How do still pictures become a moving animation?

Make a picture

The fox is an artist-created 3D character. Its rig moves the surface into different poses. Choose a pose, place it on the stage, and capture it into your filmstrip.

Give pictures time

Each captured picture has a hold. Our studio counts time in 24 ticks per second. Six ticks take one quarter of a second. A longer hold adds time while keeping that picture still.

Choose what happens between

Held pictures change at the next capture. Linear in-betweens compute new poses along a chosen transition. Ease-in/out changes the progress through that transition. The last picture keeps its hold; the loop switch adds an explicit jump back to the start.

Watch, then inspect

Your visual system processes changes across space and time. A sequence can appear to move even when you can also notice its steps. Pause and select a picture to see exactly what was authored.

The name: apparent motion

A presentation of changing images can produce perceived motion without the corresponding continuously moving object on the screen. Visual responses do persist over time, but an afterimage alone is not a complete account of directional motion. Adelson and Bergen’s 1985 model describes motion processing through patterns across space and time; our studio does not simulate a brain.

Timing and spacing measure different choices

Duration equals total hold ticks divided by 24. Between two pose onsets, represented interval rate equals position change divided by elapsed time. A gap of 20 stage units over 0.25 seconds is 80 units per second; doubling that gap doubles the interval rate. Stage units are authored coordinates, not a fox’s real-world speed.

An in-between follows a rule

For scalar placement, linear interpolation gives x = (1−u)x₀ + ux₁. Ease-in/out uses e = 3u²−2u³ in place of u. Rig rotations use normalized quaternion spherical interpolation along a short arc, following glTF rotation semantics. These rules create poses; they do not recover a uniquely true missing observation or enforce foot contact.

Output frames are not all new pictures

Eight pictures held for six ticks each make a 48-tick, two-second project. Exporting it at 24 frames per second gives 48 output frames. At 48 frames per second it gives 96. In hold mode those extra frames repeat the same captured states; interpolation is a separate choice.

No universal frame rate for seeing

Watson, Ahumada and Farrell’s 1986 experiment asked observers to distinguish sampled motion from a 1920 Hz reference under controlled display and viewing conditions. Its 75%-correct discrimination threshold depended on speed, detail and conditions. Experiment 1 included two observers; it does not prescribe a frame rate for every child, eye or phone. The authors distinguish their visibility theory from a theory of motion sensing.

A repeated pattern can hide a turn

Twelve identical spokes repeat their arrangement every 30 degrees. At eight samples per second, a model wheel turning 240 degrees per second advances exactly 30 degrees per sample: its unmarked frames match a stationary wheel. A marked spoke adds identity information. Extra full turns can still hide between samples. This mathematical ambiguity differs from perceptual reversals under continuous input studied by Kline and Eagleman in 2008.

Pictures are historical evidence

Muybridge’s 1878 Sallie Gardner plate records different horse positions. Its twelfth panel shows the horse standing. The printed caption claims about 1/25 second between negatives and 1/2000 second exposures: interval and exposure are different quantities. A scan does not establish an exact seamless movie, and the standing reference should not be silently added to a gallop loop.

Stop-motion and flipbooks

Artists change a subject or drawing between pictures. Repeating a picture adds a hold; drawing or capturing another state adds information.

Character animation

Artists choose key poses and adjust timing. Computer-generated in-betweens can help connect them, but the result still needs judgment about contact, staging and the intended action.

Games and interfaces

A character or button can change through authored states while a display refreshes at its own rate. Interpolation and playback are explicit software choices, not guarantees of a perceptually perfect result.

Eight distinct pictures each hold for six ticks at 24 ticks per second. What is the project?

• Eight pictures, 48 ticks, two seconds.
• 48 different poses.
• Eight seconds.

Pose onsets stay 0.25 seconds apart. The position gap changes from 20 to 40 stage units. What happens?

• The interval rate stays the same.
• It changes from 80 to 160 stage units per second.
• This proves the fox is twice as heavy.

Add 12 ticks to one picture’s hold. Leave every other setting unchanged. What follows?

• The clip becomes half a second longer.
• New poses appear automatically.
• All the other poses speed up.

Export every held output frame twice at twice the output frame rate. What changed?

• The duration doubled.
• Every extra output frame contains a new pose.
• There are more output frames, with the same poses and duration.

What is a computer-generated pose between your selected keys?

• The only possible real movement.
• A new pose produced by a chosen interpolation rule.
• A missing photograph recovered from the model.

Which explanation fits the evidence about seeing motion?

• Every eye stores whole pictures for exactly 1/24 second.
• Visual responses persist, and motion processing uses changes across space and time.
• Noticing steps means someone has poor eyesight.

A 12-spoke wheel turns 30 degrees between samples. Its unmarked frames can match what?

• Both that rotating wheel and a stationary wheel.
• Proof that the model wheel stopped.
• A measurement of the viewer’s brain frame rate.

Why not loop all 12 Muybridge panels and label it an exact gallop movie?

• Old photos cannot teach us anything.
• Reordering makes historical timing exact.
• The twelfth horse is standing, and an exact seamless sequence is not established.

Animation channels and rotation interpolation have defined data semantics.

Sections on animations and Appendix C. LINEAR rotation uses spherical interpolation. Playback policy belongs to the application.

A real reusable rigged character with Survey, Walk and Run clips.

PixelMannen model CC0; tomkranis rig and animation, AsoboStudio and scurest conversion CC BY 4.0. Original 162,852-byte GLB is hash-verified. Learner/Brytalearn selections, placement and timing are modifications.

Key poses and computed in-betweens are distinct parts of production.

Source-owned explanation including Tony DeRose. No Disney/Pixar artwork or footage is redistributed.

Timing, staging and pose choices are tools of the animator.

Primary practitioner account; artistic principles are not universal physical or psychological laws. No paper artwork is copied.

Motion processing can be modeled using structure across space and time.

Original research, used for a narrow conceptual explanation. This app is not a neural simulation.

Detecting differences in sampled motion depends on stimulus and viewing conditions.

Original JOSA A paper, DOI 10.1364/JOSAA.3.000300. Experiment 1 used two observers and a 1920 Hz reference. Its discrimination threshold is not a universal onset of motion perception.

A continuous-input illusion does not establish a single visual sampling clock.

Original experiments challenge that proposed explanation. Deliberate digital sampling in this lesson is a different mechanism.

Successive horse positions and the standing twelfth panel belong to one historical artifact.

Eadweard Muybridge, Sallie Gardner, June 19, 1878. LC-DIG-ppmsca-06607. Source rights statement: No known restrictions on publication. Full plate resized/WebP encoded; separate caption crop is identified.

Browser rendering opportunities and authored timeline states are distinct.

Playback derives its position from elapsed time; missed callbacks do not permanently slow the project.