Methodology
Every value each video is generated against, published in full — for parents who like to know, for professionals who want to check it, and for anyone else making things like this who might find it useful.
Last updated 9 September 2026
01
The luminance ratio between the lightest and darkest elements in a frame.
Newborn contrast sensitivity is a fraction of an adult's, which is why bold black-and-white holds attention when pastels do not.
We render ink at sRGB 10,10,10 and paper at 250,250,250. That gives a WCAG contrast ratio of 18.96:1 and a Michelson contrast of 0.9937, against 21:1 and 1.0000 for absolute black on absolute white.
Stopping just short of pure values is deliberate. Absolute black leaves the H.264 encoder no headroom below the signal floor, and hard pure-white edges ring visibly after compression. Giving up 2 points of contrast ratio buys a cleaner image at the bitrate YouTube actually delivers — and at 18.96:1 we are still far beyond any contrast an infant can resolve the limit of.
02
The measure that shifts most with your screen and your room, so it is worth setting out properly.
How fine or coarse the pattern is, in cycles per degree of visual angle.
There is a sweet spot. Too fine and the pattern averages out into flat grey. Too coarse and it reads clearly but there is not much there to hold attention.
Spatial frequency depends on your screen and how far away your baby is, so rather than pick one number we have worked out several. Our repeating patterns span 0.08 to 0.35 cycles per degree across realistic setups — see the viewing table below. The finest element we ever draw is a 14 px ring stroke, about 3.4 cycles per degree on a phone.
03
Degrees of visual angle per second.
Smooth pursuit — following a moving object with the eyes rather than in jumps — develops over the first months, so early videos have to move slowly enough to be trackable.
Measured across every video in the catalogue, peak on-screen speed is 797 px/s at 0–3 months, 959 px/s at 3–6 months and 960 px/s at 6–12 months. On a phone at arm's length that is roughly 8, 10 and 10 degrees per second; on a tablet nearer 11 to 13.
The more telling number is the median, which for most 0–3 month videos is zero. Most of the time the screen is simply resting, and movement arrives as something to notice.
04
Full-field brightness changes per second.
A safety limit before it is a design one: rapid full-field luminance change is the property associated with photosensitive seizure risk.
We render every frame at 1920×1080, 30 fps, and measured mean-frame relative luminance across the entire catalogue. Counting discrete transitions that shift luminance by more than 10% within a third of a second, the worst video in the catalogue produces 0.56 per second. Most produce under 0.15. The generally accepted threshold is 3 per second.
The highest readings come from the peekaboo videos, where a character hides and reappears. Even those sit more than five times under the limit.
05
How long one pattern holds before it changes.
Babies settle into a pattern and then gradually lose interest in it. That whole arc is the part worth having, so a scene is given time to run its course rather than being cut short.
Nothing is keyframed by hand. Every video is a tempo plus a list of sections, and each section owns a whole number of musical bars, so scene changes always land on a downbeat. Sections run 8 to 40 bars at 48–116 BPM depending on age, which puts a single scene at roughly 20 to 90 seconds for a newborn video and 15 to 40 seconds at the oldest tier.
There are no cross-fades. A gentle fade is close to invisible at newborn acuity, so transitions are either a clean cut on the downbeat or a full swap of foreground and background.
All of it
Measured across all 39 videos currently in the catalogue.
| Measure | 0–3 months | 3–6 months | 6–12 months |
|---|---|---|---|
| Videos | 21 | 6 | 12 |
| Contrast ratio | 18.96:1 | 18.96:1 | 18.96:1 |
| Palette | Ink + paper | + red | + yellow, blue |
| Tempo | 48–72 BPM | 62–88 BPM | 70–116 BPM |
| Peak motion | 797 px/s | 959 px/s | 960 px/s |
| Peak motion (phone) | 8.2°/s | 9.9°/s | 9.9°/s |
| Median motion | 0 px/s | 0 px/s | 216 px/s |
| Runtime | 4–45 min | 4–25 min | 4–30 min |
| Flash events | ≤0.16/s | ≤0.38/s | ≤0.56/s |
Resolution 1920×1080 at 30 fps throughout. Audio is mastered to −20 LUFS, deliberately quieter than the −14 streaming standard, because platforms turn loud content down but never turn quiet content up.
Screens and distance
Here is the arithmetic for three setups families actually use.
