You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
This entry is inspired by the Voynich manuscript.
The idea is to combine a few different algorithms and Transformer models to generate a book with botanical illustrations of species of flowers that don't exist, alongside some text written in a newly generated font/script type.
I'm thinking of calling it the Codex of Aldhaka because aldhaka apparently is "intelligence" in Arabic, but 'Artificial Intelligence" was too long lol.
For this the following steps are necessary:
Done:
An algorithm to generate binomial scientific names for new flowers from the combination of some hand picked existing genus + adjectives + random Latin words
Generating a description of the characteristics of these flowers using a Language Model (probably a Transformer + prompt engineering)
Generating an image with this description + Prompt engineering (possibly using Stable Diffusion for this part)
Generating realistic looking script/characters using Midjourney
Manually editing the character rasterized images into vector graphics and transforming them into a .ttf TrueType font (could call it Aldhaka.ttf maybe?)
Writing a script to generate the final text with the newly made up font.
Using Stable diffusion to make the generated text look more realistic (the font will be too uniform to look like it was written by a human)
Maybe generate a pretty cover on Midjourney
Write some simple scripts to put everything together into a pdf
To do:
Maybe combine the flower images and text in a more creative way than just hstacking.
Maybe try to add a texture to the pages using some simpler generative art techniques.
I was not sure if this would work out or not so I did a bunch of trial and error over the month and I'm now convinced it's at least doable as a proof of concept, so I'm making this submission for Nanogenmo 2022.
Here's some examples of the ongoing WIP:
The text was updated successfully, but these errors were encountered:
Oh I forgot to post what these descriptions look like in english. Here's an example
3 Flower shape is pepulated white sepals surrounding to black lobes around petals with small black-grey 'winking patches'—noted more strongly if viewed close up from this angle, earthed through them; 2 lobicles larger overall & wider along center line —towards their own axill'
I used one of those OPT Transformers to generate these descriptions. To me it's kind of funny how they invented botanical sounding words like 'pepulated ' or 'lobicles ' or 'axill'. As far as I know none of these exist.
I'm not sure if I'll be able to improve this a lot in the next 3 days, so I'm gonna call it here.
Here's the final submission: https://github.com/HylisWilk/aldhaka
It includes a txt with all the prompts if anyone is curious to read them in english. I might put up a more formatted and pretty file to read just the English prompts.
To reach the 50k word limit I did rely partly on the generated prompts, which feels a bit suboptimal, but I do think this project makes up for it with the combination of the other models/techniques, as a proof of concept.
I'm really excited about automatizing font generation, after this.
This entry is inspired by the Voynich manuscript.
The idea is to combine a few different algorithms and Transformer models to generate a book with botanical illustrations of species of flowers that don't exist, alongside some text written in a newly generated font/script type.
I'm thinking of calling it the Codex of Aldhaka because aldhaka apparently is "intelligence" in Arabic, but 'Artificial Intelligence" was too long lol.
For this the following steps are necessary:
Done:
To do:
I was not sure if this would work out or not so I did a bunch of trial and error over the month and I'm now convinced it's at least doable as a proof of concept, so I'm making this submission for Nanogenmo 2022.
Here's some examples of the ongoing WIP:
The text was updated successfully, but these errors were encountered: