I Built an AI Children's Book Generator (PictureBooks.io) - Looking for feedback!

Added a new feature today: Story World. Story Worlds let your child become the hero of beloved classic tales! Choose from worlds like:
  • Alice in Wonderland
  • Wizard of Oz
  • Peter Pan
  • The Jungle Book
  • Treasure Island
  • Arabian Nights
  • Robin Hood
Your child enters these magical settings while keeping their own identity - meeting the Cheshire Cat in Wonderland or sailing with pirates to find treasure. It's a unique way to experience classic stories with your child as the star!

Below is what the Step 2 of the Create Story wizard looks like now:

1768799581183.png


Also updated the credits section to show that it costs as low as $1 to create a personalized book:

1768797323333.png
 
Last edited:
I don't think using public domain characters as a way to avoid copyright infringement actually works if you're still using what's obviously the Disney versions of said characters, and Alex wearing a t-shirt with a very much not public domain character on it won't help your cause either
 
Most of the times I click on a link to the site or a particular book, it times out. I don't think the problem is on my end as this has happened from different locations and networks (and browsers).
 
Most of the times I click on a link to the site or a particular book, it times out. I don't think the problem is on my end as this has happened from different locations and networks (and browsers).
That's odd. On mobile or desktop?
If on desktop, could you hit F12 and see if any error in the console logs when you load the page?
 
Desktop. Can't see anything in any logs. Just times out.
 
Now a Civ related book!

A boy entering the wonderful world of Sid Meier's Civilization! :)


jeff-civ-book-png.755504
This is why they call it slop, man. If a real artist turned in a book riddled with inconsistencies between art elements and character design on every page they'd get fired.

It's substantially worse than real creation, and how are kids supposed to grasp onto anything substantial when everything about it keeps changing constantly for no reason. They deserve better than being fobbed of with this demeaning rubbish.
 
Last edited:
Can you list out some of the inconsistencies?
If you expect 0 inconsistencies, then of course you will be disappointed.

For the main characters (child, parents, sibling, etc), we include reference images (avatars) for character consistency. This is most important for us. However, for secondary characters and objects, it relies on the text description, which could be interpreted differently sometimes.

The main question is for kids, is minor inconsistency a deal breaker, or can they still enjoy the books?

Personally, my 2 year old son has enjoyed some of the books I generated enormously. Minor inconsistency doesn’t affect his enjoyment.

If you want to achieve 100% accuracy, the tool can do that as well, but it requires more detailed prompts from user and some image regeneration efforts. For example, if you want the Civ game box to look exactly the same on all pages, then I could have provided it a very detailed description for the item. It’s just that it’s not worth the effort to do that, as it’s not something a kid would notice or care about.
 
If you expect 0 inconsistencies, then of course you will be disappointed.

If you want to achieve 100% accuracy, the tool can do that as well, but it requires more detailed prompts from user and some image regeneration efforts. For example, if you want the Civ game box to look exactly the same on all pages, then I could have provided it a very detailed description for the item. It’s just that it’s not worth the effort to do that, as it’s not something a kid would notice or care about.

Inconsistency is one thing, it's another when the game box sometimes isn't even Civilization at all (page 18 and the original page 3 image) and can't consistently spell the name of the game or its creator (like on page 2, it's "Sid Mder's Cviization" on the cover)

Can you list out some of the inconsistencies?

Well on top of the game box never looking the same, the rooms the characters are in sometimes change for no clear reason, the aunt has glasses on one page but not any of the others, nobody has consistent hair styles, the kid suddenly becomes barefoot when he teleports into the game, then has sandals on the next page, then sneakers which are sometimes blue and sometimes brown, clothing details like the blue collar on the kid's white shirt or Cleopatra's dress patterns pop in and out of existence at random, Cleopatra's crown has a different shape on every page, the kid's height relative to the adult characters keeps changing so he sometimes looks 5 and sometimes looks 15, and last and foremost, the leaders' facial models differ so much from page to page they don't even look like the same art style sometimes- like, if pages 9, 10, and 11 weren't back to back in the same book, I'm not sure if I'd even be able to tell all three versions of Cleopatra are supposed to be the same character!
 
