This guide sets out the evidence behind a calming scroll app that gives every short video its own color and tone. Each design choice is graded against the research, from the colors and pitches on screen to the pace of the feed and the rhythm that plays while a viewer lingers. Parts 1 to 6 build the rules, and Part 7 runs them in a working scroll lab.
People everywhere broadly agree on what colors mean; 4,598 people in 30 nations matched one global pattern at r = .881. The mood effects themselves are small, depend on context and are often confounded with brightness and saturation2. Several famous findings failed replication3,4,5. Lowering saturation matters more than choosing the right hue.
Seven steps that never repeat a hue category back to back: muted coral, soft gold, sage, muted teal, dusty blue, lavender-grey and misty blue-grey. Chroma falls at each step and lightness stays moderate. Each color holds for 60 to 90 seconds with a 10 to 20 second crossfade in OKLCH, paced to about 6 breaths per minute. Evening room lighting, and the app's night colors, stay dim and leave out blue and teal light.
No scientific color-to-note table exists, and the historical systems contradict each other. The best-supported calming sound is slow, quiet, low to middle in register and consonant6,7. It works best when tempo, register and loudness descend together and the piece ends in silence. A descending D-major pentatonic line over a D drone, slowing from 72 to 48 BPM, applies these rules.
Each claim is rated on four levels: strong, moderate, weak or unsupported. Ratings reflect sample sizes, replication status and whether effects survived correction for publication bias.
Color meaning is broadly shared across cultures. Color effects on the body are smaller, and they depend more on saturation and brightness than on hue.
The largest cross-cultural dataset is the International Color-Emotion Association Survey. Jonauskaite and colleagues tested 4,598 participants from 30 nations who spoke 22 native languages, asking them to link 20 emotion concepts from the Geneva Emotion Wheel to 12 color terms1. Each nation's pattern closely matched the global pattern, with a mean similarity of r = .881.
Nation still predicted associations beyond the global pattern. Nations that were closer in language or geography had more similar patterns, and a machine-learning classifier identified a participant's nation at 34.4% accuracy1. Red and black were rated the most emotional colors and brown the least. Pink was linked to love, yellow to joy and black to sadness1.
| Measure | Value |
|---|---|
| Participants | 4,598 |
| Nations | 30 |
| Native languages | 22 |
| Emotion concepts × color terms | 20 × 12 |
| Mean similarity, nation vs global | r = .88 |
| Classifier accuracy for nation | 34.4% |
These studies record what people think a color means. What a color does to the body is a separate question, and the two often diverge.
Valdez and Mehrabian used Munsell color chips and the Pleasure-Arousal-Dominance (PAD) model. Saturation and brightness had strong, consistent effects, summarized by the standardized regression weights in Fig 1.18. Brightness effects were nearly identical for chromatic and achromatic colors. Blue, blue-green, green, red-purple, purple and purple-blue were the most pleasant hues, and yellow and green-yellow the least pleasant8.
Wilms and Oberfeld controlled all three dimensions on an LED display. They showed 27 chromatic colors and 3 brightness-matched achromatic colors to 62 participants while recording skin conductance and heart rate. Saturated, bright colors produced higher self-rated arousal and stronger skin conductance responses, and arousal rose from blue and green to red2. The authors stress that many earlier studies failed to control brightness or saturation2.
A virtual-reality study of red and blue rooms complicates the brightness story. Darker rooms produced higher heart-rate arousal regardless of hue, the opposite direction from small color patches26. A separate study found cognitive performance and emotion indifferent to ambient color27.
Saturation (chroma) is the most reliable lever, because lower chroma lowers arousal. Keep brightness moderate. Very bright colors raise arousal in some paradigms, very dark immersive scenes raise it in others, and darkness lowers pleasure2,8,26. Hue is secondary. Cool hues are slightly less arousing than red at matched chroma and lightness2.
People like colors to the degree they like the objects associated with them, such as clear skies and clean water versus feces and rotting food. A weighted affective valence estimate explained 80% of the variance in average US color preferences (r = .893) and outperformed cone-contrast and color-appearance models28.
