This is a fresh experimentation of noise domain warping with glitches. As many of my previous plots, this is a 2-pass spiral parametric function, displaced with perlin noise warpped into each other (see &quot;domain warping&quot;). The spiral ends up being unstable and &quot;breaks apart&quot; with intentional glitches: the line is stopped as soon as two datapoints are to far from each other, as a spiral get bigger and bigger it inevitably ending up breaking apart completely in various places which combined with the parametric function makes some sort of humanoïd shape!
I sent some video while it was getting plotted. It started from the middle and the beginning is quite interesting and could itself be made a standalone. All of these are very exploratory and may be revisited later!

This is a fresh experimentation of noise domain warping with glitches

Warp Glitch 01

This is a continuation of yesterday&#39;s plot loop to try to reuse similar noises on a more arbitrary structure. This is a crossfade of two fountain pen inks: Red Dragon and Turquoise.

This is a continuation of yesterday's plot loop to try to reuse similar noises on a more arbitrary structure. This is a crossfade of two fountain pen inks: Red Dragon and Turquoise.

Caves 01

Here is &quot;Dancing Planet&quot;, my 5th <a href="/plots/tags/plotloop">plot loop</a> <a href="https://greweb.me/2021/05/plot-loops">(see article)</a>. The main digital art is a 1920p video loop of 8 frames available as a <a href="https://www.hicetnunc.xyz/objkt/105425">Tezos hicetnunc NFT</a>. The physical art are the 8 frames, plotted with 2 fountain pens on Bristol A4 paper (250g), and offered when <a href="https://www.hicetnunc.xyz/objkt/105425">buying the NFT</a>.
There are 8 plots available for sale and there will be no other editions of these plot loop frames. They all have the same inital price and the physical piece is selected in order of buy as each frame is relatively similar!
<img src="/images/plots/150-plots.jpg" width="100%">

This is a reboot of <a href="/plots/148">plot#148</a> but with a big rework of the noise technique using domain warping as well as combination with this famous &#39;rotating dancer&#39; GIF.
It is plotted with two phases, a first pass without a lot of noise and a second pass with more noise divergence:
<video src="/images/plots/150-phases.mp4" width="100%" controls autoplay muted loop></video>
<img src="/images/plots/150-zoom1.jpg" width="100%">
<img src="/images/plots/150-zoom2.jpg" width="100%">
<img src="/images/plots/150-zoom3.jpg" width="100%">

<h2 id="early-prototype-in-pursuit-of-the-good-ink-and-paper">Early prototype, in pursuit of the good ink and paper...</h2>
I started with this first prototype of the concept:
<img src="/images/plots/150-proto1.jpg" width="100%">

It allowed me to see what I could adjust, for instance: adding more noise, more lines and scaling a bit the dancer.
I used ink &#39;Quink&#39; by Parker plotted with a fountain pen on a Canson Bristol 250g paper (A4 format).
Here is another prototype I tried: it uses a mix of two colors (dragon red and turquoise). I ended up preferring minimalism: black on white. At the end I stayed on my first ink choice but I will revisit this idea.
<img src="/images/plots/150-proto2.jpg" width="57%"><img src="/images/plots/150-proto2-zoom.jpg" width="42.8%">
<h2 id="process-to-find-the-final-plot">Process to find the final plot</h2>
To search for the good parameters I&#39;ve used a generator script with bash for loops. It kinda looked like this:
<pre><code class="language-sh">for a in 0.5 1 2 4; do
for b in 0.5 1 2 4; do
for c in 0.5 1 2 4; do
for d in 0.5 1 2 4; do
 sh $P/gen.sh --a $a --b $b --c $c --d $d
 cp ./results/out.gif out/${a}_${b}_${c}_${d}.gif
done
done
done
done
</code></pre>
Then I kept refining my &quot;range&quot; by fail and retry. It is quite slow to generate them so it&#39;s a long process.
When I was satisfied by the parameters, I then switch to a second selection: the selection of the seed! because my noise have seed and can generate very different kind of noises, I put it against my &quot;elector&quot; homemade tool as explained on <a href="/plots/143">plot#143</a>.
<img src="/images/plots/150-selection.png" width="100%">

