What happens between the frames?

Critical flicker-fusion frequency (CFF) is the frequency at which brightness fluctuations become indistinguishable from steady light for specified viewing conditions. Frequency in hertz (Hz) means cycles per second. CFF varies with brightness, modulation, area, and retinal location.

The stationary patch flashes in real time with equal light and dark intervals. The requested frequency is sampled by your display and browser, so this is not a calibrated CFF test. The available frequency range is limited by the observed update rate; the slider snaps to whole-frame light/dark intervals and shows their estimated frequency.

Successive still pictures can evoke apparent motion, sometimes called motion fusion. Frame rate (fps) counts new pictures per second. Seeing a moving character does not mean every position jump has become invisible.

10 fps is not a universal threshold. Motion depends on the pictures, their spacing, and timing. Compare rates rather than treating one number as a biological switch.

At low frame rates, a moving object jumps farther between positions. These steps can remain visible even when you recognize one moving object.

Angular step = object speed ÷ source fps. Faster objects require more motion samples to keep the jumps small. We call this frame stepping here, and distinguish it from the cadence mismatch and late frames in sections 9 and 10.

Smooth pursuit helps keep a continuously moving target sharp by keeping its image approximately stationary on the retina. On a display, the target holds each sampled position while your eyes keep moving. This mismatch adds blur from frame holding, even with instantaneous pixels.

Estimated smear = tracking speed × frame-hold time. Each frame is illuminated for its full interval here. Increasing FPS shortens that hold. Smooth pursuit is assumed; real tracking at low frame rates can include fixations and corrective saccades.

Some modern panels switch quickly enough that pixel-response ghosting is small. This example lets you explore slower switching: pixels at the old ball position fade while pixels at its new position brighten. Both positions can briefly remain visible: pixel-response ghosting. The 0 ms setting is an idealized fast-switching reference; the slow settings do not represent every current XR display.

The selected response time describes 10–90% brightening and 90–10% fading. This model uses the same speed in both directions; real panels can differ. A refresh can request a new value before a pixel finishes its previous transition. Slow response does not automatically prevent a high refresh rate.

The dashed marker identifies one pixel. Its graph compares requested brightness with modeled light output as the ball crosses it.

Section 05 changed how quickly pixels reached their requested brightness. Here, restore instantaneous switching and keep 60 fps / 60 Hz fixed. Now change how long each frame emits light. Lighting the frame for the full interval can still produce tracking smear; a brief light pulse followed by a dark interval reduces that smear.

Duty cycle = light-on time ÷ refresh interval. Here, display persistence is the light-on time per presentation. At 60 Hz, an interval lasts 16.7 ms: 100% duty lights the whole interval; 25% lights it for about 4.2 ms.

Shorter light pulses reduce the additional retinal smear during smooth tracking. They do not add motion positions or require slower pixels. Emission can be gated using a backlight or the panel’s own emission controls.

At a fixed refresh rate, lower persistence means a shorter light pulse per frame. If its luminance stays the same, the shorter pulse delivers less light per frame and the average luminance falls. To keep the same average luminance, the pulse must be brighter: the same amount of light must be delivered in less time. This requires higher peak luminance, not a higher average luminance.

Imagine a light emitting at 100 nits. If it stays on throughout a 10 ms interval, its average is 100 nits. At 20% duty, it emits for only 2 ms and is dark for 8 ms. The average is therefore (100 × 2 + 0 × 8) ÷ 10 = 20 nits.

To recover the original 100-nit average while still emitting for only 2 ms, the pulse must reach 500 nits: (500 × 2 + 0 × 8) ÷ 10 = 100 nits. Five times the luminance for one-fifth of the time delivers the same time-average light.

Average luminance = pulse luminance × duty cycle. Luminance is measured in nits (cd/m²). Lux instead describes light arriving at a surface.

Shortening emission can save energy when the peak stays unchanged, but restoring the average brightness can offset those savings. Hardware limits constrain compensation.

Compare 60 Hz and 120 Hz at 20% duty. At 60 Hz, each light pulse lasts about 3.3 ms. At 120 Hz, each lasts about 1.7 ms, but there are twice as many pulses per second. Both emit light for 20% of each second, so the same peak luminance gives the same average luminance.

The shorter pulses reduce tracking smear, while their higher repetition rate can make flicker less visible. This reduces persistence without the loss of average light caused by shortening pulses at a fixed refresh rate. In this widget, FPS follows refresh rate, so 120 Hz also means rendering 120 new frames per second.

