All our movies since 2006 have been on our Youtube channel. Go there to see some crazy video processing.

Here you'll find old movies in some of the nonstandard codecs supported by Cinelerra, before internet video became big. Some demonstrate effects in Cinelerra. Others try to show the quality of the codec. Others are simply examples of the kind of resolution achievable in internet downloads at the time.

Some were encoded using a combination of OGG Vorbis and MPEG-4. Others use a derivative of MPEG-4 where two streams interleave every other field. These could only be played in Cinelerra.

Still photo aerial timelapse using multiple Motion passes & Time Average.

Stabilized aerial video using Motion, Time Average, Sharpen.

HD version (99MB)

The Sunset Movie. 4185 JPG photos were shot over 2 years and combined in Cinelerra to make timelapse movies of sunsets. Histogram and Gamma were used extensively to correct exposure. ReframeRT was used to adjust speed after the initial timelapse combination.

For Youtube, the original 1080p footage was scaled down to 426x240, sharpened, and exported as H.264 + MPEG-4 Audio.

Video: Raw digital camera file
Resolution: 3522x2348
Aspect: 3x2
Size: 12 MB

Stardust was the first and last spaceship we'll ever see returning from beyond Mars orbit to a landing on Earth and Heroine Virtual Ltd. was there to capture it . Here are 2 EDLs and the 2 raw digital camera files of the spaceship as seen from Calif* and captured by the mighty Canon EOS 20D, 15mm F3.5, 100 ISO, 61 sec. The spaceship appeared in the middle of its expected path and took 45 seconds to travel to the clouds.

The spaceship was travelling too fast to see anything in the JPG safety image, but the raw images did better. One .CR2 file contains the sky with the spaceship and the other file contains the sky with no spaceship. They provided an opportunity to use Cinelerra's floating point tools.

When loaded into Cinelerra, one EDL applies histogram and sharpening to bring out the spaceship. The other EDL subtracts the 2 images to bring out the spaceship. Don't bother loading these files without 2 gig of RAM, a copy of Cinelerra, and a long time to process the images.

Click on an image to get a JPG of the output. Click on PROJECT FILES to get the raw data.

Video: JPEG
Resolution: 1024x1024
Framerate: 15
Aspect: 1x1
Size: 29MB

The only moving images of the Deep Impact event captured from Earth are here. Through the power of the dual Opteron and Cinelerra, we have subtracted the ambient light to emphasize any changes in brightness of the comet. Furthermore, every 4 frames were averaged in a moving average to reduce noise.

85 30 second exposures were taken by a Canon EOS 20D, 200mm F2.8 L II, and trapezoidian mount. The ISO was 1600. Extra motion tracking, white balancing, and stacking were done in Cinelerra. The location was a mountaintop somewhere in the Morgan Territory, Calif*.

Observing location:

The trapezoidian mount is a type of Scotch mount, but is light enough to carry up a mountain even if you're not a heroine.

To remove ambient light, a gaussian blur of 500 pixels was applied to each frame. The blurred images were subtracted from the original images. As the comet moved towards the horizon and the ambient light got brighter, the blurred images also got brighter. The subtracted frames appear to have the same background light even though the original movie was much brighter at the end.

Video: Variable bitrate MPEG-2
Resolution: 960x540
Framerate: 29.97
Aspect: 16x9
Size: 17 MB

When the motion tracker was developed for Cinelerra, this was the first test project. The background image, of Heroine College of course, was added when the motion tracking was done.

The idea was to make a piece of text move around the screen with rotation. The motion tracker was supposed to keep another piece of text exactly aligned with the moving text. For extra excitement a crosshair shows just how close the motion tracker got it.

This particular setting tracked the relative motion between every frame and the previous frame. This tends to lose accuracy over time. A more accurate setting tracks the absolute motion between one frame and every other frame.

Video: Variable bitrate MPEG-2
Resolution: 960x540
Framerate: 59.94
Aspect: 16x9
Size: 112 MB

This stream shows the motion tracker tracking a flashlight in a long dolly. The tracking is relative. What was a very unstable camera move looks a little more stable with the motion tracking. To improve the motion tracking, the original interlaced footage was deinterlaced and doubled in framerate with the Frames to fields effect.

