DIY-ai-art VS pyconar-talk

Here's a generative art project I did a while back: https://github.com/maxvfischer/DIY-ai-art

It's not so much about creating the generative algorithms, but more if you wanna wrap the learning around a fun project.

Last year I built an AI-art installation from scratch and posted the DIY-documentation here on HN [0][1]. A big part of why I built the installation was to stir up this exact discussion. To me, art is about emotions and making people feel, even if that feeling is their strong opinion on art.

[0] https://github.com/maxvfischer/DIY-ai-art)

[1] https://news.ycombinator.com/item?id=28221904

DIY AI art

I have a question regarding this DIY AI Art project - https://github.com/maxvfischer/DIY-ai-art

This is amazing! I recently shared a project I've been doing, building an installation visualizing ML-generated art (https://github.com/maxvfischer/DIY-ai-art). It would be amazing to try your painting model on top of my StyleGAN.

I’ve also written an extensive guide if you want to build your own installation: https://github.com/maxvfischer/DIY-ai-art

Start by copying some existing example code and running it locally, then edit it and see what changes. Comment pieces out, look at the results. Change magic numbers to understand the effect. It probably has some calls to a random number generator in it; add more calls to the random number generator.

There are lots of examples bundled with Proce55ing, on Shadertoy, on bl.ocks.org, on ObservableHQ, on Jared Tarbell's website, in the Coding Train vlog, etc. My own repo of examples using Python and PyGame is at https://github.com/kragen/pyconar-talk, but I've also done examples like http://canonical.org/~kragen/sw/dev3/tweetfract.html with (you have to click on the invisible to see it) and http://canonical.org/~kragen/sw/dev3/plotiir.html. Start with small things.

There's probably some kind of awesome example repo out there for deepdream ANN stuff but I don't know what to recommend.

But that's just where to start. Once you're doing stuff you'll want to understand what you're doing and learn about more techniques (algorithmic, software design, and interfaces to libraries and devices) so you can expand your range. There's lots of resources out there (Tarbell in particular has given an hour lecture you can find on YouTube about what techniques he finds useful) but I can suggest:

∙ Many instances of the same thing that differ by incrementing a variable. For example, you can create 64 particles that move from point A to point B at successive points in time 30 milliseconds apart, or at the same point in time at 64 different velocities, or 64 Bezier curves from point A to point B that start at 64 angles evenly spaced around a circle.

∙ Adding randomness to things. Adding randomness to pixel colors gives you "graininess"; adding randomness to object positions gives you spatial dispersion or, if the randomness varies over time, jittering; adding randomness to the angles of different objects gives you visual variety.

(to be continued)