For the many web stories we gather from the Internet, we need to give them covers to interest people in them. However, there are too many stories and too little time, so we need to leverage automation to cover for them.
The Penguin Classics template is the gold standard for how covers should be mass produced. This publisher faced a similar problem when publishing tons of printed versions of classic public domain books whose original covers were simply the book bindings.
Choose the main cover image from a variety of sources:
Images in the Story
Any images found in the story or posted by the OP should also be considered first, as long as the image is not obscene or copyrighted (or both) and has at least some relation with its contents.
New York Public Library
To depict the many scanned books without any useful images, the New York Public Library . The Processing script also does facial recognition to put close ups of the amazing engravings that many of these books have.
Abstract Patterns + Text
Sometimes for a very large story, generating Word Clouds creates quite a visually interesting.
The key to a good word cloud is all in the typeface and color schemes, as well as the layout shape that it makes. See the wordcloud bot on reddit for inspiration.
The last type of image to use is fanart. As this takes actual human labor to produce, not many covers will be able to have fanart on them.
Japanese Light Novel Covers
These would be quite good to pair with a generated image GAN with anime girl faces to depict the many /a/ stories we have. It also fits with the content well as many stories 4chan users keep archiving are wincest stories...
The key to eyecatching light novel design is calligraphy in the end. The logos are often very finely illustrated complex typographic masterpieces, with a visual style exhibited throughout the book. It will be hard to top that.
Then add in the CJK text. https://github.com/SolidZORO/zpix-pixel-font