Scanimate
The letter above is not a font being scaled. It is a photograph of a photograph: a white letter on a light table, shot by one camera, drawn on a grey television tube, shot again by a second camera, and mixed back onto the same tube. Color is added last, to the second camera's picture. Every ring behind the letter is a previous frame.
That is how Scanimate worked, and this page is a working model of it. The homepage runs the same model as a background. Here the screen is the output and the panel under it is the control board. Everything on the board does something you can watch.
The machine
Scanimate was Lee Harrison III's analog animation computer, built by his company, Computer Image Corporation, in Denver. The earliest blueprints date to 1969, and eight were built. Dave Sieg, who maintained two of them at Image West in Hollywood and later restored the last working one, puts it this way: for about ten years, if you saw computer animation on television, it probably came from one of them. Network idents, the counting films on Sesame Street (Jim Henson flew to Denver in 1970 to animate the numbers 4 and 10), and the false-color Death Star in the first Star Wars were all Scanimate. Sieg's Scanimation in the Analog Days and the 1969 operator's manual are the sources for everything here about the hardware.
The process began with a light table: a backlit sheet with animation pegs, holding high-contrast artwork, usually a Kodalith of a logo. A monochrome camera scanned it. The camera's picture was then drawn on a five-inch precision CRT, but not with a normal television scan. The deflection signals that steer the tube's electron beam came from an analog computer: ramp generators, sine and cosine oscillators, bias and gain pots, multipliers, and summing amplifiers, patched together with cords on the front panel like an early modular synthesizer. Bend the deflection voltage and the logo bends. Add a sine wave to it and the logo swings. There is no geometry anywhere. The picture is warped by rewiring the scan that draws it.
The tube was grey. In video mode a second camera rescanned it at broadcast rate and fed a colorizer, which sliced the grey levels and assigned a color to each slice. That is where every color came from: a mapping from brightness to hue, applied after the fact.
Clients paid up to $2,500 an hour and left the same afternoon with finished tape. Nothing was repeatable. A job used a hundred or so patch cords and as many knob positions, and Sieg writes that a client coming back for one small change usually got an animation that barely resembled the one they had.
Stages
The model can be tapped at each stage, in signal order. The four monitors below are live while they are on screen.
Light table. The artwork: a white letter on a black field, drawn once and left alone. It never moves. If the letter appears to move on the monitor, that is the scan being warped, not the artwork.
Mono CRT. This frame's picture on the tube, in grey. It is a mix of three things: the light table as camera A sees it, warped by the computer; the tube's own afterglow, which is the previous frame at low level; and camera B's picture of the previous frame.
Cam B. The rescan camera's view of the tube, as the mixer receives it. Camera B is aimed at the CRT and zoomed in a few percent, so what it sends back is the last frame, slightly enlarged.
Colorizer. The grey tube mapped through a palette. This is the only stage with color, and it is the same picture as the monitor.
Look at Light table and Mono CRT together. Both are grey. The colorizer sits downstream of the tube, so color never enters the loop. Switch the palette on the board: every ring changes color at once and none of them changes shape. Sieg lists the mono tube among the machine's unfortunate shortcomings. Here it is the rule that keeps the picture honest.
Why the rings
The rings are video feedback: a camera pointed at a monitor that shows the camera's own output. Sieg describes the chain one way, light table to tube to colorizer, so closing the loop is this model's addition. But feedback was a studio technique of the same era, done with a spare camera and a mixer, and that is how it is done here. Camera B's picture goes into the mixer, and the mixer's output is what camera B is looking at.
Because camera B is zoomed in slightly, each pass through the loop draws the previous picture a little larger. The tube already holds last frame's mix, which already holds the frame before, so the rings pile outward until they fade below black. The letter at the center is the only thing drawn fresh. Pull the cam B fader to zero and the rings are gone in a few frames; there is nothing left to re-photograph. Turn zoom up and they spread faster. Nudge pan H and the whole stack leans away from the letter, because the camera moved and the artwork did not.
This is a different thing from zooming the letter. A zoom enlarges the letter and its rings together, as one image. Feedback enlarges only what the camera saw last time, so the letter stays put and the copies march away from it. The first version of this page did the zoom. It looked like a Photoshop effect because it was one.
The board
The panel follows the machine's own layout, in signal order: light table, computer, sequence, oscillators, rescan camera and mixer, colorizer. The names on the knobs are the names on the 1969 panel. Knobs are where the machine had knobs. Sliders are where it had sliders: the colorizer and the video switcher.
Computer
The manual calls the deflection panel the computer chassis. Depth is its word for size: the raster is drawn as if nearer or farther away. Horiz and vert are its position. These three come in two sets, initial and final, for the reason given in the next section. Width and length set the raster's proportions and there is one of each, as on the machine. In the model they all scale and translate the light-table image as it is drawn onto the tube. Bend is not on the 1969 panel; it shifts every other scanline sideways by a sine of its position, which is what an unstable horizontal time base does to a picture, and what animators got by patching an oscillator into the horizontal channel at scan rate.
