Generative Art Setup

What is generative art?

In general generative art is defined as art that has been (at least in some part) created by using an autonomous system. I think of it as art that is created by some system, but the artist (not the system although some may argue that the system is the true artist) doesn't know exactly how it will turn out. While the artist can tweak some parts of the system (usually the inputs) they do not have full control of the generation of the artwork.

I recently discovered generative art after reading an excellent article by Anders Hoff, titled On Generative Algorithms. I found it to be a very interesting read, well worth your time, and it inspired me to make my own generative algorithms. In particular the first example, Hyphae, interested me. I thought, hey I bet I can do that in a couple lines of python!

After playing around with a couple of examples I extracted a basic framework for creating generative art that I thought was worth sharing.

The Tools of the Trade

Python, for prototyping it's hard to beat python in terms of speed of development/iteration and available libraries.

Pillow, PIL was Python Image Library until it stopped getting updated. Pillow is a fork of PIL that is actively maintained. Pillow allows us to quickly create, edit, and save images.

MoviePy, as mentioned in a previous post is a great way to generate GIFs from a series of still images. I use it to (guess what) generate gifs from still images.

Using the Tools

Here's an outline of the steps that my Python script does for each invocation

1. Create a new directory relative to this script, call it the current date so that the name won't conflict and the directory are in a sane order
2. Copy the current version of this script to the newly created directory, this allows me to jump back to a previous state if I liked the results or if my more recent changes were too drastic or if the output of a previous version was particularly interesting.
3. Do the work, for each frame/step draw on a Pillow Image.
4. Save the current image, make the name something that will allow for some kind of order such as appending the current itteration to the file name such as appending the current iteration to the file name.
5. Create the GIF using MoviePy.

The Script

import os
import shutil

from datetime import datetime
from PIL import Image

# Get the current time, use it as the directory we save everything in
SAVE_DIR = datetime.now().strftime("%Y-%m-%d %H.%M.%S")
os.mkdir(SAVE_DIR)
shutil.copy(__file__, SAVE_DIR)  # Save a copy of this script so that we can recreate stuff

# The size of the image
IMAGE_SIZE = 100, 100
# Create a new image
CURRENT_IMAGE = Image.new('RGB', IMAGE_SIZE)

# Missing: initial setup

# How many iterations?
MAX_ITERATIONS = 100
# The main loop, mess with the image for each iteration
for i in range(MAX_ITERATIONS):
    # Manipulate the image before saving it
    CURRENT_IMAGE.save(os.path.join(SAVE_DIR, "{}.png".format(i)))

    # Every so often print out something so we get a sense of progress
    if i % 10 == 0:
        print('Iteration {}'.format(i), flush=True)


def gifit(image_loc, gif_name, fps=12):
    """ Creates a GIF from all the pngs in the provide image_loc """
    import glob
    import moviepy.editor as mpy

    os.chdir(image_loc)
    file_list = glob.glob('*.png')
    list.sort(file_list, key=lambda x: int(x.split('.png')[0]))
    clip = mpy.ImageSequenceClip(file_list, fps=fps)
    clip.write_gif('{}.gif'.format(gif_name), fps=fps)


gifit(SAVE_DIR, 'final')

Example

Here's an example using the basic script from above. We first place randomly colored pixels then for each iteration we select pixels in random order and have them "fight" against an adjacent pixel. The winning pixel transfers their color to the losing pixel. In this first example we chose the pixel whose red value is larger as the winner.

import os
import random
import shutil

from datetime import datetime
from PIL import Image

# Get the current time, use it as the directory we save everything in
SAVE_DIR = datetime.now().strftime("%Y-%m-%d %H.%M.%S")
os.mkdir(SAVE_DIR)
shutil.copy(__file__, SAVE_DIR)  # Save a copy of this script so that we can recreate stuff

# The size of the image
IMAGE_SIZE = 100, 100
# Create a new image
CURRENT_IMAGE = Image.new('RGB', IMAGE_SIZE)
PIXELS = CURRENT_IMAGE.load()  # Get a modifiable representation of the image

