Persistence of vision: what it is, why it's a myth, and what really makes animation move

Menu names below are checked against the Blender 5.2 manual and source code.

What is persistence of vision?

Persistence of vision is the claim that the eye holds on to each image for a moment after it is gone. Joseph and Barbara Anderson, two film scholars who traced the history of the idea, summed up the textbook version this way: when the eye sees a rapid succession of slightly different images, each one "persists upon the retina, allowing that image to blend smoothly with the next image."

The idea grew up with optical toys. In an 1827 book, the physician John Ayrton Paris described the thaumatrope, a pasteboard disc with a picture on each side, spun between two strings, as "founded upon the well-known optical principle, that an impression, made on the retina of the eye, lasts for a short interval, after the object which produced it has been withdrawn." His example had a rat on one side and a cage on the other: spin the disc and the rat sits in the cage.

Roget, Plateau and the origin story of cinema

Film histories often credit the discovery to the English-Swiss physician Peter Mark Roget, for a paper read on December 9, 1824. When the Andersons went back to it, they found that Roget was describing the spokes of a wheel seen through a row of vertical slits, which look curved. That is a wheel in real motion looking distorted, not still pictures appearing to move. French writing on film often credits the Belgian physicist Joseph Plateau instead, whose phenakistiscope (1832) animated drawings on a spinning disc viewed through slits. Plateau's own statement of the principle reads, in the Andersons' translation: "the impressions they make upon the retina will join together without being confused, and one will believe he is seeing a single object gradually changing form and position."

The part that is true

Vision does persist, a little. In a 1980 paper titled "The persistences of vision," psychologist Max Coltheart noted that observers keep experiencing a stimulus for some time after it ends, and separated three kinds of persistence: what you see, what neurons keep doing, and what visual memory keeps. That is why a light whirled in a circle draws a ring, the very example Roget used, and why the rat lands in its cage. But look at what those examples produce: one combined shape. Not movement.

Why persistence of vision doesn't explain motion

The Andersons' 1993 article, "The Myth of Persistence of Vision Revisited" (Journal of Film and Video), followed a 1978 paper written with Barbara Fisher. Its key move is to split one question into two that early writers kept mixing up. Why does the picture look continuous instead of like a string of flashes? That is flicker fusion. Why do the figures on screen appear to move? That is apparent motion, and persistence of vision has no answer for it.

• Blending gives you superimposition. Images fusing on the retina would stack up, like Marcel Duchamp's Nude Descending a Staircase or a frame from Norman McLaren's Pas de Deux: in the Andersons' words, "a static collage of superimposed still pictures, not an illusion of motion."
• Motion is computed behind the eye. In 1912, Max Wertheimer ran experiments that refuted the afterimage theory and concluded that "we must have recourse to processes which lie behind the retina." After 1900, according to the Andersons, psychology treated movement almost without exception as a central process, handled by the brain.
• The brain can lose motion on its own. In 1983, Josef Zihl and colleagues described a patient with damage to the lateral temporo-occipital cortex on both sides. Her visual field was otherwise intact, yet she had no impression of movement in depth and did not see apparent motion either. Movement vision turned out to be a separate function, relying on brain areas beyond the primary visual cortex.

Hugo Münsterberg had already called the afterimage account "far too simple" in his 1916 book The Photoplay. The Andersons argue it survived as the creation myth of film studies, with a "passive viewer upon whose sluggish retina images pile up."

Phi phenomenon vs. beta movement

Wertheimer's 1912 monograph, "Experimentelle Studien über das Sehen von Bewegung" (experimental studies on the seeing of motion), is usually cited as the founding text of Gestalt psychology. He flashed two separate lines one after the other and varied the gap in time between them. The Andersons list three outcomes:

• Beta movement: the line appears to travel from the first position to the second. Steinman and colleagues call it optimal apparent movement, the kind that resembles real movement.
• Partial movement: each line seems to move a short distance.
• Phi phenomenon: pure, objectless motion. The lines look stationary as they flash on and off, yet movement is felt between them.

Wertheimer never described phi very clearly, which caused lasting confusion, as Robert Steinman, Zygmunt Pizlo and Filip Pizlo explain in a 2000 review with a blunt title: "Phi is not beta." Some film scholars simply swapped "phi phenomenon" in for persistence of vision without understanding it any better. But a character on screen does not stay put while a ghostly motion passes beside it. It moves, which looks like beta, not phi, and the Andersons file film motion under short-range apparent motion, covered next.

Short-range apparent motion

In 1974, Oliver Braddick described a "short-range process" in apparent motion using random-dot patterns: people could report the direction of motion only when the dots shifted less than an upper spatial limit, now written dmax. The Andersons concluded that film, where changes from frame to frame are small, falls in this short-range category, and that "to the visual system the motion in a motion picture is real motion." Not everyone accepts two separate processes: in 1989, Patrick Cavanagh and George Mather argued for a single kind of motion analysis followed by a more general inference stage. Either way, motion comes from mechanisms that detect change from one image to the next, not from a sluggish retina.

Flicker fusion: the other half of the illusion

The critical flicker fusion frequency is the rate at which a flickering light is perceived as continuous. It has no fixed value. A 2021 review in the journal Medicina lists the amplitude of the flicker, average brightness, the position on the retina, color, ambient light, viewing distance and stimulus size, plus age and fatigue. The same review puts human flicker detection at 50 to 90 Hz, with some reports of people distinguishing modulated from steady light up to 500 Hz.

