クリエイティブ・創作
How a Nursery Teacher Made a Picture Book with AI in 3 Months — Conversations and Practice That Set an Abandoned Dream in Motion
機械翻訳 / Machine-translated
❤️ 0
🔖 0
⚡ 0
機械翻訳 / Machine-translated
One night in January 2026, Sawai (34) stared at her smartphone screen, unable to move for about fifteen minutes. A "picture book cover made with generative AI" that a colleague had posted on social media had stirred a dream she had given up on ten years ago.
She was in her eighth year as a nursery teacher. Reading picture books aloud was part of her daily work. She understood, better than anyone, the professional picture book author's craft — how to choose words that reach children, when to turn the page, how to structure a story so it stays in the heart. And yet the wall of "I can't draw" had always been right there in front of her.
Going to an art university had been an option. In the winter of her second year of high school, she consulted her homeroom teacher, who told her, "It's tough to make a living on artistic talent alone," and she chose a faculty of education instead. Even after becoming a nursery teacher, the pattern repeated: she would fill notebooks with picture book ideas on her days off, then close them with a "but I can't make it real."
In mid-January 2026, things changed. Her colleague's post led her to discover that, using image-generation AI and conversational AI, she could give shape to a story even without the skill to draw. The very next day, after work, she signed up for a generative AI service and typed in her first prompt.
"A young squirrel lost in the forest, gentle expression, picture book style"
Twenty seconds later, what appeared on screen was indeed a squirrel — but it was not "that little one" she had been picturing in her mind. The shape of the ears, the size of the eyes, the tones of the background — everything was subtly off. She generated a second image, then a third, but the same squirrel never appeared twice. Without consistency in the character, the book couldn't work as a picture book.
She kept at it until one in the morning that night and, in the end, couldn't salvage a single usable image.
The following week, Sawai changed her approach. Rather than chasing a perfect illustration from the start, she decided to first solidify the skeleton of the story. She consulted a conversational AI — "I'd like to work together to build a plot for a picture book aimed at four-year-olds" — and spent a week carefully developing the theme, characters, and page structure.
The story's title was Chiisana Risu no Okurimono (The Little Squirrel's Gift). The protagonist was "Non," a squirrel living in the forest. The story was about Non sharing precious acorns with friends before winter arrived. The structure was set: 32 pages in total, 16 double-page spreads.
The next challenge was locking down the character design. Sawai asked the conversational AI to "describe Non the squirrel's physical features as a bulleted list," then used that description to build her image-generation prompts.
By including this list in every prompt, Non's appearance began to stabilize. By the end of January, she had more than twenty illustrations of Non in hand.
February was devoted to generating backgrounds and scenes. The forest at morning, midday, and dusk. The moment snow begins to fall. The scene where Non hands acorns to a friend. She generated an average of 15 to 20 images per scene and chose the single one that best served the story. Every weeknight after work she spent thirty minutes on this; on weekends, two to three hours.
The conversational AI was also useful for checking the story's internal consistency. When she asked, "Will a four-year-old understand this line?" or "Does this plot development feel too abrupt?" she received concrete suggestions for improvement. Sawai treated the AI like a "co-editor," keeping all final decisions for herself.
On March 20, every page of the picture book was complete. She laid her own text over the 32 illustrations created with image-generation AI, adjusted the layout using a free editing tool, and a single data file — one complete book in PDF format — was finished.
The following week, with the director's permission, Sawai held a test read-aloud session at the nursery where she worked. She read Chiisana Risu no Okurimono to 18 children in the four-year-old class.
"Non is so kind." "Is she giving away her acorns?"
The children's response was wonderful. When she closed the final page, three voices called out, "Again!" In that moment, Sawai decided to self-publish.
In early April, she used an on-demand printing service and printed 20 copies. The cost was approximately ¥800 per copy — ¥16,000 in total. The day a bound A4-size picture book arrived at her home, Sawai felt, for the first time, that it had truly taken shape.
The production period was roughly three months. The main tools she used were three: an image-generation AI, a conversational AI, and free editing software. The technical barriers had dropped dramatically compared to ten years ago. Even so, Sawai is emphatic: "AI doesn't do everything for you."
"The time spent fine-tuning prompts, the time spent developing the story, the patience to generate and regenerate again and again. AI is a tool, and without the intention of the person using it, it won't go anywhere. If you're not clear on what you want to make and who you want to reach, things won't work out in the end."
In fact, during the first month, she came close to giving up multiple times as expectations collided with reality. But the feeling of "I want to set that abandoned dream in motion again" kept her going back to the work every night for thirty minutes.
Sawai is currently developing ideas for her second book. The theme is "parting from a friend." She wants to create a story that can gently encourage the children who leave to attend a different nursery.
Through this experience, Sawai came to realize that "what you want to say" matters more than "the skill to make it." AI can generate images, but the heart of the story is decided by the human. What do you want to deliver to a four-year-old? What feeling do you want to leave behind? She feels that continuing to face those questions is the true essence of creative work.
For those who want to deepen their skills in using AI, structured learning environments such as HAIIA's certification program are also an option worth considering. A perspective grounded not only in technique but in AI as a vehicle for self-realization aligns closely with HAIIA's brand philosophy of "setting abandoned dreams in motion."
And if you, too, feel that "there's something I want to do, but I have reasons I can't" — even if you can't draw, even if you can't write, even if you lack the skill, if there is something you want to make, try it just once. That one attempt might set something in motion that has been frozen for ten years.
Sawai's next book is scheduled to be completed in the summer of 2026. Communities where people with the same aspirations can learn together — like joining HAIIA — are also expanding as a future option.
In Sawai's case, paid plans for an image-generation AI and a conversational AI came to approximately ¥3,000 per month, and printing cost ¥800 × 20 copies = ¥16,000, for a total of roughly ¥25,000. By making use of free plans, it is also possible to start with only the printing costs.
Yes. Using image-generation AI, you can create illustrations even without the skill to draw. What matters is the ability to put into specific words "what kind of image you want," and the time to keep trying and refining.
Sawai took about three months, working at a pace of 30 minutes a day on weeknights and 2–3 hours on weekends. If you work intensively, it's possible in one to two months; if you proceed at a relaxed pace, it may take six months or more. The important thing is to decide that you will finish it and keep going.
This article is reprinted from HAIIA Notes.