<h2 id="some-highlight-of-the-rustlang-code">Some highlight of the Rustlang code</h2>
There is nothing really &quot;new&quot; from my previous plots, especially all the parametric plot when I started this &quot;growing parametric&quot; exploration between <a href="/plots/114">plot#114</a> and <a href="/plots/148">plot#148</a>.
What is new here however is the use of a GIF as an input. It&#39;s very easy to do: my script takes basically an loop percentage which I can transpose to looking up a frame in the rotating dancer loop. I can then use this function I&#39;ve just added to my utilities:
<pre><code class="language-rust">use image::AnimationDecoder;
use image::RgbaImage;
use image::gif::GifDecoder;
use image::io::Reader as ImageReader;

pub fn image_gif_get_color(
 path: &amp;str,
 index: usize
) -&gt; Result&lt;
 impl Fn((f64, f64)) -&gt; (f64, f64, f64),
 image::ImageError,
&gt; {
 let file_in = File::open(path)?;
 let decoder = GifDecoder::new(file_in).unwrap();
 let frames = decoder.into_frames();
 let frames = frames.collect_frames()?;
 let img = frames.get(index % frames.len()).unwrap();
 let buffer = img.buffer();
 return Ok(dynamic_image_get_color(buffer.clone()));
}

pub fn dynamic_image_get_color(
 img: RgbaImage
) -&gt; impl Fn((f64, f64)) -&gt; (f64, f64, f64) {
 let (width, height) = img.dimensions();
 return move |(x, y): (f64, f64)| {
 let xi = (x.max(0.0).min(1.0)
 * ((width - 1) as f64)) as u32;
 let yi = (y.max(0.0).min(1.0)
 * ((height - 1) as f64))
 as u32;
 let pixel = img.get_pixel(xi, yi);
 let r = (pixel[0] as f64) / 255.0;
 let g = (pixel[1] as f64) / 255.0;
 let b = (pixel[2] as f64) / 255.0;
 return (r, g, b);
 };
}
</code></pre>
Rust really shines in having great library, this is just the classical &quot;image&quot; package here.
Apart from this, the Rust code (you can see main.rs link on top of this page) is pretty straightforward. There is just a lot of work in the parametric function which starts to be quite big:
<pre><code class="language-rust">let parametric = |p: f64| {
 let p1 = (splits * p).floor();
 let p2 = splits * p - p1;
 let t = (p1 + split_threshold * p2) / splits;
 let mut t2 = (p1
 + split_threshold * p2.powf(pow))
 / splits;
 let initial = 1. / spins;
 t2 =
 (t2 - initial).max(0.) / (1. - initial);
 let scale = 1.0
 - t2 * (1.0
 - i as f64 * opts.size_diff / size);
 let s = spins;
 let mut p = (
 scale
 * amp1
 * ((s * 2. * PI * t).sin()
 + amp2
 * mix(
 (s * f1.1 * PI * t).sin(),
 (s * f2.1 * PI * t).sin(),
 t,
 )),
 0.07
 - scale
 * amp1
 * ((s * 2. * PI * t).cos()
 + amp2
 * mix(
 (s * f1.0 * PI * t).cos(),
 (s * f2.0 * PI * t).cos(),
 t,
 )),
 );
 let noise_angle = p.1.atan2(p.0);
 let noise_amp = 0.003 * perlin.get([
 opts.a * (progress * PI).sin() +
 4.8 * p.0 + perlin.get([
 7.8 * p.0,
 4.2 * p.1 + opts.b * (progress * PI).sin(),
 40. + opts.seed
 ]),
 4.8 * p.1 + 0.8 * perlin.get([
 4.5 * p.0 + opts.c * ((1. - progress) * PI).sin(),
 6.8 * p.1 + perlin.get([
 20.5 * p.0 + opts.d * (2. * PI * progress).cos(),
 20.8 * p.1,
 200. + opts.seed,
 ]),
 20. + opts.seed,
 ]),
 100. + opts.seed + i as f64 * opts.seed_diff,
 ]) +
 0.03 * (1. - t) * perlin.get([
 0.7 * p.0 + perlin.get([
 2.9 * p.0 + opts.e * (2. * PI * progress).cos(),
 1.7 * p.1,
 2000.0
 ]),
 0.7 * p.1 + perlin.get([
 3.1 * p.0,
 2.5 * p.1 + opts.e * (2. * PI * progress).sin(),
 2100.0
 ]),
 1000.,
 ]);

 p.0 += noise_amp * noise_angle.cos();
 p.1 += noise_amp * noise_angle.sin();
 p
};
</code></pre>

8 frames plotted making an animated loop. A 1920p video and A4 physical art is available as an NFT.