Exploring ChatGPT 4o's New Image Generation from an Engineer's Perspective

2025-04-01 · 3 min read AI generative-ai engineering

As an electrical engineer, I don't usually spend time with fantasy storytelling or visual design — but I've always been curious about where AI is heading, both in creative and engineering domains.

Recently, I tested ChatGPT-4o's new image generation feature by reimagining scenes from the fantasy series: The Wheel of Time. The results were fantastic, which led me to further explore if this tool could be functional in the engineering field as well.

What I Explored¶

Fantastic: - Can the new tool not just draw, but visually narrate a moment with emotion, structure, and storytelling? - Can it finally generate text correctly within images (a known limitation in previous models)?

Functional? - Can this tool be useful in engineering, not just in art?

Fantastic — Scenes I Reimagined¶

"You are the Dragon Reborn" — Rand hears his destiny, rendered in dramatic manga style with an emotional panel layout.

Mat's Corruption — A haunting, shadowy depiction of Mat falling under the dagger's influence.

The White Tower — A traditional, calm, symbolic fantasy cover style.

Rand Awakening at the Eye of the World — This one was my favourite. I tried three versions:

A single-scene fantasy poster:

A 3-panel comic showing Rand's raw power:

The same comic, now with magical sound effects and monologue added:

What I Learned¶

This tool is a major leap beyond previous DALL-E models. It maintains scene and character consistency, and you can adjust layout, proportions, and tone — almost like working with a designer. Plus, the generated text is no longer a random assortment of characters; it's meaningful and coherent.

If you've used image generators before, you'll know how big that is.

Functional? — Applications in Engineering¶

I wanted to see whether this tool could also be helpful in engineering communication — where we often need to simplify and visualise complex ideas.

Here are some promising use cases: - Technical illustrations - User-focused dashboards - Interactive learning content - Prototyping concepts

I asked it to generate images representing these points — and it did surprisingly well.

But There Are Limitations¶

When I tested it with basic theory illustrations — like Ohm's Law (U = IR) — it worked fine, even with realistic object symbols.

But when I asked it to generate something more complex, like the pi model of a high-voltage overhead line, the result was visually impressive but technically inaccurate.