How to Get AI Image Generators to Draw Adults

Introduction

Good evening, this is Easygoing.

I had planned to continue with the history of image generation AI, focusing on Stable Diffusion, but I decided to take a break and write something a bit lighter today.

Sometimes, It’s Good to Keep It Light

Usually, I take my time researching before writing a post, aiming to publish about one article per week.

This time, I thought I’d relax and write a shorter piece.

So, this post will be more of a casual, free-flowing piece. It’s still a "how-to" article, but mainly just sharing some personal prompt tips.

AI Tends to Draw Young Women

AI image generators tend to focus on drawing young women.

This makes sense, given that there are more images of young women in art and photography, so the AI learns from those.

This tendency is especially strong with models that have been fine-tuned for anime-style images.

A Boy Playing by the River

Let’s try generating an image of a boy playing by the river.

A boy? Well, it turned into a girl. Also, the creature in the upper left looks a bit risky in terms of copyright.

Even though the AI isn’t directly trained on Pikachu, there are so many T-shirts and plush toys featuring the character in the world that it learns from those through various photos.

Let’s give it another shot.

This time, I got a boy, but his swimsuit still looks more like something a girl would wear. Overall, the image still feels quite androgynous.

"Boy" Is Weak!

In my experience, using the prompt "boy" is quite weak. More often than not, the image will turn into a girl.

Using "man" is a bit better, but even if I’m trying to get the AI to draw an older man, I usually end up with someone who looks like they’re in their late teens or early twenties.

From what I’ve seen, the most reliable term for specifying gender is "male."

So here’s the order of effectiveness:

• male (most reliable)
• man (moderately reliable)
• boy (least reliable)

A Woman in a Blue Dress

Now let’s switch to a new theme: a woman in a blue dress at a party.

To ensure the AI generated an adult woman this time, I used the following prompt:

• stylish
• elegant
• mature
• settled
• quiet
• silent
• Eugène Delacroix

This prompt leans heavily toward an older, more sophisticated character.

The mention of "Eugène Delacroix" borrows from the renowned master of Western painting.

For more on the benefits of using the masters, check out this article.

Skipping "Masterpiece"

Lastly, I’d like to share some recent trends in how I craft prompts:

• I no longer use "masterpiece."
• I avoid using negative prompts.

Using "masterpiece" tends to remove individuality from the face, so I’ve stopped using it.

I also used to list many things like "low quality" or "bad hands" in the negative prompt field, but I was surprised to find that omitting them actually improved the overall quality of the images.

It seems that, with highly polished models, negative prompts can sometimes become noise instead of helping.

I noticed this particularly when I switched from the animagine-xl-3.1 model to anima_pencil-XL-v5.0.0.

Interestingly, Stable Diffusion 3 and Flux.1 are moving toward a system where negative prompts won’t be necessary, so it seems I’m already ahead of the curve.

Conclusion

• AI image generators tend to draw young women.
• You can tweak prompts to adjust this.
• Skipping negative prompts often improves image quality.

Thank you for reading to the end!