AI Illustration Study Notes: Eyes Edition | Organizing Eye Styles with 4 Maps [ChatGPT]

When generating illustrations with AI multiple times, sometimes the result doesn't match the "style I had in mind" and feels a bit off.

Recently, I've been observing facial parts one by one. As the first installment of this, I've compiled the “Eyes Edition” here.

Since the eyes aren't the only element that determines the art style, I plan to organize the nose and mouth next. This is just my own personal study note, but I would be happy if you could take a look at it with me.

1. Why I observed the "eyes" first

It is said that when looking at a person's face, the eyes are the first thing the gaze lands on. Even in AI illustrations, I feel that just a slight change in the eyes significantly shifts the atmosphere, such as “anime-style,” “realistic,” or “Korean-style.”

Of course, the art style isn't determined by the eyes alone. However, since it's a part where the effect is easy to see when you start working on it, I chose it as the theme for the first study session.

2. Four axes for categorizing eyes

After looking at various examples, it seemed that differences in eyes could be explained by roughly four axes.

• Realism … From deformed to photorealistic

• Cultural Sphere … Japanese anime, Korean, Chinese social games, Western cartoons, etc.

• Era … Trends change across the 80s/90s/2000s/2010s/2020s

• Commercial Genre … Shojo manga, Shonen manga, light novels, Webtoon, Vtuber, social games

I felt that having these axes makes it a bit easier to put into words when translating a "vaguely liked art style" into a prompt.

3. Visual map of eyes (gradient of realism)

The first thing I want to look at is the “Realism” axis. While it's not something that can be strictly defined, arranging them in a rough gradient from 0 to 10 made it easier to grasp the atmosphere.

📷 Visual map of eyes (0-10 gradient)

Even if you vaguely say "anime-style eyes," you can see that there is quite a wide range within that. Just by being conscious of "which zone I am aiming for" when writing a prompt, the inconsistency in the output seems to have settled down a bit.

4. Notes on the Vibe of Each Zone

I've jotted down my own impressions of each zone in my own words. The boundaries are fuzzy, so please view these as rough guidelines.

• Around 0-1: Dot eyes, symbolic eyes, similar to LINE stickers. Gives the impression of a minimal configuration with almost no iris or sclera.

• Around 2-3: Retro anime and gag manga. Single-color iris, one highlight, leaning towards flat.

• Around 4-5: So-called "general anime." Common in late-night anime and light novels, this zone features 2-3 layers of iris and starts to show some depth.

• Around 6-7: Modern light novels, Korean illustrations, Webtoons, and streaming thumbnails. Features transparency, gradients, and plenty of light—a style often seen on social media lately.

• Around 8-9: Chinese social games and thick-painted semi-realism. Very high information density, with glass-like eyes, multiple reflections, and a strong sense of spherical volume.

• Around 10: Live-action, concept art, and film style. A hyper-realistic realm where even blood vessels and tear ducts are drawn in.

I feel like I can use this as a personal scale, thinking things like, 'Around 6-7 feels modern' or '8-9 has a strong commercial social game vibe.'

5. Eye Shape and Contour Map

On a different axis from realism, there is also the topic of 'eye shape.' This is the world of commonly heard expressions like droopy eyes, cat eyes, round eyes, and almond eyes.

📷 Eye Shape and Contour Map (Broken down into 15 items)

In this map, I've broken down the eyes into 15 items: shape, openness, upper/lower eyelids, outer corners, inner corners, pupil size, sclera ratio, eyelashes, depth, pupils, highlights, line art, era, and cultural sphere.

Once I broke it down this far, I felt like the resolution had finally reached a level where it could be translated into prompts.

6. Impressions by Shape

Here are my personal notes on the impressions I get from the shapes alone. Please keep in mind that this is just 'how I feel about it,' so take it as a reference.

• Droopy eyes: Healing, protective. Gives an impression that fits gentle or young characters.

• Round eyes: Energetic, approachable. A versatile and easy-to-use vibe.

• Standard Anime Eyes: The most familiar shape, close to the "ordinary protagonist" archetype.

• Slender Eyes: Easily conveys intelligence, coolness, and maturity.

• Slightly Upturned Eyes: Strength of character. Perhaps closer to the image of a heroine or an older sister character.

• Sharply Upturned Eyes: Combat-oriented, queen-like, or aggressive atmosphere.

• Half-lidded Eyes: A languid impression. Seems to fit slice-of-life or sleepy characters.

When you're aiming for a gentle impression but the AI keeps outputting stronger eyes, I felt that adding vocabulary from here to your prompt makes it easier to steer the result.

7. What does "Style" actually mean?

Terms like "Korean style," "Webtoon style," and "Light Novel style" are often seen on social media. They are convenient, but I was using them based on a vague feeling, and my own understanding was a bit fuzzy.

When lined up, there seem to be these differences.

📷 Eyes by Genre (9 types including Shojo Manga style / Light Novel Cover style / Social Game style / Vtuber style / Webtoon style, etc.)

• Korean style: Translucency, heavy use of gradients, redness on the lower eyelid. Semi-realistic, leaning towards realism.

• Chinese Social Game style: Luxurious, flashy, glass-like eyes, multiple reflections. An impression of having a high density of information.

• Light Novel style: Moe-leaning, delicate, larger pupils. Emphasis on looking good on a cover.

• Webtoon style: Designed for readability in vertical scrolling. Lines are relatively simple.

• Vtuber style: Designed for streaming. Highlights are large and easy to see.