Film projectors deal with flicker, not motion, through the shutter. F. H. Richardson's 1927 handbook of projection put it plainly: "What we call 'flicker' is the visibility of the period during which the screen is dark." The main blade blocks the light while the film advances; a single blade would leave the dark periods too far apart, so the screen goes dark two or three times per frame, depending on whether the shutter has two or three blades. Richardson adds that flicker gets worse as the picture gets brighter and larger. At 24 frames per second, a two-blade shutter gives 48 flashes per second and a three-blade shutter 72, while the audience still sees only 24 different images.

Persistence of vision examples: from optical toys to film

October 28, the date of Reynaud's first shows, has been marked as International Animation Day since 2002.

Why 24 frames per second

24 fps is an engineering standard, not a perceptual limit. At a 1929 meeting of the Society of Motion Picture Engineers, one speaker pointed out that the 16 pictures per second in the silent-film standard was "not very rigidly in practice," while "in sound film, 24 pictures is rigidly definite and susceptible of standardization." Encyclopaedia Britannica dates the switch to 24 fps, forced by sound, to 1927. It is still the frame rate of a new Blender scene.

What this means when you animate in Blender

If motion comes from comparing one image with the next, an animator really controls the change between frames: the frame rate, the number of distinct poses, how far apart they sit and what each frame contains. Keyframes and the Graph Editor are covered in the guide to animating in Blender.

Setting the frame rate

The frame rate lives in Output Properties > Format > Frame Rate, "the number of frames that are displayed per second," as the manual puts it. The menu runs from 6 to 240 fps, including 23.98, 24, 25, 29.97, 30 and 60, plus Custom, which shows FPS and Base. The actual rate is FPS divided by Base, for standards such as NTSC that need a more precise value: the 23.98 preset is 24 divided by 1.001. New scenes start at 24 fps.

Keys are stored on frame numbers, so changing the frame rate after animating speeds up or slows down the whole shot. The Time Stretching subpanel of Frame Range remaps an animation's length with its Old and New values, but leaves the start and end frames for you to fix. Frame rate also drives render time: one minute is 1,440 frames at 24 fps and 1,800 at 30, a quarter more to render. The steps for output are in the guide to rendering an animation in Blender.

Animating on ones and twos

3D animation runs on ones: Blender evaluates every curve on every frame. Hand-drawn animation often works on twos, holding each pose for two frames. The playback rate stays the same; only the number of distinct poses drops.

The Stepped Interpolation F-Curve modifier gets you there without touching your keys. The manual says it "lowers the curve's frame rate by letting it change its value less frequently, producing choppy movement as a result." In the Graph Editor, select the curves and choose Channel > Add F-Curve Modifier > Stepped Interpolation. Set Step Size (2 by default), use Offset to line the steps up with your poses, and Start Frame and End Frame to limit the effect. If you key every pose by hand, Constant interpolation (Key > Interpolation Mode, T) holds each value until the next key. The stop motion guide covers both methods and how to render only every other frame.

On twos, each change is twice as big as on ones for the same speed. This is where perception research meets practice: the bigger the jump between frames, the more a move risks reading as a series of hops. Animators close that gap three ways: in-betweens, motion blur and smears. Spacing and breakdowns are explained in the guide to in-betweening in Blender.

Motion blur

By default, Blender renders an animation as a sequence of perfectly still images. The manual calls that "great for stop-motion and time-lapses" but unrealistic, since fast-moving objects look blurred along their direction of travel, in a movie frame as in a photo. Turn it on with the checkbox in the header of Render Properties > Motion Blur, in Cycles or EEVEE.

• Position: Start on Frame (the shutter opens at the current frame), Center on Frame (open during it, the default) or End on Frame (it closes at the current frame).
• Shutter: how long the shutter stays open, in frames. At 1.0 the blur spans a whole frame; the default, 0.5, spans half a frame.
• In Cycles, each object has its own motion blur settings in the Object tab. There is no camera motion blur with an orthographic camera, lights are treated as stationary during the frame, and particles and physics should be baked before rendering.
• In EEVEE, the frame is split into time Steps that are accumulated, with a post-process blur filling the gaps. Each step re-evaluates the whole scene and can add a lot of render time. In the viewport, blur only shows during playback, with a simpler algorithm.

Blur is paid for on every frame of an animation; other ways to claw that time back are in the guide to rendering faster in Cycles.

Smear frames

Hand-drawn animation has its own answer to moves that are too fast: the smear, a frame where the character is stretched or multiplied along its path with clean outlines. It bridges the gap between two positions the way blur does, in a graphic style. Blender has no smear tool, but the guide to smear frames in Blender shows four ways to build them.

Judging timing at real speed

When viewport playback cannot keep up, the frame counter turns red and the Sync setting in the Timeline's Playback menu decides what happens: Play Every Frame shows them all even if playback slows down, Frame Dropping skips frames to hold the rate. For a faithful preview, 3D Viewport > View > Render Playblast renders every frame with the viewport renderer, in Workbench or EEVEE. For the bigger picture of a production, see the guide to 3D animation.