# Initial setup, make each pixel a random color
for x in range(0, IMAGE_SIZE[0]):
    for y in range(0, IMAGE_SIZE[1]):
        PIXELS[x, y] = (random.randint(0, 255), random.randint(0, 255), random.randint(0, 255))


def get_random_adjacent(loc):
    """ Gets a random adjacent pixel that is not out of bounds"""
    _x = loc[0]
    _y = loc[1]
    adjacent_x, adjacent_y = [(_x, _y+1), (_x, _y-1), (_x+1, _y), (_x-1, _y)][random.randint(0, 3)]
    if adjacent_x < 0:
        adjacent_x = IMAGE_SIZE[0]-1
    elif adjacent_x > IMAGE_SIZE[0]-1:
        adjacent_x = 0
    if adjacent_y < 0:
        adjacent_y = IMAGE_SIZE[1]-1
    elif adjacent_y > IMAGE_SIZE[1]-1:
        adjacent_y = 0
    return (adjacent_x, adjacent_y)


def determine_winner(one_pixel, another_pixel):
    """Pick a winning color given two colors"""
    if one_pixel[0] > another_pixel[0]:
        return one_pixel
    return another_pixel


# How many iterations?
MAX_ITERATIONS = 100
# The main loop, mess with the image for each iteration
for i in range(MAX_ITERATIONS):
    for x in random.sample(range(CURRENT_IMAGE.size[0]), k=CURRENT_IMAGE.size[0]):
        for y in random.sample(range(CURRENT_IMAGE.size[1]), k=CURRENT_IMAGE.size[1]):
            location = (x, y)
            opponent_location = get_random_adjacent(location)

            winner = determine_winner(PIXELS[location], PIXELS[opponent_location])

            # Update pixels
            PIXELS[location] = winner
            PIXELS[opponent_location] = winner

    CURRENT_IMAGE.save(os.path.join(SAVE_DIR, "{}.png".format(i)))

    # Every so often print out something so we get a sense of progress
    if i % 10 == 0:
        print('Iteration {}'.format(i), flush=True)

...

gifit(SAVE_DIR, 'final')

Here is the result of this script:

You might first expect that the system would become a single large red square, however it resolves into some red, some pink, and even some yellow pixels. This is because each of them have the same value for red and towards the end the winner is simply whoever was the first chosen pixel.

The beauty of our setup is that we can quickly change a couple lines and create a completely different system. Let's modify the determine_winner function so that only the truly "red-ist" pixel wins.

def determine_winner(one_pixel, another_pixel):
    """Pick a winning color given two colors"""
    one_sum = one_pixel[0] - one_pixel[1] - one_pixel[2]
    other_sum = another_pixel[0] - another_pixel[1] - another_pixel[2]
    if one_sum > other_sum:
        return one_pixel
    return another_pixel

... and here's the result, the other colors quickly disappear and it becomes a fight to the true red (255, 0, 0)

Finally for your enjoyment here's a large version of the previous red fighting

Bonus

Some suggested additions

for i in range(MAX_ITERATIONS):
    fought = {}
    for x in random.sample(range(CURRENT_IMAGE.size[0]), k=CURRENT_IMAGE.size[0]):
        for y in random.sample(range(CURRENT_IMAGE.size[1]), k=CURRENT_IMAGE.size[1]):
            location = (x, y)
            opponent_location = get_random_adjacent(location)
            if location in fought:
                continue
            if opponent_location in fought:
                continue

            winner = determine_winner(PIXELS[location], PIXELS[opponent_location])

            # Update pixels
            PIXELS[location] = winner
            PIXELS[opponent_location] = winner

            # Update fought map
            fought[location] = True
            fought[opponent_location] = True

    CURRENT_IMAGE.save(os.path.join(SAVE_DIR, "{}.png".format(i)))

    # Every so often print out something so we get a sense of progress
    if i % 10 == 0:
        print('Iteration {}'.format(i), flush=True)

import subprocess

# Or chrome or okular or ...
subprocess.run('preview final.gif')

# Or, in Windows
subprocess.run('start final.gif', shell=True)

Final Thoughts

Have fun and mess around with this script. I've found this setup very conducive to quick iteration and experimentation that can quickly produce some very interesting results. If you made anything similar I would love to see it, I find these kinds of animated images very interesting and relaxing.