I feel that if you put into words what "style" means to you before writing it in a prompt, the output will be less inconsistent.
*I'm sure there are many other "styles" as well.

8. Examples of tags when actually combining them

English tags were also useful when testing outputs. These are some combinations I've put together myself.

📷 List of Eye Direction Tags

Looking at the text alone, one might think 'both are anime-style eyes,' but when I generated them, there was a clear difference in realism. I remember it clicking for me, thinking, 'Oh, maybe this is what the boundary between around number 6 and around numbers 7-8 feels like.'

I felt that for things like this, it's faster to generate and compare them once than to think about it in your head.

9. Examples of actual creations

Based on the maps so far, I tried generating some with random combinations myself. It was interesting how even with the same 'eyes,' the impression changed significantly depending on the combination of tags.

Pattern Example 01

【現代ラノベ表紙風】

丸目
少しタレ目
虹彩3層
透明感強
下まぶた赤
細線

印象:
可愛い
現代アニメ感
万人向け

近い系統:
電撃文庫
富士見ファンタジア系

Pattern Example 02

【韓国イラスト風】

半目
切れ長
細長い虹彩
ハイライト小さめ
淡色グラデ

印象:
儚い
空気感
オシャレ

特徴:
下まぶた赤
まぶた厚め
彩度低め

Pattern Example 03

【中国ソシャゲ風】

軽いツリ目
虹彩超多層
多重反射
ガラス眼
光沢強

印象:
豪華
強い
高級感

近い系統:
原神系
崩壊系
鳴潮系

Pattern Example 04

【少女漫画風】

超タレ目
縦長
まつ毛大量
キラキラ反射
白目少なめ

印象:
感情表現強
恋愛感
可憐

Pattern Example 05

【90年代アニメ風】

巨大黒目
単色虹彩
白ハイライト巨大
太線
平面塗り

印象:
懐かしい
元気
レトロ

近い系統:
セーラームーン初期
昔のギャルゲ

Pattern Example 06

【2000年代萌えアニメ風】

超丸目
黒目巨大
白目少
ハイライト大量
まつ毛少

印象:
萌え特化
幼い
守りたくなる

近い系統:
深夜萌えアニメ初期
美少女ゲーム

Pattern Example 07

【VTuber風】

丸目
虹彩多層
発光ハイライト
透明感
感情表現強

印象:
映える
配信向き
感情が伝わる

特徴:
サムネ映え重視

Pattern Example 08

【Webtoon風】

標準目
少し切れ長
均一線
グラデ少なめ
読みやすい瞳

印象:
スマホ向け
現代感
量産向き

特徴:
縦読み最適化

Pattern Example 09

【アメコミ風】

小さめ瞳
白目広
強い陰影
太線

印象:
力強い
大人
劇画感

特徴:
リアル筋肉系と相性良

Pattern Example 10

【絵本・北欧雑貨風】

点目
超省略
輪郭ゆるい
虹彩なし

印象:
やさしい
ナチュラル
雑貨感

特徴:
ミニマル

When lined up, I felt once again how interesting it is that the worldview changes this much just by the eyes. This kind of comparison seems quietly effective for checking if you are 'getting close to the target zone.'

10. Thoughts after doing the Eyes Edition

Looking at it this far, I feel that AI illustration might be closer to the work of translating the image of the art style in your head into words rather than 'the work of typing in spells'.

By keeping axes like realism, cultural sphere, era, genre, shape, and impression in mind, I feel like my vocabulary grows every time I rewrite a prompt.

Next time, I plan to organize other parts like the nose and mouth in the same way. I would be happy if I could gradually create something like a 'notebook for breaking down and verbalizing art styles'.

Thank you for reading. See you in the next article.

👇 The sequel is out.

★
★
★

I have compiled tips on AI image generation, prompt creation, game development, and efficiency.If you think 'I want to see more' please check out my magazines.

📚 Magazines↓↓
▶ [AI x Information]
Organizing basic AI knowledge, terminology explanations, and how to use generative AI for beginners.

▶ [AI x Illustration]
Compiling practical notes on AI illustration production, including prompt creation, image generation, composition, color schemes, and character design.

▶ [AI x Parenting/Housework]
A collection of ideas for making life easier by incorporating AI into parenting and housework. Introducing AI utilization methods for the home, such as time-saving techniques and childcare support.

▶ [AI x Chrome Extensions]
A magazine introducing Chrome extensions. Focusing on extensions that improve work efficiency.

▶ [AI x Tools]
Introducing AI tools that can be used for work and daily life, such as ChatGPT, Claude, Gemini, and Chrome extensions.

▶ [AI x Design]
Compiling ideas for creative production using AI, such as banners, thumbnails, and presentation design.

Thank you very much.

📱 SNS
X・Threads・Instagram I also post about AI illustrations and production notes. I would be happy if you could follow me. You can tap each link to go to the SNS.

💬 Related Tags
#AI #AIUtilization #DailyImprovementWithAI #DailyImprovement #OrganizingLife #LifeImprovement #Efficiency #WorkEfficiency #BusinessImprovement #WorkStyle #WorkTips #LifeHack #DigitalUtilization #ToolUtilization #Prompt #PromptDesign #LivingWithAI #LifeWithAI #AIForBeginners #AILearning #LearningRecord #TriedIt #TriedItWithAI #ExperienceReport #Insights #Mindset #HabitFormation #note #LikeIt #MidnightLatte

Midnight Latte | AI x Create
Recently, I have been tidying up my note page little by little and trying out things I can do with AI. Your likes and follows encourage me to keep updating!!