Depth is clamped away from zero. Through zero the deflection reverses and the picture flips. On the machine that was a real effect. In an early version of this model it was a bug that flipped the letter every half cycle. Inv does the flip on purpose, as a switch.
The knobs do not jump. Each control voltage slews toward its knob position over a fraction of a second, the way an analog stage settles, so turning depth hard leaves a short trail of intermediate sizes in the loop instead of a cut.
Mouse is not on the machine. It adds the pointer's position to depth and width, to position, or to the two mixer faders, as an offset on top of the knobs.
Sequence
This is how Scanimate animated. The operator set up an initial image with one set of depth and position pots and a final image with the other, and looked at each by flipping a switch between them. Throwing the switch to FINAL started the sequence: a ramp generator ran from one voltage to the other and the deflection followed it, so the picture moved from where it was to where it was going. Everything else on the panel, oscillators included, ran on top of that. There were two rate pots, A and B, and a switch to pick one, and the manual says you could flip between them mid-run. Sieg describes a sequence button that could also be fired from timecode on the videotape, so the move landed on the frame the editor wanted.
The model does the same. Initial and Final are the switch; the bar under them is the ramp. Throw it and the letter travels between the two images at rate A or rate B. Throw it back and it returns. The default final image is the initial one at a larger depth, so the sequence is a zoom. Set the final horiz off to one side and it becomes a fly-in. Because the rings are camera B looking at the tube, every position the letter passes through stays in the loop for a moment, and a fast sequence leaves a trail.
Oscillators
The 1969 machine had three, one each for the horizontal, vertical, and depth channels, with an on–off switch, a frequency pot and an amplitude pot. Later machines let you patch them anywhere. This board has two, each with the same three controls and a row of jacks for where it goes. By default both drive width, at different rates, which is why the letter breathes. Patch one into bend and the picture wobbles; into horiz and it sways.
Each oscillator also had a ramp select switch, and so do these. Const ignores the sequence. Max→min multiplies the oscillator by the ramp turned upside down, so it swings hardest at the initial image and dies out as the sequence completes, which is how a logo wobbles into place and locks. Min→max is the reverse: still at first, shaking by the end.
Rescan and mixer
Phos is the tube's afterglow: how much of the last frame is still on the phosphor when the next one is drawn. Zoom, pan H and pan V are camera B's controls. The camera looks at the tube; it does not look at the artwork, so panning it moves the rings and not the letter.
The mixer is a video switcher with two faders. Cam A is the light-table camera, the letter itself. Cam B is the rescan camera, the feedback. They add. Sieg's studio had a switcher under the colorizer, used mostly to super a grid over the output for preview.
Light table
The four keys load a preset. The text field takes any short string in one of three faces. Outline strokes the letter; Kodalith fills it, after the high-contrast film studios used for logos. Mount puts a photo on the table. The photo is thresholded to pure white and clear, because that is what a Kodalith is and the camera has to see the table as on or off. Eject takes it off. Cycle walks the presets; Window wipes the artwork in from the left.
Nothing on the light table moves. Position, size, and zoom belong to the computer and to camera B.
Colorizer
Sieg's colorizer worked by slicing. Black sliders set the grey levels where one color ended and the next began; RGB sliders set the color of each slice; and each slice had a linear turn-on and turn-off so the edges did not fry. The model's colorizer is a table: 256 grey levels in, one color each out. The palette keys stand in for the RGB sliders. Phosphor is the first look this page had, yellow through vermilion and magenta to cobalt. Cobalt is the default. Fire and B/W are the other two.
The three slice sliders are the black ones. Each sets the grey level where a quarter of the palette begins, so pushing one up holds the color below it over more of the rings, and pulling it down brings the next color in sooner. At 25, 50, 75 the table is used as drawn.
References
- Dave Sieg, Scanimation in the Analog Days, scanimate.com, 1998. The history, the signal chain, and the working conditions.
- Dave Sieg, How Scanimation Was Done, scanimate.com. A tour of the machine in pieces: light table, CPU, patch panels, CRT rescan, colorizer. The colorizer page is the source for the slice sliders.
- Computer Image Corporation, Scanimate Manual, first revision, Denver, 1969. Contributed to the Internet Archive by Dave Sieg. Chapter III, table 3-1, lists the panel controls: the initial and final depth, horizontal and vertical pots, width and length, the INITIAL–FINAL switch, RATE A and B, and the three oscillators with their on–off, frequency, amplitude, and ramp select switches.
- Jim Henson's Red Book, 9/10–12/1970, The Jim Henson Company, 2014. Henson in Denver animating the numbers 4 and 10 for Sesame Street.