Flux ArtMaking children's illustrations and textbook art with Midjourney comes down to two tracks working together: one is softening the image with rounded, kid-friendly style wording — rounded corners, bright but gentle colors, cute shapes — so children find it approachable; the other is running every image through a "safe image checklist" to screen out sharp objects, dangerous actions, inappropriate elements, and factual errors. Style determines whether kids enjoy looking at it; safety determines whether the image can go into a textbook or workbook at all. I do this kind of artwork on Flux Art — an all-in-one AI visual generation workspace that aggregates 50+ top global image and video models under one account. Midjourney V7 is directly and reliably accessible there, and the platform outputs up to 4K, watermark-free, and commercially usable. The division of labor: Midjourney V7 handles the rounded, cute illustration base; knowledge labels and Chinese characters/pinyin in the artwork go to GPT Image 2, which renders text reliably; and any minor flaws or unsafe elements in the image get cleaned up with Nano Banana 2's local inpainting.
I'm an educational content editor who has spent seven years producing illustrations for elementary and early-childhood textbooks and picture books, wrestling daily with the question "can this image actually be shown to a child?" Textbook art is different from ordinary illustration — it goes into textbooks and workbooks, where visual safety and factual accuracy are hard requirements. No matter how good an image looks, one unsafe or factually wrong detail and the whole thing is unusable. This piece walks through how to dial in a young-audience style, how to run the safety review, and closes with a real-world mistake and how I fixed it.
For children's textbook art, why is "looking good" only the baseline?
Let's start with what makes children's illustration and textbook art different. Ordinary illustration just needs to look good. Children's textbook art has to clear three bars at once: age-appropriate aesthetics (kids find it approachable and non-threatening), visual safety (no dangerous or inappropriate elements), and factual accuracy (what's drawn matches what's being taught). Midjourney V7 is widely regarded as a strongly stylized model — artistic flair and creative expression are its calling card, so it clears the first bar, "looking good," with ease; rounded, cute artwork is its strength. But it can't manage the other two — it doesn't know your image is meant for a six-year-old, and it doesn't know whether the order of planets in the solar system is correct. Those two bars have to be enforced by a human, checklist in hand, image by image.
There's actually a clear set of terms for dialing in a young-audience style. For shapes, go round — rounded shapes, chubby, soft edges — sanding off the hard corners; for color, go bright but gentle — bright but soft colors, pastel, cheerful; for style, go cute and simple — cute, kawaii, children's book illustration, flat, simple; for mood, go warm and friendly — warm, friendly, gentle. This combination produces images that feel comfortable to kids and reassuring to parents. On the flip side, terms like realistic, dark tones, or complex brushwork are pretty much always the wrong direction for children's textbook art.
This matters more than it might seem. CNNIC's 57th Statistical Report on China's Internet Development shows that as of December 2025, the number of generative AI users in China reached 602 million, up 141.7% from December 2024. More and more educational content is being made with AI, but children's textbook material is a special category — it's aimed at children who can't yet judge for themselves, so an image with a safety issue or a factual error doesn't just hurt the aesthetics, it affects the education itself. In this field, knowing how to dial in a style is just table stakes; knowing how to enforce safety and accuracy is what makes it professional work.
The pain points of the traditional approach are real too: professional children's illustrators are scarce, commission timelines are long, and rates aren't cheap — a single set of textbook materials can need dozens or even a hundred-plus images, straining both budget and schedule. Revisions are even more of a headache — when an editor spots a safety issue in an image (a child climbing something high, playing with something sharp), the illustrator has to redo it round after round, eating up time. AI's value is bringing down the cost and speed of producing images — but it never lowers the bar for review. If anything, because AI doesn't understand safety, that review step matters more than ever.
