I spent some time earlier in the week refining the drawing algorithm even further, focusing on color palette selection and line weight distribution. I think I’m getting closer to something the produces pretty consistent results.

My next step — or perhaps experiment — will be to create some color palettes based on actual color fields that I produce using real paint or some other image source. I want finer control and intention over the possible color palettes, and while toxiclib’s color utilities are pretty amazing, I’m finding that the ColorRange constraints produce less-than-desirable results most of the time (too much or too little saturation and brightness at undesirable moments).

Here’s a demo of the Processing app so you can see the renders in realtime:

And here is some more hi-res output from the app:

+ View hi-res.

+ View hi-res.

+ View hi-res.

+ View hi-res.

+ View hi-res.

+ View hi-res.

+ View hi-res.

+ View hi-res.

+ View hi-res.

+ View hi-res.

+ View hi-res.

Here’s a continuation of the Line Sketching post I did last week. What you’re seeing here are larger, higher-resolution renders that use a further-tweaked drawing algorithm, concentrating more on form and color.

The key to these pieces is seeing them in their native resolution. There’s so much texture and pseudo-form that emerges from subtle shifts in a single line over time. Here’s an example:

Full View

+ View it large!

Detail View

I need to continue tweaking the drawing routine and color choices before I’ll be happy enough to take these to a printer. Also, I’m going to be sticking any additional interesting output from this thing into a Flickr set called “Line Topography”.

In the mean time, here are some interesting ones that came out of the last 50 renders I did yesterday:

+ View it large!

+ View it large!

+ View it large!

+ View it large!

+ View it large!

+ View it large!

+ View it large!

+ View it large!

+ View it large!

+ View it large!

So I’ve been feeling less than creative and less than productive for the past 45 days or so. Yesterday I realized I needed to make something even if it was crappy.

Examining some of the recent things I’ve been inspired by and even some of my older posts here, I drew upon the following line-heavy references as my starting point:

• visp Output Sample 001 (from 2007)
• “flow” from Merdan Agayev
• “line drawing, unstable/uneven surfaces: carpet”
• All the work of Jean-Pierre Hébert

I’m in love with line. I’m especially in love with really dense linework that yields faux-3D forms and false topographies.

So I jumped right into Processing with a simple idea:

1. Let the user pick a horizontal line of pixels from camera input.
2. Analyze those pixels, deriving an array of values based on pixel brightness.
3. Create a vector line whose peaks and troughs are based on this array of values.
4. Use a bunch of Math.random() and Perlin noise and magic to draw this line on the screen from top-to-bottom.

Four hours later, interesting stuff started coming out. I’m making heavy use of GSVideo for camera input and toxiclibs for Perlin noise and color.

My goals for the end of the week will be to finesse the color logic, have the “input” line morph through two or three forms throughout the process of the render, and do some post-processing in Photoshop to get the colors to unify even further. I want to make some large prints. I think I also want to do a couple hand-drawn renditions based on some of the digital output.

Here are some more Instagram’ed crops from the first wave of output:

Finally we’ve arrived at the final image to be used for the album art of Mode7‘s upcoming EP “Dusk Drops” (see above). There were a lot of contenders, but in the end, this image satisfied a certain visual balance and the general feeling of ominousness that we were looking to convey.

For these final renders, 2850×2850 images were generated by Sunflow, using the special Thinlens camera which allowed me to simulate a shallow depth of field. Each image took about 12 hours to render on a quad-core Intel i5 Windows 7 box.

Further post-processing was done in Photoshop to simulate things that we’d see in traditional photography: vignetting, chromatic abberation, and a subtle level of visual noise. Some color tweaking and other manipulations were done as well just to make the images a bit richer.

For the CD label itself, we’re going to use the following image:

+ View the piece on Flickr.

And then there were a few more images that I decided to render and treat, just because they were compelling to me and I wanted to see what they’d look like if I took them through the entire process.

+ View the piece on Flickr.

+ View the piece on Flickr.

+ View the piece on Flickr.

I’ll be following this post up with a video demoing the Processing app that I created for generating these compositions once they EP has been officially released in a few days.

After the last post, I went in and tweaked a couple of parameter ranges and how the noisy terrain was rendered and generated another 3,500 compositions, which you can see here:

+ See it super large on Flickr.

After the computer churned these out, I spent a few hours looking through them and managed to cull the set down to an even 350. But as I mentioned before, the next step was to render these using Sunflow, which I did right before I left for Minneapolis. Rendering the 350 images at 1280×1280 with no anti-aliasing and a very low sample setting took about 29 or so hours on a quad-core Intel PC. I had it dumping the output into a Dropbox folder so I could check on the progress on my phone during my trip.

Tonight, I took the next step and chose a final set of 29 images that are contenders for the final render. In that render, the visual noise will fall away considerably, and I’ll be generating a 2850×2850 image suitable for a 300dpi CD cover print.

You can see the 29 images I picked in this Flickr set. As of right now, these are my five favorite renders:

+ View the piece on Flickr.

+ View the piece on Flickr.

+ View the piece on Flickr.

+ View the piece on Flickr.

+ View the piece on Flickr.

That is, they’re amazing at working tirelessly and generating variations, given the proper constraints.

I’m in the process of creating some album art for a friend of mine, Mode7, for an upcoming EP release. I wanted to create something that was simultaneously intentional (i.e. I had ideas for an image) and generative (i.e. I’d let the computer make stuff up too). For the last couple weeks, I’ve been slowly building a Processing app that helps me generate compositions given a curated range of possible random values.

