You wouldn't curl an NFT...

[hornc]

// trigger warning NFTs (non-serious use of NFTs for meta-generative art commentary. Digital art-nonsense.)

This is a pretty terrible idea... but what if a 50K word text was generated from a series of NFTs?

Maybe if there was hypothetically some NFT series with maybe 10000 totally unique artworks, which each could be described in 5 words? Maybe with a small script which downloads every NFT, uses imagemagick to determine the background colour, and converts it to an colour name using the X11 rgb.txt colour names

#!/bin/bash

list=https://ipfs.io/ipfs/<REDACTED>
while read nft; do
  # echo $nft
  color=$(curl -s $nft |
          egrep "^\s*$(convert - -alpha off -depth 8 txt:- |
	  fgrep -m1 '70,0:' |
	  sed 's/^.*srgb.\([0-9,]*\).$/ \1,/;s/[0-9]\{2\},/\\w\\w,/g;s/ //g;s/,/\\s+/g')" rgb.txt |
	  head | shuf -n1 |
	  cut -f3,4  |
	  head -n1 |
	  sed 's/ /-/g;s/[0-9X]//g;s/\([A-Z]\)/\L\1/;s/\([A-Z]\)/-\L\1/g')
  out=$(shuf -n1 phrases.txt |
    sed "s/APE/$(shuf -n1 apes.txt)/;
         s/IS/$(shuf -n1 is.txt)/;
	 s/BORED/$(shuf -n1 bored.txt)/;
	 s/ON/$(shuf -n1 on.txt)/;
	 s/WITH/$(shuf -n1 with.txt)/;
	 s/GROUND/$(shuf -n1 ground.txt)/;
	 s/BEING/$(shuf -n1 being.txt)/
	 s/COLOR/$color/")

  echo ${out^}
done < <(curl -s $list | grep -o "https://ipfs.io/ipfs/\w*")

(edited to improve the colour picking algorithm. It's still a bit poor. Let's call it 'subjective'. Should be less prone to confusing olives and oranges.)

Vocab words and phrase structure are picked from txt files using shuf, the colour is determined by ... hypothetically ... downloading an image and getting an approximate colour value from a background pixel.

A hypothetical output might look like this:

Unimpressed primate against green-yellow setting. Critter on snow wall: miserable. Gloomy critter against snow colour. Character on slate-blue colour: inattentive. Chimp against slate-blue colour: neutral. Gloomy creature with midnight-blue background. Monkey seems neutral on midnight-blue. Midnight-blue with dude - appears disenchanted. Dulled gorilla against slate-blue setting. Dull monkey on green-yellow background. Monkey looking inattentive with snow. Green-yellow behind dude being souless. Simian by green-yellow wall: uninspired. Indian-red behind critter - appears miserable. Chimp on snow setting: deflated. Monkey against snow colour: oblivious. Primate is jaded on snow. Oblivious individual on snow wall. Critter with snow wall: inattentive. Indolent dude with green-yellow field. Critter with midnight-blue setting: oblivious. Green-yellow with dude - appears depressed. Primate looks unheeding on slate-blue. Character against indian-red setting: oblivious. Green-yellow with simian being pathetic. Simian looking glum on snow. Midnight-blue with monkey appearing indifferent. Creature by midnight-blue colour: neutral. Critter against snow field: unheeding. Green-yellow with critter - appears blase. ...

Probably a terrible idea. I'd have to figure out how to get the correct ipfs addresses directly from an Ethereum contract to execute this properly, and I haven't found a neat API / url to do that from the command line yet. Annoyingly, the metadata for the tokens has the background colour already listed, and other parameters. I discovered that after I wrote the imagemagick colour naming sed lookup thing. Also frustrating that I couldn't see a way to reduce the palette to css or X11 colours without using another pre-existing image.

The 5 words per-line was meant to be haiku-like... These lines all seem pretty fungible.

[hornc]

Better (hopefully) colour matching produces:

Beast against sandy-brown wall: uninspired. Monkey by snow wall: slack. Light-blue with creature being sad. Character against light-pink wall: dissatisified. Dismal creature on rebecca-purple wall. Sad monkey by dark-orchid color. Chimp appears slack on cadet-blue. Medium-sea-green behind chimp - appears torpid. Deflated gorilla by rebecca-purple color. Monkey appears wearied with sandy-brown. Chimp being pathetic against sea-green. Creature looking languid on sandy-brown. Torpid critter against chocolate setting. Pink behind critter appearing unimpressed. Simian on sea-green wall: deflated. Depressed individual with sea-green color. Monkey by navajo-white setting: dissatisified. Sea-green with monkey being disenchanted. Critter on sea-green setting: vacant. Unheeding beast by goldenrod wall. Gorilla against dim-gray wall: languid. Peru with monkey - appears glum. Apathetic gorilla with light-slate-gray setting. Individual appears inert on medium-violet-red. Chimp by goldenrod background: impassive. Primate looks miserable against light-blue. Slack gorilla against pink background. Primate seems wearied on light-pink. Sea-green with chimp - appears deflated. Spritless creature by goldenrod color.

[hornc]

Ok, this is done now: 50k words generated by describing 10,000 NFTs, located directly from their Ethereum blockchain contract. In bash.

• Project repo: https://github.com/hornc/NaNoGenMo/tree/master/2021/youwouldnt The code is also listed in this output pdf here:
• Output: https://github.com/hornc/NaNoGenMo/blob/master/2021/youwouldnt/NaNoGenMo_colorpaper.pdf

Every coloured sentence in the pdf links to the ipfs:// URL of the original NFT it describes, to verify whether the algorithm's description is accurate. Github (where it's hosted currently) won't let you follow the links, and following an ipfs:// URL is unlikely to work from whatever other pdf viewer you might have unless you have the correct browser / viewer plugins. I don't want to make this too easy, but the links are there in theory at least...

https://ipfs.io/ipfs/ is an ipfs -> https gateway, if anyone needs it.

[hornc]

@disastrid , hey look, I was doing silly NFT download art before it was cool ;p