Capability is there to generate the game box more accurately. Just didn't find it worth to do it. Right now for the game box, it just has a simple description.

A game box for Sid Meier's Civilization

1769221594869.png


Change the description for that box item to be super detailed, like below:

A photograph of the vintage, slightly worn cardboard game box for the PC game "Sid Meier's Civilization" by MicroProse. At the top center, in blue text, are the words "Sid Meier's". Below that, in large, gold, block letters with a black outline, is the title "CIVILIZATION". Underneath the title, in black text, is the phrase "Build An Empire To Stand The Test Of Time". The central artwork is a diorama-style illustration showing a cross-section of the earth. Above ground is a beige, paper-model city skyline featuring a mix of modern skyscrapers and classical buildings like a Greek temple, under a pale sky. Below ground, in a sandy excavation pit, are classical ruins including a large bust of a Roman head, fallen fluted columns, and a small temple structure. The bottom right corner features the red and orange "MICRO PROSE" logo. The edges and corners of the box show signs of wear, scuffing, and creases.

and this can recreate a game box art that looks a lot more like the game box for Civilization:

1769222056501.png


So if you really want to create something as accurately as possible, the capability is there.
For maximum accuracy, could even provide a reference image via the character avatar feature which passes the reference image to the image generation AI.

CivGameBoxDetailed.png
 
Last edited:
Capability is there to generate the game box more accurately. Just didn't find it worth to do it. Right now for the game box, it just has a simple description.

Bloody hell AI is making people too lazy to even use AI well
 
I am not looking to publish the book and sell to people on Amazon, so they don’t need to be perfect with zero inconsistency.

I mainly use the tool to create personalized books to read together with my son. Like if I go to a holiday train show with my son, we make a book about it. It’s something that we enjoy doing and he has read way more books as a result, in addition to the books we buy from bookstores.

People who are looking to use AI tools to create books for self publishing on kindle, they can put in more effort.
 
I am not looking to publish the book and sell to people on Amazon, so they don’t need to be perfect with zero inconsistency.

But you are using examples like these to try to get people to pay you for the better version of the AI. Why would someone want to if the product looks bad?

Also even in the repaired image, why did Millie's glasses disappear, why did the background change so much (it even deleted a window), and despite the detailed description the text is still blurry and the logo in the bottom right says "Mcro Pbsqe" and the side text is complete gibberish, because the AI does not know the difference between text that would make sense to a human and text-like gibberish!
 
I’m not well-versed on the technical side of things but I don’t understand why it would not be better overall to have a two-step generation process where it creates a “sprite layer,” that is, main focus objects that appear in every/most scenes, generated once and adapted, and a background layer generator that is generated individually for each page.

Presumably this would take away some of the consistency problems since the same object is used/adjusted for each page.
 
I’m not well-versed on the technical side of things but I don’t understand why it would not be better overall to have a two-step generation process where it creates a “sprite layer,” that is, main focus objects that appear in every/most scenes, generated once and adapted, and a background layer generator that is generated individually for each page.

Presumably this would take away some of the consistency problems since the same object is used/adjusted for each page.
that's how I'd do it, but generation costs would be higher as you'd need a lot more images for each book.
 
Page 7 works as example for what i dislike about AI images..
The boy and Cleo have the same face basically..just one AI male and one AI female.
And those Genghis Riders :lol:

Really young kids prolly don't care, but they are being "robbed" of real art impressions.
When i think about Walt Disney comics (Donald Duck was what i did read most)..it's just night and day.
I would be concerned about them memorizing 0815 images, and judging soulless AI as standard & normal.

Why? Cos they will grow up with an ever increasing amount of AI stuff, with the way things develop currently.
Sorry if all of that sounds a bit harsh..but i don't think it's healthy.
 
Back
Top Bottom