Making simple stop-motion animations using Linux and a DSLR camera

movie clipThis articles describes how to make simple stop-motion animations using the most basic of equipment -- and ordinary digital SLR camera, and Linux. This method won't give you results that will rival the CGI sequences in Star Wars, needless to say. Rather, it's something you might like to try with your kids on a rainy Sunday afternoon -- it's fun, and inexpensive, and will produce something you can put on YouTube.

What you need

Basic principle

The principle of stop-motion animation is extremely simple. You position your figures and take a picture. You move them very, very slightly, and take another picture, and so on. For reasonably smooth animation you'll need at least 10 pictures per second of action -- 25-30 is more typical. That equates to 600-1800 picture per minute of action. The way to get smooth movement is to make very, very tiny movements of the animated figures between each shot.

When you've got your large number of photos, they can be edited, if necessary, and assembled into a movie clip. Under Linux, you can do the assembling using the mencoder utility, among other things.

Taking the pictures -- photographic issues

The three main problems with table-top stop-motion work are (1) keeping everything still between shots, (2) getting decent depth-of-field, and (3) focusing on the right spot.

Keeping everything still is a mechanical problem rather than a photographic one. It requires practice, patience, and a very solid base. I use a piece of solid wood about two inches thick that I scrounged from a kitchen fitter, screwed to a pair of decorator's trestles. You wouldn't want it to fall on your foot, but the assembly is heavy enough to resist all but the most determined attempts to knock it. Sadly, the animated scene itself can't be stabilised by brute force -- you need a mixture of diligence, patience, ingenuity, Blu-Tak, and glue.

Depth-of-field problems

Getting decent depth-of-field is, however, a technical problem, caused by the fact that the camera lens is typically very close to the action. With inadequate depth-of-field you'll find that, for example, a figure's nose can be in sharp focus, while its eyes are blurred. Of course, sometimes restricted depth of field can be used for creative effects, but normally you'll want at least all parts of a subject figure's face in focus at the same time.

Increasing depth-of-field in close-up work typically means stopping down, often far more than seems comfortable. Depending on the lens and the distance to the action, you might find you need f/22, where f/8 or even f/5.6 would do perfecly well for an ordinary scene in the same lighting levels. With such tiny apertures, you're going to need very long exposures to get enough light into the sensor for a low-noise image. It's tempting to turn up the ISO sensitivity of the camera to reduce the exposure time, but in many cases the increased noise in the image renders this ineffective. With my Nikon D50, I've found that images are less noisy if I allow the camera to underexpose, and increase the exposure in software afterwards. For this to work you'll probably need to capture raw images and batch-convert them, which is another step in the workflow.

A further complication is that, to get the best results, you won't really be able to use daylight. It takes a long time to do stop-motion work, and if you use daylight it will create two problems. First, the level of illumination will change. This means that you'll have to either use some sort of exposure compensation in the camera, or normalize the exposure after capture (another workflow step). In practice, the automatic exposure mechanisms in cameras are too coarse-grained to cope with gradual changes of illumination, and post-processing really is the only option. The second problem with daylight is that it's bluer at midday than morning or afternoon. Our eyes automatically adjust to these shifts in white balance, but the camera won't. To some extent you can post-process this away as well, but it's difficult and time-consuming, and only works if the scene itself doesn't have many colour changes.

In practice, I've found that decent results require a black-out and artificial light, with the camera set to manual operation with fixed shutter and aperture. With two 150W halogen lights illuminating the scene, I find that I need exposures of around a second to get best results.

Tricks for increasing depth of field

I've found that my Nikon doesn't handle tiny apertures and very long exposures partciculary well. I suspect this may be true for other cameras as well, especially when there are extreme colour contrasts in the scene. So there's a practical limit to how long you can expose for, and it might not be enough.

There are couple of tricks that can be employed in this situation. First, most modern DSLRs are capable of recording a much higher dynamic range than an LCD screen can show. Even if the result on the preview screen looks quite dark, you might be able to increase the exposure in software without introducing too much noise. This means that you can reduce the time the shutter is open without increasing the aperture. But this approach needs experimentation and careful previewing, because you'll get a noisy image if you overdo it. Some movie editing applications allow exposure adjustments to be made to the assembled clips, but in my experience it's much better to do this with the raw data that comes from the camera, if it has a 'raw' mode. You'll need a way to do this in batches, because there's going to be an awful lot of raw images to process.

The second trick takes advantage of the fact that most modern DSLRs shoot with a much higher frame resolution than you need for a movie. A standard DVD only has about a half a megapixel frame resolution, where even a modest camera will typically provide in excess of six megapixels. What this means is that you can position the camera much further away from the scene than you need to include all the action, and then crop out the relevant area in software. Again, you'll need a way to batch-process this. By putting the camera further away, you reduce the range of distances that have to be kept in focus, so you can get away with a smaller depth-of-field. It isn't that you get a better depth of field with the camera further from the action, but the depth-of-field you do get will be more adequate.

