ebook cover generating api #idea

[ilyakava]

Something like this but better and online. Maybe a request comes in with some text and a cover in different styles is generated (could be abstract generative art, or noun based photo collages).

The reason for needing this is:

1. many textbooks have no meaningful cover anyway, just a solid color or gradient
2. it is a pain to search for actual covers for many books at once while these could be generated passively
3. Some famous books have many covers anyway, and many of them can be boring, prompting the user to search for the cover anyway

Would be very cool to use from within calibre but an online app or just api endpoint would be nice to use as well

[D-32]

Sounds interesting. What kind of style were you thinking of? Simple solid colour / gradients? Or random stock photos?

[ilyakava]

@D-32 as I talked with @ranveeraggarwal someplace else, there are many directions to go in, up to your preference, but the abstract images, or single photo, should be the easiest.

photocollage

The photocollage would be great, but it would be tough to get right. The wordsworth books have the photocollage feel, but automating it to look pleasant would be herculean.

I guess this would be the dual problem of the popular captioning challenges in the image processing world. Maybe this is being done already, the learn a model to go from text to photo collage.

A modification of this idea is not to use the inherent data in the text submitted to search for images, but to use metadata. Many websites relate books to each other, and may distill a genre and keywords for groups of books. If you could identify which genre and keywords are relevant, you could use that subset of covers, or just that information, as input to the image search/generation.

art appropriation

Something also worth trying is the nypl like art covers with an art api like artsy, which can be searched for keywords and can return images.

abstract

On the abstract side is that this is in a way an identicon problem, trying to come up with a visual representation of information that is both informative and unique to each entity (I would prefer something more informative than the nypl abstract covers). This is what github does for new users in a simple way, and what some generative art people do in more complicated ways.

The initial inspiration for this idea a couple years ago for me was the old Dover reprints of classics, and the covers that they came up with. I was making a bunch of PDFs about different subject areas, and I was wishing that I could just type "probability" and get a bunch of Dover type covers.

usecases

@ranveeraggarwal mentioned that this could be useful for the plethora of new authors out there self publishing books, ebooks, even small online serials. This could be the literary version of logojoy which got some attention on HN lately. So if someone wanted to own this and try to monetize it, there is the potential to do so.

[ggerhard]

@ilyakava @D-32 This thread is excellent, the kind of networking and creativity I had in mind.

I don't know much about this domain, but from the 1PPM point of view, there are two things you could do:

You have some ideas about the market/need and about the technical complexity of your product. So you could try to test or get more information on either one:

A - the "lean startup" approach: try to learn more about your customers and the business with a technically simple, but rolled out solution (xyebookcover.com). Maybe a kind of mechanical turk. Customer submits an abstract and you send to them a coverpage the next day via email. Or auto-generate only very simple covers. Eventually you will have a list of potential beta-testers for your AI solution + knowledge how the AI results should look like.

B - the technical approach: learn more about current machine learning frameworks with a simple, related, 1 month ML project. By the end of the month you will have a better understanding how complex a potential technical solution would be and can build on that.

Would both be fun projects :)

Hope this was helpful, Cheers

[ranveeraggarwal]

@ilyakava @D-32 maybe this is longer than a 1-month-project. We could break this down into smaller chunks and get a cool and useful product by the end. The techniques used in this idea could potentially have several uses apart from those discussed above.
For example, this could be used as an easy event-poster generator, or a wallpaper-generator.
We could all have a slack discussion on how to get started and see where it goes 😄

[ilyakava]

@D-32 or @ranveeraggarwal by all means take the lead. I am in a different field now, and have many ideas like this from the old days but no ability to pursue them. Just thought I'd post a few since 1ppm got traction on hn :)