Week 4 had us studying a coloring algorithm I’d never had any luck with: Rose Range Lite. I’d tried it a few times and produced nothing interesting, leading to a complete abandonment of it in favor of easier results elsewhere. My first dozen homework images also sucked, and I was concerned I was getting nowhere. Finally, stuff clicked and I produced what I think is a pretty hip image. I turned it into another little artistic movie study:

Orca Rose

1280×720 still image of Orca Rose

[See post to watch QuickTime movie]

Low: 8 MB
Med: 24 MB
High: 268 MB

The music is produced with an interim setup — I’d gotten totally sick and tired of not being able to use my favorite sound libraries on my fire-breating Intel MacPro, so I installed my music stuff on a G5. I get to it via Remote Desktop, which really sucks (particularly because a headless Mac only goes up to 1024×768 — a tiny window on my giant 30″ Cinema Display!) but at least I can play with libraries that haven’t been updated to Intel yet! The final two straws were when I read that Spectrasonics wasn’t going to update Atmosphere until this fall, and my desire to get the amazing closeout deal on the East West Quantum Leap Platinum XP bundle prior to their Play update. So, a lot less productive in terms of efficiency, but I’ve got great sounds! Last week’s artistic study featured EWQLSO, which sounded fantastic; this week’s features a bunch of my all-time fave plugin, Atmosphere, plus two cool sounds from EWQL’s Ra.

Week 5 moved on to one of my favorite coloring algorithms, Orbit Traps. These, however, weren’t the ordinary ones in the standard formula file that I’d always used — they were the souped up expert versions from Damien Jones. Whoa, there’s a boatload of parameters!

Janet had us do all sorts of whacky things that helped push the creativity. Very fun!

Brushed Tentacle

Same Julia as from week 3 using different trap shapes on multiple layers.

Arjuna Leaves For The Front

Same Mandelbrot as from week 3 using the same trap shape with different parameters on multiple layers.

The Steed of Vipacna

Same Julia as from week 3 and the other Julias of this week, using a few trap shapes with limited iterations.

Marbling 1

Same Julia as from week 3 and the other Julias of this week, multiple trap shapes on single layers.

Our online pharmacy is the perfect resource for people to get their drugs without any hassles or awkwardness. buy cialis We work hard to make sure you save money every time you shop with us. buy levitrabuy soma At our online store, you pay less and get more. buy viagra

[See post to watch QuickTime movie]

Low: 4.4 MB
Med: 9.4 MB
High: 162 MB

This study extends one of my still images from week 3 of the “Working With Ultra Fractal” class. In it, I had used textural layering techniques to create an interesting rusty-metal look. This study animates most of those layers, but in a loosely coupled way instead of rigidly creating a single image. It gives the movie a neat “flying” feeling. This inspired the music, which played up on that feeling rather than my original “corroded” concept of the still image.

Nitty gritty and technical mumbo-jumbo

This study was actually fun to do, as opposed to my technical ones that have been routinely exasperating. I didn’t pay a lot of attention to the machinery, which left space for some art — nice change of pace!

Music

This was the first time I’d used the symphony orchestra instruments from East West Quantum Leap - Platinum. They are amazing! I’m only scratching the surface here — so many articulations I feel lost. I can easily hear spots that aren’t quite convincing, but the overall effect from just blindly choosing things that are in the right ballpark is pretty inspiring. This is going to be a key library for me, and will take a lot of time and attention to master. I can *finally* produce orchestral music that sounds good! I’ve been waiting 25 years for this…

The particular piece is (once again) just tossed off in a handful of hours. The ending is particularly bumbling, but I really like the theme. I’m sure to use something like this in a real composition, and its likely to make its way into my movie. I know I’m on the right track when my wife says “it sounds like something that should be in the beginning of a PBS documentary.”

Image

I wish I could take credit for the overall effect, but I have to admit it was totally luck.

I started with the “Corossus” image and started by simply rotating in. I’d forgotten that I had several layers that were “disconnected” from the others by different rotations and zoom settings, which was crucial for creating the still image’s texture. When I saw the first test renders I knew I was on to something — because of the decoupling, the various layers moved somewhat independently. I tweaked this to emphasize the pseudo-parallax effect: different layers moving at somewhat different speeds makes it seem as if the faster rotating/zooming layers are closer to you, with the slower ones farther away.

I also learned a “new” feature of Ultra Fractal, or at least to me: you can rotate and pan the image with your mouse. Duh. I’m sure that’s how everyone else in the known universe does it, but I had painstakingly tweaked the location and rotation numbers of various layers by hand to get the slight offsets that give both Corossus and The Fields of Island One their pseudo-3D effect. Much, much easier to isolate location edits to a single layer and simply grab it and move it with the mouse. I can’t believe I’ve used this program for almost four years and never knew about that, even though it’s blatantly obvious from the on-screen help!

In any event, that made lining up the different sized julia-dusts about 2/3’s of the way through trivial instead of nearly-impossible. I think it’s a neat effect. The last frame of the movie shows three layers of julia-dust converging perfectly together, whereas this other one shows off the self-similarity of julia zooms by superimposing two that are nearly the same but not quite.

Shake

I stupidly rendered the final HD version at the wrong resolution — final crop was incorrectly set to 1280×760 instead of 1280×720. Even my fancy render queue can’t save me from myself when I’m an idiot. An additional problem was that I’d originally rendered 33 seconds of animation, and when the music was done, it was a 39 second cue. I didn’t want to take another 24 hours to re-render everything, so I turned to the magic of Shake.