Last night, I asked my app to show me 10,000 variations on a theme. Here’s what it came up with:

+ See it hi-res on Flickr.

And here’s a zoom of a 6×6 grid from that output:

And here are some random selects that I never would have come up with on my own:

These images are rendered in an OpenGL context, so the lighting is basic and even throughout. But this is only the first step in the process of creating the final image. The Processing app is a tool for creating compositions and color schemes. In the end, my selects are saved out as Sunflow scene files and rendered in Sunflow with global illumination at a super-high resolution. Here are a couple test renders to give you an idea of the end possibilities:

There are a couple things I want to note about this whole process.

Symbiosis

What has happened in this process of creation is the emergence of a symbiotic relationship between me and the computer. I started with clear intention for the type of image I wanted to create. As I began programmatically building the structures of the elements I wanted to use in the composition, it wasn’t until I saw the way the computer rendered my forms that I would continue moving forward. These renders influenced my direction and provided a sort of course correction.

The same happens when I ask the computer to provide me with a series of choices. I become the curator and the computer becomes the creator. It seemingly influences me as much as I influence its output. This is especially true when you introduce Math.random() into the relationship. That singular function is the computer’s very powerful voice.

This whole process is like a dance or a dialogue. And I’m beginning to learn how powerful this can be in discovering possibilities for creative output that I wouldn’t have been able to achieve on my own.

Discovery

When I tell the computer to give me 10,000 choices, the process of reviewing those choices makes me feel like I’m a scientist or an archaeologist exploring unseen worlds or newly revealed information. I’m seeing something wholly new. A world that had the possibility of existing, but only up until last night, never actually existed. And here I get to discover that new world.

There’s something so utterly enchanting and delightful about this. Flipping through these images brings me closer to some level of innocence and awe that I simply don’t encounter very often in my life. And now I’m hooked.

+ View large on Flickr.

+ View large on Flickr.

And tonight, got those vertex normals calculated. Much easier this time around — probably just because I understand it all conceptually now.

I’m really enjoying how this stuff is looking. Finally, I feel like this digital generative work feels as organic and alive as the source of the data itself. I can’t wait to add the UI controls for “sculpting” the composition and colors… and can’t wait to add more structural elements.

Now that I’m unpacked and settled in my new apartment — and since I have a 50 minute presentation coming up at FITC in just over a month — it was time to gain some momentum and get back on top of this stuff for Flow.

I’ll be the first to say that these definitely bear a strong resemblance to Evan Roth‘s latest Graffiti Analysis video called “Graffiti Analysis: 3D” — because they do. And because I was definitely inspired by it (just like I was inspired by the original project when I first saw it back in 2005 at FITC).

But this is just one layer of what I’ve finally settled on for my first real “pieces” to come out of this Flow project. I’m focusing on isolation of a few strokes, creating a strong composition, and building something structural — almost architectural — out of these energetic moments of frozen time.

The hurdle today was learning to draw a tube in space…this took the better part of the day. I would’ve used the GLE Tubing and Extrusion library, but unfortunately, that has dependencies on GLUT, which Cinder does not use. And I didn’t think that I’d be able to vary the thickness of the tube over its length, so I decided to try it from scratch. And we’ve got [decent] success. I do want to implement proper per-vertex normals to smooth the sucker out some more. I think I’ll be able to rip the per-vertex normals code from the old Noise Tube experiments, since it’s ultimately the same mesh structure.

And yeah, this is all OpenGL still. Once I get the other structural elements in place, I’ll be writing some sort of Cinder Block for creating a Sunflow scene / scene file and rendering this stuff out in Sunflow. The final aim is to create some prints from this thing. I’m giving myself till the end of the week to wrap up this visual direction, so I can focus on refactoring and documenting all the behind-the-scenes Flow applications for open source release around mid-August.

You can see the full set of this first bunch of renders in this Flickr set.

+ View large on Flickr.

+ View large on Flickr.

+ View large.

+ View large.

Lots of new stuff learned over the weekend. Here’s a summary:

• Really understand basic OpenGL lighting now.
• Understand how to do basic material coloring now.
• Learned how to texture a quad!
• Starting to learn some things about the Boost C++ libraries.
• Finally remembered that OpenGL is a state machine.

I think I actually (finally) have enough know-how to execute what I’ve got going on in my head. It’s going to require some refactoring of code, but I don’t feel like I’m up against an unclimbable wall anymore, which is precisely how I felt by the end of the day on Sunday.

I still think this looks boring / uninteresting, but it’s just another step closer and a lot of lessons learned.

Here’s a video render of what it looks like in realtime:

Update on 04-29-2010: The C++ framework that I’m using has just been released to the public. It’s called Cinder. Big ups to Andrew Bell, Robert Hodgin, and The Barbarian Group for all the time and effort put into releasing such a powerful and fun open source framework for creative coding!

Below is a 60 second render from the data set captured from this drawing.

Aesthetically, this is completely off the mark from what I’ve got in my mind’s eye. But in doing each of these exercises, I’m getting more familiar with C++ programming and openFrameworks and the various add-ons that exist out there.

And I’m also refining my general goals for the first set of “pieces” I want to create with this Flow system:

1. Show the energy and motion of each stroke. This needs to be visible even in a static render.
2. Avoid random as much as possible. The work needs to exhibit clear intention.
3. Create something dimensional, structural.
4. Work within “moments” of the data set, not its entirety.

I’m going to try to make sure whatever it is I come up with adheres to these main goals for now.