Spatial frequency only means something once you fix a screen size and a viewing distance.
| Setup | Screen width | Distance | Field of view | Pixels per degree |
|---|---|---|---|---|
| Phone, arm's length | 14 cm | 40 cm | 19.9° | 97 |
| Tablet, propped up | 23 cm | 50 cm | 25.9° | 74 |
| TV, across the room | 95 cm | 2.5 m | 21.5° | 89 |
The three land within a narrow band, 74 to 97 pixels per degree, which is why the same patterns hold up on all of them. Translated into what a baby's visual system actually sees:
| Pattern | Phone | Tablet | TV |
|---|---|---|---|
| Stripes, 7 bars | 0.35 c/deg | 0.27 c/deg | 0.33 c/deg |
| Checkerboard 8×5 | 0.20 c/deg | 0.15 c/deg | 0.19 c/deg |
| Checkerboard 4×3 | 0.10 c/deg | 0.08 c/deg | 0.09 c/deg |
| Main dot, angular size | 4.2° | 5.5° | 4.6° |
Newborn contrast sensitivity peaks at very low spatial frequencies and falls away quickly above roughly 1 cycle per degree in the first months. Our patterns sit at 0.08–0.35, comfortably inside that window.
Under the hood
The specification is the code itself, rather than a document the code is meant to follow.
Videos are not edited. Each one is defined as a tempo and a list of sections drawn from a shared library of motions, and rendered deterministically — the same definition always produces a byte-identical file. Ink, paper and the per-age palettes are single shared constants, so the contrast values stay the same everywhere by construction.
Music is generated separately and then measured before it is allowed in: tempo drift against the video's grid, downbeat accent strength, dynamic range, high-frequency content and lead-in silence. A track that fails is regenerated rather than corrected.
Evidence
Each figure traces back to peer-reviewed work, linked here so you can read it for yourself.
Norcia, Tyler & Hamer (1990), Development of contrast sensitivity in the human infant, Vision Research. Contrast threshold at low spatial frequencies falls from about 7% at 2–3 weeks to roughly 0.5% by 9 weeks. Grating acuity by sweep VEP rises from around 5 c/deg in the first month to 16.3 c/deg at 8 months, against an adult mean of 31.9.
Banks & Salapatek (1981), Infant pattern vision: a new approach based on the contrast sensitivity function, Journal of Experimental Child Psychology 31, 1–45, and Atkinson, Braddick & Moar (1977), Vision Research 17, 1037–1044. Both establish the band-pass shape of the infant contrast sensitivity function and its peak at low spatial frequencies, which is the basis for keeping our patterns between 0.08 and 0.35 c/deg.
Behavioural estimates run lower than VEP ones: Fantz and colleagues found discrimination of a 1.5 c/deg grating by one month and 3 c/deg by two. We design against the lower behavioural figures rather than the more generous electrophysiological ones.
Eye tracking in the newborn is largely saccadic; the capacity to track smoothly begins around 6–8 weeks and approaches adult quality by 4–5 months. This is why horizontal tracking motion only appears in the later videos.
Development of chromatic discrimination in early infancy (1995). One-month-olds discriminated red from green but neither from yellow; discrimination of red from yellow appeared only at three months. Our tier 2 introduces a single red at three months for exactly this reason, and holds yellow and blue back until the 6–12 month tier.
WCAG 2.3.1 "Three Flashes or Below Threshold", also published as ISO/IEC 40500:2012, sets the general limit at three flashes in any one-second period. Broadcast practice via the Harding analyser classifies a transition as an inversion when 80% or more of the screen changes brightness by 20% or more, and permits up to three such transitions per second.
Our own measurement is deliberately stricter: we count any change above 10% of mean-frame relative luminance within a third of a second, across the full frame, at 30 fps. On that harsher test the worst video in the catalogue reaches 0.56 per second. See International Guidelines for Photosensitive Epilepsy: Gap Analysis and Recommendations for how the various national standards compare.
None of this research says that watching a screen benefits an infant. It describes what the infant visual system can and cannot resolve. We use it to avoid showing a baby something they physically cannot see — not as evidence that they should be watching at all.
Being clear
These are visual stimulation, not an intervention.
We make no claim that watching them improves development, accelerates milestones, or produces any measurable outcome, because no such claim would be supportable.
Paediatric guidance is cautious about screens in the first two years, and faces, movement and physical high-contrast cards do things a screen cannot. This specification describes what is on the screen when a parent chooses one. It is not an argument that they should.