Introduction

Lately, I’ve been spending so much time using FreeCAD to design hobby robots that my shoulders and neck are completely stiff...

Thinking “I need a bit of an escape from reality,” I stopped by a bookstore — and something caught my eye:

> “Stable Diffusion Start Guide: Introduction to Image Generation AI”

The cover featured a cheerful-looking anime girl character.

Honestly, I hadn’t been all that interested in image generation AIs until then — but before I knew it, I was heading straight for the register.

It’s kind of scary how momentum works, huh?

My Impressions After Reading It

Surprisingly, it was a fun read!

One big plus: with Google Colab Pro, it seemed like I could run it in my own environment without much trouble.

It especially seems like a great tool for people who love illustration.

I remembered ChatGPT once said, “If it’s for anime-style art, Stable Diffusion is your best bet.”

Writing prompts in English might be a bit tough (just my personal opinion),

but overall, it gives the impression of being a very flexible tool —

you can specify the art style, give detailed instructions, and more.

How Is It Different From Other AI Tools?

I was also curious: how is this different from the ChatGPT I usually use?

Roughly speaking, there seem to be two main types of tools:

1. Text-based AIs that control image generation AIs

　→ This includes tools like ChatGPT and Gemini. You give them instructions in text, and they pass those along to an image generation AI.

2. AIs with built-in image generation capabilities

　→ Some apps and Grok fall into this category. You input prompts directly, and the image is generated internally.

Using the example prompts from the book as a reference, I tried preparing a prompt specifically for an image generation AI:

Sample Prompt I Created

> A vivid, fantasy-style anime illustration.

A single girl with long silver hair, wearing a long black robe and black boots, sitting cross-legged in a flower field with a gentle smile, looking this way.

She’s wearing a large witch’s hat.

It’s a sunny day, and a castle is visible in the distance.

I then tried using this prompt with Gemini and Grok to generate images.

Generation Results

Gemini’s Result

Grok’s Result

ChatGPT’s Result

Only ChatGPT required some retraining with a revised prompt:

Additional Prompt:

> Please research current fantasy and anime art styles online, then use the following prompt to generate an illustration.

Also, image generation through ChatGPT can be slow due to heavy traffic.

And as ChatGPT itself admitted: “I’m not the best at moe-style illustrations.”

Conclusion

Image generation AIs turned out to be much more fun than I expected.

Personally, I’m more of a “blueprint drawer,” so if I get too into this, I might never go back to FreeCAD...

So for now, I’ll rein it in and return to my design work.

That said, if you enjoy illustration, or if you’ve ever wanted to turn your ideas into visuals,

Stable Diffusion is a really compelling tool.

If you’re curious, it might be worth trying out — even just as a way to refresh your mind.

Trying AI Art for the First Time