Focusing issues

For close-up photography, small errors in focus are very apparent. And in animation, you won't have to move the subject very far to make all the difference between being in sharp focus and being blurry. This, of course, is just the depth-of-field issue we discussed above. But since we do have limited depth-of-field, we need to be especially careful with focus.

Ideally, every frame would be individually manually focused. This is very time-consuming, and not very accurate when you're working in less than perfect lighting level. So long as the action of the animation is centered on a single figure, and that figure remains in centre shot for the whole scene, then you might be able to let your camera's autofocus take care of it. That's assuming, of course, that there's enough light and the scene has enough contrast for the autofocus machinery to work. I find that sometimes it is necessary to illuminate the subject that needs to be in focus with a very bright torch, so that the autofocus can lock on. Then set the camera to manual focus before taking the shot.

Batch editing the images

For each second of action in the scene, you'll have 12-30 images. Before they can assembled into a movie clip, there are various batch-processing steps you may need to apply:

Raw data conversion

If your camera has a raw data mode, it's almost certainly best to use it, even though it creates additional post-photography work. If you allow the camera to convert to JPEG, it will use its own idea of exposure setting, and you'll get exposure variation from frame to frame. In addition, the JPEG format is not capable of carrying the full dynamic range of a camera sensor, so if the light level is low you'll get details lost in the shadows which won't be recoverable later.

The ufraw utility for Linux can do batch raw conversion for many camera models, and it has a batch mode, so you can convert a whole folder-full of images in one operation. The batch mode uses settings derived from the graphical user interface, so you can adjust the contrast and exposure interactively, and then apply the same settings to the whole batch.

Cropping

The ImageMagic utility has a tool called mogrify that can batch process a folder-full of images in one operation. But be aware that this process is destructive -- the images are converted and stored in place, so you'll need to take backups if you think you might change your mind.

To crop a region 640x480 pixels out of each frame, with the top left corner of the crop region 200 pixels across and 100 pixels down the frame, do this: 
% mogrify -crop 640x480+201+100 *.jpg
This tool won't tell you the region you need to clip. You could get this imformation from the Gimp image editor, by selecting the cropping tool and making a note of the coordinates it displays when you select the crop region.

Exposure adjustment, etc

The ImageMagick utility can also batch-adjust a whole range of image settings. For example, to increase the brighness by 10%, use: 
% mogrify -modulate 120 *.jpg

Assembling the images into a movie clip

This bit is easy, using mencoder. 
% mencoder mf://*.jpg -ovc xvid -xvidencopts bitrate=2000 -fps 25 -o clip.avi
This assembles all the JPEG images in the current directory into an AVI file, using XVID encoding at a bitrate of 2000 kbps, 25 frames per second. It goes without saying that the filenames of the images have to be in alphanumeric order for this to work. If not, you'll just get a jumble. Most cameras store images with sequential filenames, so it's not usually a problem.

Of course, you can encode into different formats than XVID, and experiment with the frame rate and bitrate to get the best results.

The results

Click here for a short clip from the Lego movie I made with my son one rainy Sunday afternoon. Sorry it's a low bandwidth encoding, but I have to pay for my Web hosting. We're not going to be competing with George Lucas any time soon, but with a bit of patience the results you can get aren't bad, for something made with free software and a consumer digital camera.

What next?

In this article I've described a simple process for creating an animated movie clip using Linux utilities and a camera. But building a whole movie involves much more than this. Most obviously, you'll need to cut many clips together and edit to make a proper cinematic narrative. You'll probably also want to add music and sound effects. There are Linux tools for this -- the best established is probably Cinelerra -- but that's for a different article.

Another interesting technique is to create visual effects by drawing directly onto each frame using a tool like Gimp, before assembling the frames into clips. This can be used to create, for example, laser beams and explosions, as well as more peaceable effects. This process can be very time consuming, and isn't easy to automate other than by writing specific software.

It's also even possible to use blue-screen techniques to replace the backgrounds of images, so you can make your animation take place against a different background -- a drawing, perhaps, or a photograph of a real scene. As in studio blue-screen, the trick is to photograph the action against a background of uniform (or near uniform) colour and brightness. This is typically bright blue or bright green. Then you can use software (for example ImageMagick again) to remove the background pixels leaving a transparent area, and layer the modified image on top of a background of your choice. However, to make this work using table-top techniques requires very careful attention to lighting -- if the lighting isn't even it becomes very hard to identify the background among the image pixels.