Video: Variable bitrate MPEG-2
Resolution: 1280x720
Framerate: 23.976
Aspect: 16x9
Size: 70 MB

If you're flat broke, unemployed, or American you probably can't afford a 35mm movie camera. There is still a way to make pretty high quality movies using the motion tracker and a digital still camera. In this movie 104 high resolution images were taken with a digital camera.

They were tracked using relative motion tracking. A specific part of every frame was matched to the same part in the previous frame. The rotation and translation was compensated to convert a series of extremely uneven frames to a reasonably viewable dolly. The Interpolate effect finally increased the frame rate by averaging frames.

Video: Variable bitrate MPEG-2
Resolution: 1280x720
Framerate: 59.94
Aspect: 16x9
Size: 48 MB

This movie shows just how little the motion tracker needs to track a target. The original footage had changing brightness which made absolute tracking impossible. The target separated from an object behind it, which made the motion tracker want to track the wrong object. Finally the target moved around wildly, which required a very large search radius.

Only by tracking an extremely small region of the frame, the NASA logo on the side of the missile, did the motion tracker stay locked on. That allowed it to track the missile for the entire duration.

Video: Variable bitrate MPEG-2
Resolution: 1280x720
Framerate: 24
Aspect: 16x9
Size: 107 MB

Point a digital camera at the sky and you won't see anything. Combine massive numbers of exposures and you'll see extremely faint objects. The exposures need to be motion tracked and rotated to compensate for the Earth's rotation. Unlike normal motion tracking, this motion tracking needs to be extremely precise to avoid blurring.

Here 347 exposures were hand matched with motion curves to within 10 pixels of exactly one exposure. This didn't match accurately enough to enhance the image but reduced the search radius for the motion tracker, making it faster.

Absolute motion tracking matched all 347 frames to the starting frame. The translation step tracked a single star. The rotation step tracked a very large area to get the most accuracy. Since logarithmic motion tracking doesn't work with discrete points of light like stars the search steps needed to be exhaustive.

The time average effect was applied to the matched frames. To conserve memory the time average effect was run with subtraction disabled. The time average was also performed in floating point to expand the dynamic range.

Finally the resulting floating point image was brightened and sharpened to reveal objects millions of light years outside our own galaxy, far below the limit of the camera's CCD.

Video: Variable bitrate MPEG-2
Resolution: 1920x1080
Framerate: 23.97
Aspect: 16x9
Size: 88 MB

One day someone asked what it would take to launch an oil tanker to Mars. While not as good as Japanese or Indian animation, Rocket Tanker depicts one possible scenario in full HD.

It's 20 years since the last time we ever attempted to do cell animation at Heroine Virtual. Back then the most a flat broke programmer could do was redraw the same image thousands of times on a paper pad and flip through the pages real fast.

The technology improved considerably since then. Today real cell animation can be achieved without expensive plastic sheets. Different layers can move at different speeds. Changes to the timing and stacking are now automated. Most of all, cell animation can be made at extremely high resolution without an expensive camera.

Rocket Tanker was the simplest possible cell animation. It took 2 days. All the cells were pencil drawings on paper. Here are all the cells in Rocket Tanker.

Using 3.5" x 2" frames made the drawing a lot faster and cheaper, though it would have been easier if each frame was denoted by crosshairs in two corners, not rectangles.

The cells were all photographed on a 2560x1920 digital camera. They were stacked in Cinelerra using multiply operations. The interpolate effect was used liberally to create the illusion of inbetween frames. Camera and projector curves created all the translation motion. Background rendering was crucial to getting the timing right.

Heroine Statue
Video: 720x540 1 MBit/sec MPEG-4
Audio: 44100 130kbit Vorbis.
Composited still photo movie. TV commercial soundtrack.

Title Demo
Video: 640x480, 29.97fps, 1 MBit/sec, MPEG-1
Titler animation.

(C) 1997-2017 Starving, flat broke Programmers
Hosted all these years by Sourceforge