The previous lesson showed the blur–flicker–brightness tradeoff. More frequent pulses can reduce visible flicker, and 24 fps does not require 24 light pulses/s. Repeating each frame twice or three times gives 48 or 72 pulses/s while retaining 24 motion positions/s. Dark intervals remain; the light–dark cycles simply occur more frequently.

More frequent pulses can reduce flicker, but do not create new poses or guarantee smooth motion. Each source frame receives the same number of equal light pulses; motion steps can remain. These are idealized global-emission examples.

Evenly captured frames can be displayed for unequal durations. 24 fps on 60 Hz alternates groups of three and two refreshes per source frame: a 3:2 cadence.

“Judder” and “stutter” are used with overlapping meanings. Here we name the cause: cadence judder means unequal repeats caused by rate conversion; late-frame stutter means a missed frame deadline. These are not separate mechanisms of vision.

A late frame leaves the old position on screen too long. When a fresh frame arrives, the object catches up with a larger position change.

The model introduces one rendering stall every 0.6 simulated seconds (every 6 seconds of playback at 0.1×). During the stall, source arrivals stop; display refreshes and illumination continue using the previous frame.

Frame pacing means keeping presentation intervals consistent. A high average fps can hide occasional long intervals.

Solutions include reducing rendering workload, preparing assets and shaders ahead of time, and scheduling frames consistently. A higher refresh setting alone cannot make a late frame arrive on time.

A briefly presented stimulus can remain perceptually visible after it disappears. This is visible persistence; it has no single fixed duration for all conditions.

Persistence can keep a stimulus briefly visible, but motion perception also depends on where and when the image changes. Motion-sensitive circuits respond differently to left-to-right and right-to-left sequences. Their activity helps encode direction and speed; a lingering image alone does not explain that selectivity.

In a flipbook, each page contains a frozen pose. The character's position or pose changes between pages, and the sequence can produce apparent motion.

The faded marks are an adjustable illustration of lingering traces. Use them to compare the sequence of positions with the traces left behind. This demo illustrates timing and spacing; it does not independently demonstrate why persistence alone is insufficient or predict what you will perceive.

A headset must measure motion, render an image, send pixel values to the panel, and emit light. These operations take time—and work for different frames can overlap.

Every row shares the same 0–120 ms time axis. The Scene frames ready row identifies completed frames; Display refreshes shows which frame each presentation uses. Each frame keeps its color through the operation rows. Saturated stages belong to your selected refresh; faded stages belong to other work. The moving black line only indicates the time being inspected.

Writing pixels and emitting light are different operations. Transfer/scan loads pixel values, often row by row. With rolling emission, different rows emit at different times. With global emission, the panel emits together after the data is loaded. This timeline uses a global-emission example.

An IMU is an inertial measurement unit: a motion sensor. The pipeline queries an early pose for rendering and a late pose for reprojection, correcting the viewpoint without rendering the whole scene again. These are uses of the tracking system’s pose estimate, not two separate IMUs or its only two measurements.

Motion-to-photon

Head movement → the corresponding change in display light. The bracket here starts at the pose timestamp and ends at first light. This pose age is not a full physical motion-to-photon measurement: sensing delay and pose prediction are omitted.

Interaction latency

A physical action → its visible result. Our controller-input marker starts later, when the application samples the input at render start; device and input-delivery delays are omitted. Reprojection refreshes head pose, not the action’s effect on the scene.

Display latency

Display input → emitted light. Our bracket starts at pixel transfer and ends at the first global light pulse.

90 Hz means 11.1 ms between refreshes. It does not specify the total delay from motion or interaction to light.

Photon-to-photon is a different measurement for camera passthrough: light reaches a camera → the corresponding image emits light from the display. Camera exposure, readout, and processing add delay; they are not simulated here.

Pipeline adapted conceptually from Hoffman & Lee (2019). Passthrough terminology: Antmicro.

The scene is stationary in the world. As the head turns, its image must shift across the headset display. A frame based on an old head pose puts it in the wrong place.

Reprojection reuses the rendered pixels and remaps their positions using a newer head measurement. For head rotation, this can avoid recalculating geometry, lighting, and shadows.

It still takes time. Here, the corrected pose is sampled 1 ms before the latest refresh—a chosen teaching value, not a hardware benchmark. The source frame also has a configurable render delay.

The three rows compare the continuous target, the old rendered view, and the reprojected view. Reprojection helps head-motion stability but cannot perfectly reveal unseen content or update independently moving objects.

A moving object can be sharp but jumpy, smooth but smeared, or flickering. Describe the visible problem first, then inspect its timing.