Color meaning depends on context. Red signals danger or failure in achievement settings and attraction in romantic ones29. The theory is coherent, but several of its flagship predictions have not replicated well (section 1.5).
These accounts point to red faces in anger, redness in sexual signaling, and blue sky and green vegetation as safe environments. Jonauskaite and colleagues read some shared pairings, such as a red face in anger, this way1. The accounts are plausible for a few associations and hard to test directly.
Language, symbolism and tradition shape color meaning, such as red as good fortune in China. Linguistic proximity predicts the similarity of national patterns, which supports a learned layer on top of shared tendencies1.
Physiological color effects are small and fragile. Luminance easily confounds them, because a red and a blue stimulus are rarely equally bright. Studies that equate luminance still find saturation and brightness effects on skin conductance, while hue effects on heart rate are inconsistent across studies2.
Colored light differs from colored surfaces. Light changes the illuminance at the eye and the pupil size, and short wavelengths also drive the melanopsin-based circadian system that regulates melatonin30. A surface or screen color changes only the visual scene.
Associations vary with nation and language1, with gender2, and with age and current mood. Among people with Northern European ancestry, red-green color vision deficiency affects about 1 in 12 males and 1 in 200 females31. A calming sequence should never rely on red-green contrast alone.
| Claim | Original evidence | What followed | Verdict |
|---|---|---|---|
| Red impairs intellectual performance | Early laboratory studies of achievement contexts | A meta-analysis of 67 effect sizes from 38 samples found early studies much larger than later ones, and no evidential value after correction for publication bias3. | Unsupported |
| Red increases attractiveness | Laboratory rating studies | Men rating women: d = 0.26 (95% CI 0.12 to 0.40, N = 2,961, I² = 89%). Women rating men: d = 0.13 (N = 2,739), with clear upward bias10. Student replications (n = 640) found no effect of red backgrounds on women's ratings of men11. | Weak |
| Red helps detail tasks; blue helps creative tasks | Mehta and Zhu, 69 participants12 | A direct replication with 263 participants found no color-by-task interaction4. | Weak |
| Green boosts creativity | Four experiments: a glimpse of green beat white, grey, red and blue13 | Ambient green versus red in the workspace showed no effect on creativity, t(126) = 1.2414. Ambient red and blue showed no effect either32. | Weak |
| Baker-Miller pink reduces aggression | Claims of reduced strength and aggression33 | 59 inmates randomly assigned to pink or grey cells showed no difference in aggression5. In a 46-student study, a pink room lowered self-reported anxiety but left grip strength and motor precision unchanged15. | Unsupported |
Each card pairs a typical saturated version of the color (top) with a calmer, lower-chroma version (bottom). Meters show relative tendencies at typical saturation and lightness; they shift as chroma and lightness change.
One constraint applies throughout. Each consecutive color comes from a different hue category. No controlled study has tested a multi-hue calming sequence, so the principles below extrapolate from the evidence on single colors, light and music.
Twelve volunteers completed the Montreal Imaging Stress Task and then lay under blue or white light while ECG and EEG were recorded. Under blue light, stress reached its minimum about 1.1 minutes into the block, compared with about 3.5 minutes under white light. The advantage disappeared after 3.5 to 5 minutes9. Color effects therefore look front-loaded and transient, which favors changing colors every minute or two.
Low chroma reduces arousal2,8, and very dark immersive scenes can raise it26. Slow, minor-mode, soft music is matched with darker, greyer and bluer colors, so a congruent audiovisual pairing for calm moves toward cooler, less saturated colors19.
Photosensitive epilepsy. WCAG 2.3.1 prohibits content that flashes more than three times in any one-second period unless it stays below the general and red flash thresholds34. Crossfades of 10 to 20 seconds sit far below flash rates. Never strobe or pulse saturated red.
Color vision deficiency. The sequences change lightness and chroma as well as hue, so they still read as a progressive change for people with red-green deficiency31.
Evening light. Short-wavelength light is the strongest driver of melatonin suppression30. Keep blue and cyan light out of the hours before sleep and use dim, long-wavelength light instead.
| Step | Category | Color | OKLCH (L, C, h) | Dwell + fade | Rationale |
|---|
The same logic with blue and cyan light removed and progressive dimming. Tiles fade toward the page background to show relative intensity; on-screen colors only approximate LED output. The app uses this sequence for its night colors (7.5).