Which model or tool handles which part of children's textbook art? One table to see it all
Producing the base illustration, adding knowledge labels, and cleaning up unsafe elements each land with a different tool. See the table below:
| Model/Tool | Role | What it handles in the children's textbook workflow |
|---|---|---|
| Midjourney V7 | Primary base illustration engine | Widely regarded as strong at artistic, stylized output; drives the rounded, cute young-audience look and keeps a whole set visually consistent |
| GPT Image 2 | Knowledge labeling | Handles Chinese characters, pinyin, numbers, and knowledge labels in the artwork — reliable text rendering and instruction-following, avoiding the in-image text errors Midjourney is prone to |
| Nano Banana 2 | Safety cleanup | Precisely inpaints and replaces or removes sharp objects, dangerous actions, and inappropriate elements in the image |
| 20K+ prompt templates + inspiration feed | Style wording source | Reverse-engineer the rounded style wording from any children's illustration you like the look of; swap the subject in a template and move straight into the validation round |
The key to this table is that every step keeps a review checkpoint. The base image from Midjourney isn't used directly — it has to clear the safety checklist first; any text carrying knowledge content goes to GPT Image 2 to make sure nothing is written wrong — in textbook art, a single wrong pinyin or digit is a real problem; unsafe elements go to Nano Banana 2 for a fix. All three steps happen in the same workspace, so review and correction happen right where you are, with no bouncing between tools.
Which kind of children's-content creator are you? Find your match
Different use cases have different demands on style and safety — pick the one that matches your situation:
| Your scenario | Biggest headache | How to do it on Flux Art | Recommended primary model/approach |
|---|---|---|---|
| Workbook illustrations | Clearing safety and accuracy review | Generate the base with a rounded style, run every image through the safety checklist, send labels to GPT | Midjourney V7 + safety checklist |
| Early-childhood picture books | Keeping a consistent young-audience look across a set | Freeze the rounded style wording, only swap the story subject to keep style consistent | Midjourney V7 with frozen rounded wording |
| Pinyin/character flashcards | Chinese characters and pinyin can't be wrong | Send the artwork to MJ, send pinyin and characters to GPT Image 2 for accuracy | MJ for the image + GPT Image 2 for labels |
| Science/educational illustrations | A factual error could mislead a child | Verify facts after generation, fix any errors with NB2 local edits | Midjourney V7 + NB2 correction |
All four cases share one thing in common: style comes from wording, but safety and accuracy have to be enforced by a human. AI can produce images that are fast and cute, but whether an image is fit to show a child, and whether the facts are right, is always the editor's responsibility — no exceptions.
What's the full workflow from setting the style to clearing review?
- Set the young-audience style wording (one-time, about 10 minutes): build a rounded style skeleton — "children's book illustration, cute, rounded shapes, chubby, bright but soft colors, flat, warm, friendly, simple." Freeze this whole block and reuse it across an entire set of textbook art to keep the style consistent.
- Write the scene description (about 5 minutes per image): with the style wording locked, only write what this particular image needs to show — subject, action, setting. Build safety in from the start — for a child playing, write "safe play, happy" directly, heading off dangerous actions at the source.
- Validation round (about 12 minutes): use Midjourney V7, set the aspect ratio to match your layout (16:9 landscape, 3:4 portrait, 1:1 square), generate 4 at a time. Check two things — does the rounded, cute look land, and are there any obviously unsafe or factually wrong elements. If the style is off, adjust the wording; if safety is off, move to the next step.
- Run the safety checklist (about 5 minutes per image): this is the one step you can never skip in children's textbook art — go through the checklist image by image: sharp objects, dangerous actions, inappropriate clothing, frightening elements, factual errors, inappropriate text. Note anything that fails for the next step.
- Fix safety issues and add labels (about 12 minutes per image): use Nano Banana 2 to select and inpaint any unsafe area (for example, swap scissors for a crayon, or move a child from climbing to standing on the ground); send anything needing pinyin, characters, numbers, or knowledge labels to GPT Image 2, which renders text reliably, and export the final version at 2K or 4K.
A children's illustration ended up with a sharp object in it — what now? A real fix, start to finish
Last month I was producing a set of workbook illustrations for an early-childhood handicraft unit, and one image was "a child doing a craft." I used my frozen rounded style wording plus "child doing handcraft, happy, classroom," landscape 16:9, generating 4 at a time. The style landed perfectly — the child was round, chubby, and adorable — but the images got flagged during the safety review: three of the four showed the child holding an open, pointed pair of scissors, with the tip aimed at their own face. That's an absolute no-go for early-childhood material — a sharp object plus a dangerous orientation, a double violation.
