion, as its style aligned more closely with my taste.
Regardless of the platform, the most challenging part back then was the prompt itself. To get the image I wanted, I had to combine specific words and refine the phrasing carefully. It was a period that required a form of prompt engineering. Each model interpreted language differently, and a prompt that worked well in Midjourney often produced unrelated results in Firefly. Creating separate prompts for each model became a skill on its own.
Around 2024, the landscape began to shift. Across categories, from web-based productivity tools to content platforms, nearly all of them began offering built-in image generation. By 2025, Grok added it as well, and it became difficult to find a platform without AI tools.
With this shift, prompts became significantly simpler. Long chains of keywords were no longer necessary. Plain natural language was enough, and the models were able to interpret intent accurately and produce near-finished images.
Post-processing changed as well. Until early 2024, AI images often lacked texture or detail, so Photoshop corrections were still a major part of the workflow. But with newer models such as Firefly 2 and 3, Midjourney 6, and Imagen 3, the quality of the initial output improved dramatically. Even Photoshop’s introduction of Nano Banana made post-processing more enjoyable and expressive.
Today, it is normal to move through two or three tools to create one image. What used to be a necessity has become an option, and that option is remarkably powerful.
Looking back at the recent history of AI image generation, it is clear how quickly everything has changed. There was a time when writing a good prompt was a key skill. Now we have entered a period where simply describing what we want is enough. Image generation has become a natural part of everyday workflows rather than a separate task.
(About False: People fall for false beliefs not because of personal flaws, but because cognitive biases are amplified by today’s information environment. Algorithms, scattered media, and social dynamics reinforce what we already want to believe, making misinformation feel credible. The key is to focus not on individuals, but on the psychological, social, and technological systems that shape everyone’s beliefs.)