No universal assignment exists. Historical systems disagree with each other, and the physics-based mapping has no perceptual basis. The reliable links are broader cross-modal tendencies shared by most listeners.
The systems contradict each other. C is violet for Newton, blue for Castel, and red for Scriabin and Rimington. Red falls on D for Newton, on G for Castel, and on C for Scriabin and Rimington.
Newton chose his seven spectral bands by musical analogy. He split the spectrum into segments proportional to the intervals of a Dorian scale, placing orange and indigo at the half steps39. Scriabin's assignments follow the circle of fifths (C red, G orange, D yellow, A green, E sky blue, B blue, F♯ bright blue), so his rainbow encodes key relationships rather than pitch height40.
From Concerning the Spiritual in Art (1911)41. His scheme sits closer to modern findings, with yellow high and bright and dark blue low, than any note table.
The method converts wavelength to frequency (f = c / λ) and halves that frequency 40 times, dividing by about 1.0995 × 1012, to land in the audible range.
The visible spectrum, about 400 to 700 nm, spans a frequency ratio of only 1.75, less than one octave, so the whole rainbow collapses into roughly G4 to F5. The choice of 40 octaves is arbitrary, hue perception is not linear in wavelength, and purples are not spectral at all.
The mapping also runs against perception. It puts red lowest and violet highest, while listeners match the lightest hue, yellow, with high pitch and dark blue with low pitch17,18.
| Color | λ (nm) | Light (Hz) | ÷ 240 | Nearest note |
|---|
Sound-color synesthetes are more consistent over time and more specific in their color choices than non-synesthetes. They still follow the same underlying mappings as everyone else: higher pitch with lighter color, and timbre affecting chroma17. Ward and colleagues concluded that synesthesia uses cross-modal mechanisms common to all people17.
Different synesthetes do not agree on which color goes with which note. A shared true color for C does not exist even among synesthetes.
Two multilevel meta-analyses of 104 randomized trials (327 effect sizes, 9,617 participants) found that music interventions reduce physiological and psychological stress (Fig 4.1)22. Music therapy delivered by qualified therapists showed a larger effect across 47 studies42. A Cochrane review found that music listening reduced preoperative anxiety43.
Tempo is the main lever. Bernardi, Porta and Sleight recorded 24 young adults, half of them musicians, across six music styles. Breathing, blood pressure and heart rate rose with faster tempi and simpler rhythms, and a two-minute pause lowered heart rate, blood pressure and minute ventilation below baseline6. The authors concluded that tempo drives most of music's arousal effect and that relaxation is clearest during a pause6.
Across studies of vocal and musical expression, calm and tenderness are carried by slow tempo, low sound level, lower and less variable pitch, soft timbre, legato articulation and slow, smooth onsets7. Relaxing music before a laboratory stressor changed autonomic recovery but did not dampen the cortisol response, so music is not a uniform buffer44. Ambient music whose loudness swelled at a slowed, personalized breathing rate lowered breathing rate, heart rate and skin conductance38.
| Parameter | Recommendation | Evidence |
|---|
The sequence opens near the listener's state and then slows, descends, softens and lengthens its notes over a sustained drone, one phrase per breath6,7,36. The scale is D-major pentatonic (D, E, F♯, A, B). It contains no semitones, so any combination of its notes stays consonant. Standard tuning (A4 = 440 Hz) is used, since the evidence for alternative tunings is weak24.
Each phrase lasts one breath. Amplitude swells on the inhale (about 4 seconds) and decays on the exhale (about 6 seconds), following the amplitude-envelope approach in breathing-paced sound38. The melody falls about an octave and a half overall.
| Phase | Notes (Hz) | Tempo | Values | Level | Breaths |
|---|
Colors move from warm, more saturated and lighter toward cool, desaturated and mid-light. The sound moves from brighter, faster, higher and louder toward darker, slower, lower and softer. These are the directions found for emotion-mediated music-color matching19,20.
| Step | Time | Color | Notes (Hz) | Tempo | Level | Transition |
|---|
Lower chroma reduces arousal. Slower tempo, softer sound and silence reduce physiological arousal. Music reduces stress overall. Slow, minor music is matched with greyer, bluer and darker colors. Breathing-paced sound slows breathing.
