An AI drawn maze usually has no solution. Printed, that is a dead end on paper and a one star review. Activity books in this plugin solve that by not using AI for those pages at all.
The four algorithmic page types:
- Maze: generated as a perfect maze, meaning exactly one path exists between any two cells. It cannot dead end and it cannot be unsolvable.
- Dot to dot: numbered dots trace a real outline in the correct order, with the numbers placed outside the shape where a child can read them. The built in shapes are star, fish, house, rocket, butterfly, sailboat, tree and a cat face.
- Counting: distinct, non overlapping groups of objects with an answer box row underneath.
- Tracing: a solid model shape followed by dashed repeats to trace over.
Difficulty ramps automatically:
Difficulty is derived from the page position. The first third of the book is easy, the middle third is medium and the final third is hard, which is how a real activity book is edited.
Puzzles are deterministic. Each page is seeded from the project ID and the page number, so the same book always regenerates the same puzzles rather than shuffling on you.
Which activity pages still use AI?
Picture based activities such as spot the difference, matching and plain coloring pages are generated by the image engine. That means a locked character can appear in them too, while the algorithmic pages cost you nothing in generations.
Because the algorithmic pages are vectors rather than pixels, they print razor sharp at any size, they add almost nothing to the PDF file size, and they never need an upscale pass. The kid facing instruction above each puzzle is drawn as real text in the PDF layer.