(These are not strictly needed to achieve the same result; this is just what I know works. Feel free to port this piece to other OSes or automate/optimize it)
If you wish to follow along, please read and understand these two articles first:
-
Databending using Audacity by Antonio Roberts
-
Databending with Audacity: What I do Differently/Required Reading by me
I'll assume you understand the basic idea of converting images into raw data, importing them into Audacity, applying effects to it, exporting the results, then opening the raw data in a suitable image viewer. That's covered by the first one.
The second one covers how I do databending with Audacity, and how my method differs from the one detailed in the first article.
To replicate my results, go to each respective image's source, and download the "Medium" size.
Paulstretch is an effect suitable for extreme stretching of audio, 50x or more. It essentially "smooths out" audio. Try it yourself in Audacity; open a file with some music on it, run Paulstretch at a Stretch Factor of 10 and a Time Resolution of 2 seconds, and listen to how it just "blurs" the sound together.
With that in mind, I thought to myself, "Wait, what happens if I tried this with image data?". Let's get today's test subject.
You know the drill. Convert to planar RGB RAW, open Audacity, import as shown here:
...and you'll see this. Select the entire thing.
Go under Effect > Paulstretch. You'll get this:
The Time Resolution's not important here yet; just set it to 50 seconds first. Stretch Factor, however, must be 1. The image becomes unusable nonsense if it's not 1. Once that's done, press OK.
You can listen to this if you want. It just sounds "blurred". For the eagle-eyed among you, you'll have noticed that Paulstretch has changed the length of the file. This won't affect anything, not in this article.
Let's open Irfanview, export this and open it with the correct parameters...
What? How? It's a spectacular effect, for sure. I didn't do this beforehand; I databent the image while I was writing this article. You'd think that an effect that "blurs" sound would just be a blur effect on images, but this is something else.
Let's split it into individual R, G and B planes to understand what's going on. I've done it such that R is the first vertical third, G is the 2nd third, and B is the final third.
It's like it copies the textures in the original image over and over, overlaying them into a nice final texture. I did it, and you can, too!
Paulstretch seems to follow the "texture" of the source image. Let's pull out some more test subjects:
Paulstretch them with the same settings as before, and they turn into this:
They all faintly resemble their original pictures, but it's certainly a massive change. It's completely different. (Protip: Apply the Amplify effect at default settings after Paulstretch to prevent clipping of colors.)
Let's try it on images with less-obvious textures. I'll take these ones:
Put it through the Paulstretcher, and we have:
So far, I've been using a Time Resolution of 50 seconds, because I think it produces nice results. Here's a guinea pig.
Let's experiment on it(s picture). First off, Paulstretch at 1 second.
2:
4:
8:
16:
32:
64:
128:
It seems that the higher the Time Resolution, the less "liney" the resulting image is. However, the higher it is, the more likely it is that the blue component just gets cut off entirely. I think a safe value for "reasonably sized" images is somewhere between 32 and 64 seconds, so I was justified in using 50 seconds before this.
Now, I generally don't like interleaved because it tends to lose color here, as it does with most things databending. I'll do it for the sake of completeness. I'll just pull a picture of a cat here:
Convert this to interleaved RAW, run it through Paulstretch, export it...
The color's just mostly gone, and it's pretty good pareidolia material; quite spooktacular if you told people that it was haunted or something. You'd have better luck opening this as planar RAW:
I'd say this one's way better.
This is a Paulstretched image in planar RGB. Can you figure out what the original image was? Can you reconstruct the original? This is a PNG, so it's an exact pixel-for-pixel copy of what I got from Paulstretching; no compression artifacts here.
Paulstretch... it's a wacky effect. Just transforms images into layers upon layers of textures pasted on top of each other. Try it for yourself; realizing that you've pulled off that with nothing but Audacity and Irfanview is an experience.
If there's anything you'd like to add or change, fork (and add a pull request) this repository or find my contact info on GitHub at multiplealiases.
This work (excluding the images that are explicitly credited to Unsplash) is licensed under a Creative Commons Attribution 4.0 International License.
