About PopStory
PopStory publishes short illustrated stories, free to read, in 12 languages and at every reading level from absolute beginner to fluent. There are 70 of them so far.
Where this came from
PopStory started in 2022, at home, as a way of getting my own children reading in two languages again after the lockdowns. The English half was easy: there is an enormous amount of writing which I understand. The Welsh half was not so easy for me to understand or trust! Believe me, there are many poorly translated childrens books...
That was before generated writing became a category on the internet. The aim since has not changed, but we have learned a lot.
The main thing several years has taught us is that the novelty wears off fast, for everybody. Machine story telling is no longer interesting. What lasts is much more ordinary: whether there is something worth reading, at a level that fits, that somebody picked for themselves.
This is not a substitute for sitting down and reading with a child, and or doing the hard work of learning the grammar and vocabulary of a new language. You still have to do those things, but maybe here you can find the "right piece".
Welsh also showed up the limits of this. Large models are fluent in Welsh in the way a confident learner is fluent: convincing, and wrong in ways a first language speaker catches at once. We have had those errors pointed out and the people pointing them out were right. So we now collect corrections and publish them openly, for anyone training or testing a model to use. Welsh, Irish and any others are short of exactly that kind of material. If you are able, please help us grow this dataset.
Reading level and audience are separate things
Almost everything written in simple language is written for small children. So an adult learning Japanese gets picture books about ducks, and a twelve year old who reads slowly gets the same.
Here they are unrelated. You choose how hard the language is, and separately who the story is for. An adult reading Japanese at beginner level gets a story written for adults, told in words they can actually read. A confident nine year old can read something with real stakes in it without being handed adult subject matter.
Levels are CEFR, and where a language has its own widely used scale we use that instead, because those encode script difficulty and CEFR does not. A Japanese story at A2 is written to JLPT N4: only N5 and N4 kanji, and everything else in kana. A Korean story at B1 is written to TOPIK II level 3. Length follows level rather than audience, because an A1 vocabulary cannot carry eight hundred words.
Yes, it is generated
Most generated writing on the web is worthless, and the reason is not that a machine wrote it. It is that nobody was meant to read it. It exists to be indexed and counted by algorithms, in volume, with no constraints.
The constraint here is different, and it can be checked. A story at JLPT N4 either stays inside N5 and N4 kanji or it does not. An A1 story either keeps to one clause and eight words a sentence or it fails. A story that misses its level is a failed story however nice the prose is, because the reader it was written for cannot read it.
The stories are also written in their own language rather than translated. Asking for a story in Japanese, drawing on Japanese storytelling, produces something noticeably different from writing in English and translating. That difference is most of the point.
And we publish the workings! Every story has a production notes page showing its cast, its setting, the five-part plan behind it, the reference drawings, and what it took to make.
How a story is made
Nothing is written in one go. The cast is designed first, then the setting, then a five-part plan, and only then the prose, one part at a time. Each step is kept and can be looked at.
The pictures work the way an animation studio works. Before any scene is drawn, each character is drawn once as a model sheet, and the setting is painted once as a background key. Every scene is then drawn against those references, which is why the same knight is recognisably the same knight in all five pictures instead of changing between them.
So far that comes to 733 drawings and 54808 words. All documented in the production notes.
What we do not do
There are no accounts, nothing to sign up for and no advertising. Nothing is generated from anything you type, because nothing here runs live: the stories are made in batches and the site is a set of files. Every story is labelled with who it was written for, so nobody stumbles into something written for adults while looking for a bedtime story.
This is not a course and it will not replace a teacher. It is reading practice at a level you can manage, with pictures, which is the thing learners run out of first.