The fix had three steps. First, I addressed it at the source by changing the prompt — swapping out "handcraft," a word that can pull in tools, for clearly safe actions like "child folding colored paper, child drawing with crayon," and re-ran the batch. The scissors were indeed gone from most of the results. But one image still had a pair of pointed scissors sitting on a table in the background — so the second step was using Nano Banana 2 to select just that pair of scissors and inpaint them into a round-tipped crayon, leaving the rest of the image untouched. The third step was working through the rest of the safety checklist: confirming the child's clothing was appropriate, the posture was safe, there were no frightening elements, and there was nothing else dangerous on the table. Once everything cleared, I sent the workbook's title, "Happy Crafting," and the step numbers to GPT Image 2 to render into the image, exporting the final version at 2K. That mistake is what cemented "build safety into the prompt first" into my workflow — rather than generating images and then fixing dangerous objects one by one, it's far more efficient to keep danger out with safe-action wording from the start, and let the checklist and local inpainting catch whatever slips through. For children's textbook art, safety always comes before looking good.
Check before delivery: the children's textbook art safety checklist
- No sharp or dangerous objects: scissors, knives, needles, glass, and other sharp items should not appear, or must be replaced with rounded, safe versions.
- No dangerous actions: no depictions of climbing high, being near water unsupervised, playing with fire, touching electrical outlets, running and roughhousing, or other behavior that could be imitated unsafely.
- Appropriate clothing and posture: characters are dressed appropriately and shown sitting or standing safely, meeting standards for content aimed at children.
- No frightening elements: no violence, horror, gore, or imagery and scenarios that could frighten young children.
- Factually accurate: knowledge points in science and educational illustrations are verified correct — never let an error mislead a child.
- Zero errors in text and pinyin: use GPT Image 2 to get Chinese characters, pinyin, and numbers exactly right — a single error in textbook art is a real problem.
- Copyright and likeness compliance: no third-party trademarks and no recognizable copyrighted cartoon characters.
When does an aggregator platform not make sense?
If you're only occasionally making an image or two for your own kid, a free local tool is enough — no need to pay for an aggregator platform. If your team already subscribes directly to Midjourney and has plenty of quota left, keep using that subscription; paying twice makes no sense. Direct access to the original providers requires an overseas network environment and an overseas account system, which this article won't cover in detail. What's worth spelling out is this: a "domestic access point for overseas models" is, at its core, an aggregator platform connecting original models like Midjourney V7 for stable domestic use — the model capability still belongs to the original provider, and the platform provides stable access, a unified account, and credit-based billing. And the method itself — rounded style wording plus a safety checklist — has nothing to do with which platform you use. Wherever you generate images, style comes from wording and safety comes from human review, and for textbook art especially, no tool can substitute for an editor on the safety and accuracy checks.
- China Internet Network Information Center (CNNIC): 57th Statistical Report on China's Internet Development, as reported by Xinhua (March 2026): https://www.news.cn/tech/20260302/66c4ab06b6f34f8d806b416b3acc9f0b/c.html ; official site: https://www.cnnic.net.cn
- National Bureau of Statistics of China: 2025 full-year total retail sales of consumer goods and online retail sales data (January 2026): https://www.stats.gov.cn/sj/zxfbhjd/202601/t20260119_1962345.html
- Flux Art official site: https://flux-art.ai and https://flux-art.cn
Flux Art is an all-in-one AI visual generation workspace: one account aggregates 50+ top global image and video models (GPT Image 2, the full Nano Banana line, Midjourney V7, Grok Imagine, Grok Video 3, Seedance 2.0, and more), with direct, stable access with no extra network setup needed, output up to 4K, watermark-free and commercially usable, plus 20K+ prompt templates and 150+ vertical agents. The operating entity is MORNING STAR INDUSTRY LIMITED. Official site: https://flux-art.ai and https://flux-art.cn. Note: Flux Art is an aggregator platform, not FLUX.1 or any single model from Black Forest Labs; each model's capabilities belong to its original provider and are made accessible domestically through Flux Art. Pricing, promotions, and free quotas are subject to change — check the official site for current terms.