The iso-principle applied to color; the specific hue order, dwell times and crossfade lengths; synchronizing the two channels; and the specific notes and decibel levels. None of these has been tested as a combined sequence.
How many short videos does a person see in 15 minutes, and how can each one carry its own color and pitch while the session as a whole still follows the calm arc?
At 5 to 10 seconds per video, a 15-minute session covers 90 to 180 videos. The midpoint pace of 7.5 seconds gives 120.
Measured behavior is slower. TikTok's browsing-history export records when each video starts, so researchers infer viewing time from the gap between consecutive entries46. In 107 donated histories, sessions averaged about 8 minutes, and session sizes in the paper's robustness analysis ran from about 24 to 32 videos, or roughly 3 to 4 videos per minute46. In timed 10-minute browsing sessions, 68 participants averaged 22 to 31 videos47. Together these imply about 35 to 60 videos in 15 minutes.
The 5-to-10-second assumption therefore describes a fast skimmer who leaves most videos early. Most participants in a 347-person data-donation study watched only 30% to 50% of their videos to the end49. The median video in another donated dataset ran 23 seconds, with the middle half between 11 and 60 seconds50. Swipes cluster near the start and the end of a video. In one crowdsourced study, 29% of swipes came in the first fifth of a video and 42% in the last fifth51. Industry panels put a typical session near 11 minutes, though those figures are not peer reviewed52.
The session splits into seven groups that follow the calm arc from Part 6. Inside each group, colors and notes are drawn at random under rules that keep the overall trajectory calming. Every draw produces a new sequence.
Scroll, swipe up, or press ↓ or Space for the next video. Stay on a video to linger. Esc ends the loop.
Press Play or Preview to hear the sequence, or Scroll this loop to watch it like a feed.
No loops recorded yet. Press Scroll this loop and move through the feed. Stay longer on any video that feels good; the next loop will bring it back more often.
| Group | Time | Hue categories (weight) | Note pool | Chroma |
|---|
Dwell time is a useful but ambiguous signal. Recommender systems use it beyond clicks once it is normalized across devices and contexts53, and watch-time models remove the bias that longer videos earn longer watch times before learning from it54.
Here the main bias is position. Later pages are calmer and the viewer may be drowsier, so dwell drifts upward through a loop. Each page is compared with the median of its 16 neighbors on a log scale. It counts as a linger when it is at least 1.5 times that median and its robust z-score reaches 2.
A long stay can mean liking, calm, distraction or sleep. The lab cannot tell these apart, so every long stay counts, and the decay described below lets a stray distraction fade. Time spent in another tab or app is not counted. Three minutes without input is logged as possible sleep. The note keeps sounding, and that page counts once like any other linger.
TikTok exploits what it has learned about a user for roughly 30% to 50% of the videos it shows55, and users rate sessions with novel content higher than sessions close to their history47. The lab caps remembered pages at 25% of each group and draws the rest fresh.
Each remembered page keeps a Beta score of lingers against non-lingers. A random draw from that score decides whether the page returns: proven favorites usually come back, uncertain ones sometimes do, and pages that stop drawing lingers fade out. This is Thompson sampling56. Hue categories and notes that drew lingers in a group also get a bounded weight boost (0.6 to 1.8 times) in fresh draws. Scores decay 15% per loop, so old favorites make room for new ones.
Across 300 simulated viewers, each with one favorite hue and one favorite note, remembered pages drew lingers about twice as often as fresh pages (about 16% against 8.5%). By the tenth loop about a fifth of pages came from memory and the favorite hue's share rose from 14% to 18%. The calm arc caps how far the feed can drift: each group allows only three hue categories and never repeats one back to back.
Remembered pages stay private to you and are saved in this browser. Remembered pages return in their original group, so every new loop still follows the calm arc.