The more I use this tool, the more indispensable it becomes. It was a simple matter to set up a resize node for the right output dimensions, and to retime the incoming file to 33/39’s (=0.8462) of the speed. Voila, problem solved with about 2 minutes of setup and 15 minutes of Shake rendering.

One of the keys I’ve learned with this pipeline is to render images somewhat larger than what I need — about 10% bigger. I then crop it down (which my queue does automatically for me.) The reason for this is that several Shake nodes muck up the edges. The warp node used in Study #4, for example, left wavy artifacts on the edges that you can see if you look closely. The wonderful retiming feature also makes the edge look all wonky because it has a hard time figuring out the pixels’ vectors when they disappear off the edge. Thus, rendering a bit big, running stuff through the Shake tree, then cropping ensures that the image is good all the way to the edge.

This retiming thing is huge for me. As I calculated in the render requirements post, I need about 186 Ghz worth of processing power (about 55 3.4 Ghz machines!) to render my movie — at least with the assumptions of that post. I can’t afford that, so I need to change the assumptions. I’ve done some test retimings at a ratio of 10:1, and for slow stuff, it looks fantastic. Fast stuff doesn’t work so well. I figure I can average 2:1 or maybe even 3:1, which dramatically reduces my computing needs.

Combined with the possibility of cutting the movie down to about 60 minutes, I’m now at 186 * ( 2/3 length ) * ( 1/2 retiming ) = 62 Ghz with no other artistic compromise. Throw in some additional restrictions — I really do have to limit deep zooms and expensive coloring algorithms, and might need to more aggressively retime less important texture layers — and I can probably reduce by another 50%. So now we’re in the 30 Ghz total range, which is close to what I’ll have once I bring all my existing computers online. I’ll put off ordering any new machines until the fall when I have better metrics on actual throughput as well a real need for speed. Currently, my creative input is the bottleneck, not the rendering throughput.

Final Cut and Codecs

I installed the newest version of Final Cut Studio last night. Can’t say that much has changed on the surface, but underneath it’s got some key improvements. First and foremost for me is that the new Apple ProRes 422 HD codec rocks. The quality is excellent, and all the actions inside of Final Cut don’t require rendering. My workflow is reasonably fast now.

Most important is that the output I’m seeing when rendering HD output is good. I think Corossus isn’t as hard to make look good as some other potential images, but I’m much happier with the results I’m seeing so far. It doesn’t look like roadkill.

I didn’t render individual layers this time — instead, I had UF render the flattened version. I’m sure it would have been fun playing around with some additional tweaks with separate layers, but I’ll save that for the next artistic study.

I tried using Compressor to create the final movies but failed miserably. It looks like it will be absolutely terrific once I figure out how to use it — single source, multiple targets, exactly what I need! I currently have to render final output for the various resolutions one by one now, which is a bit of a pain. Compressor can easily do multiple output resolutions, and can even move them automatically to my blog server when done. I’m also very keen on getting Compressor spreading its load over all machines — compressing video is very time consuming.

Render Queue

This was the first real piece to go through my new render queue. I’ve achieved most of my requirements for the queuing system, although I have a bit more to go. It was quite a relief to simply submit the job and not think about it. What’s done is:

• Submit a render job by simply saving a UF file to disk
• Automatically spread over any number of computers, with approximately linear scaling
• Automatically handle render computers coming and going
• Easily monitor progress
• Render computers automatically submit chunks of work to Ultra Fractal, and automatically recover from errors.
• Completed renders automatically sent to Shake for assembly into a Quicktime movie, including cropping to final image size
• Automatic file management, including consistent enforcement of filenaming standards and guaranteed traceability of source files to final deliverables.

What’s left to do for the render queue is:

• Easily pause or delete jobs. Currently can do this manually, so not a huge deal, but certainly want to add it.
• Automatic splitting of layers into multiple jobs. Currently can do this manually, but this is a major must-have.
• Reduce queue overhead — currently a bit over 10% because either my command-line driver or the virtual machine environment makes UF crash a lot. Although the queue automatically handles these sorts of errors to guarantee the output is OK, it slows things down a bit. This is a nice-to-have.
• Get all of my available machines participating. I’m close to feeling confident enough in the setup to spread it out over all of the family Macs and four dedicated render machines.
• Automatically retime jobs. As discussed above, retiming in Shake is amazing and crucial. I’ll end up doing this frequently, so I’d like to integrate this into the automated flow.
• Automatically transcode into Apple ProRes 422. Finally, found a codec that doesn’t suck! This is obviously a vital timesaver in the pipeline.
• Get Qmaster working. Technically not part of my render queue, but a vital part of the pipeline. This would enable me to spread out Shake, transcoding, and final compression jobs over all available Macs. Apple provides a slick interface to do all this, but haven’t gotten it working yet.

I’ll be sure to write a complete description of the queue when I finish up the loose ends.

We’ve done the first three lessons, and even though I started making fractals in 1989, I have learned a lot. The first two lessons focused on really getting familiar with the Mandelbrot (and its Julia sets). Although I thought I already knew about things like the Seahorse and Elephant valleys, there’s a lot of subtleties I’d never picked up on. Even little stuff like reliably finding mini-mandelbrots — it was hit or miss for me up until I went through the lesson. Now I can spot one a million miles away (literally, if one takes the zoom ratios inherent in fractal zooming seriously…)

The third week has been a bit more artistic. We finally opened up to different coloring algorithms and layers. The two coloring methods (Gaussian Integer and Smooth) we’re using are easy, bread and butter ones that I know like the back of my hand — or so I thought until working through the exercises! I found out that I had previously been scraping only the easy surface on these, and that all sorts o