The page's note keeps sounding for as long as a viewer stays. Past the page's planned time it starts to breathe, easing to half level and rising again once per breath, each breath a little longer than the last. The schedule follows breathing-pacing research, because no controlled study of tone-interval schedules for hypnosis was found.
Slow breathing, under 10 breaths per minute and most often 6, raises heart-rate variability and self-rated relaxation, although the 15 studies behind that conclusion are small and none is randomized57. Resonance-frequency protocols step breathing down by 0.5 breaths per minute, from 6.5 to 4.5, hold each rate for 2 minutes and favor a longer exhale58. Whether people breathed at 6 or 12 per minute, a short inhale with a long exhale (ratio 0.42) produced more relaxation than the reverse59.
Gradual leading works. Device-guided breathing plays one tone for inhaling and one for exhaling and lengthens mainly the exhale tone; in one case report breathing fell from about 16 to 6 per minute during 15-minute daily sessions60. Music whose loudness swelled at 75% of a listener's breathing rate slowed breathing as well38.
Rhythm can ease the way into sleep. Rocking at 0.25 Hz, one cycle every 4 seconds, eased the transition into sleep and boosted spindles and slow oscillations61. Twenty minutes of breathing at 6 per minute before bed shortened sleep onset in people with insomnia62. Sleep playlists are slower and softer than music in general63.
Predictability matters too. Responses fade with repetition, and a novel stimulus restores them, a process called dishabituation64. Each change in the rhythm should therefore stay small.
Hypnosis research offers no timing rule. Repetitive drumming did not lower cortisol more than meditation music, and its dreamlike experiences depended on the instructions given65. People respond to suggestion almost as much without a hypnotic induction as with one, and expectancy explains much of who responds66. No controlled study supports Fibonacci-style or other special interval sequences, and a direct test of the iso-principle, with 9 participants, found no clear advantage67.
The rhythm should meet the viewer before it leads, so the first breath lasts 5 seconds. That is 12 breaths per minute, the slow end of the 12 to 20 per minute that adults breathe at rest68. A shorter first breath would sit inside or above ordinary resting breathing and offer nothing to follow. A first breath of 10 seconds would jump straight to the 6-per-minute target, a step of 50% or more from a resting pace and far larger than the steps that pacing protocols use58. Each breath then stretches by 6%, close to the 8% steps used in resonance-frequency protocols58, so no single change is large enough to pull attention back through dishabituation64. The stretch slows to 3% at 10 seconds, the 6-per-minute rate with the most support37,57, and holds at 12 seconds, inside the adult resonance range58. The exhale grows faster than the inhale, because a short inhale with a long exhale relaxed people more than the reverse59. Gradual lengthening is also how device-guided breathing slowed one user from about 16 to 6 breaths per minute60.
| Breath | Elapsed | Length | Per min | In / out |
|---|
At night the app trades the screen colors of Sequence A for the evening colors of Sequence B (2.5). A sideways swipe switches: right for night, left for day. Everything else in a session stays as it is.
Each video keeps its group, note, timing and place in the session; only its color changes. It takes its group's Sequence B color, turned by up to 6° of hue according to where its day color sits in its hue family, so neighboring videos still differ a little.
Dimming replaces falling chroma as the main lever. Each group's color is dimmed to its Sequence B intensity, from 1.00 for warm white to 0.05 for the deep red night-light, and the session follows one luminance curve through those levels, stepping geometrically from one group's center to the next. The curve never rises, so no video is brighter than the one before it. Dimming scales the light itself, which keeps the hue and lowers OKLCH lightness and chroma by the cube root of the dimming factor.
No hue from 160° to 300°, teal and cyan through blue, appears at night. Text keeps a contrast of at least 4.5 : 1 and turns light on the darker videos. The sound, the pace fit, the linger rhythm and the learning loop work the same by day and by night, and the app notes which videos were seen in night colors.
Faint arrows at the left and right edges show the swipe and fade out once it has been used, and a short notice names the mode as it changes. The choice carries over to the next session and can also be set in Settings. The Done for now screen explains both modes.
| # | Start | Dwell | Group | Hue | Color | OKLCH | Note | Hz | Level | Fade | Source |
|---|
Numbers in the text link here. Secondary and non-peer-reviewed sources are marked.