Skip to content

Schlieren 0.1.0: the warm air rising off the live picture made visible, as a schlieren camera shows it, in relief, color, shadowgraph lines or fringes - #34

Merged
kevinraymond merged 2 commits into
mainfrom
schlieren-0.1.0
Oct 30, 2026
Merged

kevinraymond merged 2 commits into
mainfrom
schlieren-0.1.0

Conversation

@kevinraymond

@kevinraymond kevinraymond commented Oct 30, 2026 •

Copy link
Copy Markdown
Owner

The warm air rising off the live picture made visible, as a schlieren
camera shows it (board #4847). What stands in front in the depth map or
mask, the picture's lights and whatever moves warm the air above them;
the warm air rises in a plume that sways and curls, stirred by what
moves through it, and the picture is seen through it, the subject and
the lights opaque.

  • The air: stable fluids (Stam 1999; Harris 2004) at a quarter of the
    picture's size, vorticity confinement, lib incompressible() with 20
    passes, its speed capped at 0.4 frame heights a second. It rises in
    a column above the picture's warmth (a mipmapped march: single-cell
    taps rose in a ladder of bands), never above the heat it carries.
  • The heat: half the picture's size, carried by MacCormack's step with
    Selle et al.'s clamp (2008), heated toward the smoothed warmth with
    a sharpened edge (unsmoothed, the movement's 1/8 cells came out as a
    grid of bumps), cooling as it mixes into the room.
  • The looks (Settles 2001; Brownlee et al., TVCG 2011): Knife Edge
    (relief added to the picture, tanh cutoff; multiplied, it vanished
    over a dark room), Rainbow (a spectral hue wheel by the bending's
    direction, Wyman et al.'s CIE fit from lib), Shadowgraph (the
    Laplacian), Interferogram (straight fringes the heat bends), and
    Backdrop 0, a gray schlieren photograph with the subject black.
  • What warms the air: lib warmth() from the depth's "what stands out",
    the picture's lights (bright and brighter than what is around them,
    so a bright sky isn't one) and movement past a dead zone.
  • Music: beats puff the warm air up through the flash gate, weaker
    where more of the picture is warm; loudness heats faster; the seven
    bands heat seven heights; bars turn the blade, at most once every
    2 s. Unwired, a soft white light circles over Movie.2 (Fluid's
    trick), since the still banana warmed nothing.
  • Flashing: a beat's puff goes through the flash gate (Flashing
    setting). Stress beat on the dancer (every style), flowers, plaza,
    hands and koi: each within its setting's rule, the dancer at the
    clip's own everywhere, plaza and koi at Off's half-flash margin, no
    red flashes (the red guard on every pixel). On the way: lights made
    opaque and the shift capped (a lamp's point flashed 16 times a
    second), puffs weighted by how much is warm, the blade turned at most
    once every 2 s, Rise rescaled and the air's speed capped. Strength,
    Rise and Beat Puffs all at 1 on the flowers measured 2.0, 3.5, 3.5,
    half a flash over WCAG's 3: documented as a corner.
  • 3.1 ms GPU (95th percentile 4.1) and 0.74 ms CPU a frame at 1280 x 720
    on the M2 Pro, 60 fps, the median of three; nothing cooks while unused
    (0 of 292). The Mac was short of memory this session (Blob Track alone
    read 7 ms against its page's 2.3), so these may read high.
  • Patents: clear by design (private/patents/schlieren.md, board
    #4903). MIT's US 9,710,917 and US 10,636,149 measure real air from a
    background: nothing is measured. ByteDance's US 12,694,597 advects a
    displayed fluid color by a plain backtrace: the heat is never shown
    as a color and is carried by MacCormack's corrected step. ByteDance's
    US 12,182,906 moves a fluid by a tracked object: nothing is tracked.
    Huawei's pending US 2025/0117558 takes the velocity from the carried
    density: the lift comes from the picture's warmth.

Also: Showreel 0.1.32 with twenty-nine effects (idle 0 of 8585, flash
within each setting; its page keeps 0.1.31's cost, since this session's
GPU times ran high), the README row, status and hero (2, nine rows of
three, one of two), the prototype (dev/protos/schlieren) and two probes
(count_pars, schlieren_look).

docs/BUILDING.md gains Schlieren's three flash lessons (a bright point
the bending shifts, puffs over a picture warm all over, relief that
turns or churns) and a note that the Mac's memory counts when
measuring.

…e, as a schlieren camera shows it, in relief, color, shadowgraph lines or fringes

The warm air rising off the live picture made visible, as a schlieren
camera shows it (board #4847). What stands in front in the depth map or
mask, the picture's lights and whatever moves warm the air above them;
the warm air rises in a plume that sways and curls, stirred by what
moves through it, and the picture is seen through it, the subject and
the lights opaque.

- The air: stable fluids (Stam 1999; Harris 2004) at a quarter of the
  picture's size, vorticity confinement, lib incompressible() with 20
  passes, its speed capped at 0.4 frame heights a second. It rises in
  a column above the picture's warmth (a mipmapped march: single-cell
  taps rose in a ladder of bands), never above the heat it carries.
- The heat: half the picture's size, carried by MacCormack's step with
  Selle et al.'s clamp (2008), heated toward the smoothed warmth with
  a sharpened edge (unsmoothed, the movement's 1/8 cells came out as a
  grid of bumps), cooling as it mixes into the room.
- The looks (Settles 2001; Brownlee et al., TVCG 2011): Knife Edge
  (relief added to the picture, tanh cutoff; multiplied, it vanished
  over a dark room), Rainbow (a spectral hue wheel by the bending's
  direction, Wyman et al.'s CIE fit from lib), Shadowgraph (the
  Laplacian), Interferogram (straight fringes the heat bends), and
  Backdrop 0, a gray schlieren photograph with the subject black.
- What warms the air: lib warmth() from the depth's "what stands out",
  the picture's lights (bright and brighter than what is around them,
  so a bright sky isn't one) and movement past a dead zone.
- Music: beats puff the warm air up through the flash gate, weaker
  where more of the picture is warm; loudness heats faster; the seven
  bands heat seven heights; bars turn the blade, at most once every
  2 s. Unwired, a soft white light circles over Movie.2 (Fluid's
  trick), since the still banana warmed nothing.
- Flashing: a beat's puff goes through the flash gate (Flashing
  setting). Stress beat on the dancer (every style), flowers, plaza,
  hands and koi: each within its setting's rule, the dancer at the
  clip's own everywhere, plaza and koi at Off's half-flash margin, no
  red flashes (the red guard on every pixel). On the way: lights made
  opaque and the shift capped (a lamp's point flashed 16 times a
  second), puffs weighted by how much is warm, the blade turned at most
  once every 2 s, Rise rescaled and the air's speed capped. Strength,
  Rise and Beat Puffs all at 1 on the flowers measured 2.0, 3.5, 3.5,
  half a flash over WCAG's 3: documented as a corner.
- 3.1 ms GPU (95th percentile 4.1) and 0.74 ms CPU a frame at 1280 x 720
  on the M2 Pro, 60 fps, the median of three; nothing cooks while unused
  (0 of 292). The Mac was short of memory this session (Blob Track alone
  read 7 ms against its page's 2.3), so these may read high.
- Patents: clear by design (private/patents/schlieren.md, board
  #4903). MIT's US 9,710,917 and US 10,636,149 measure real air from a
  background: nothing is measured. ByteDance's US 12,694,597 advects a
  displayed fluid color by a plain backtrace: the heat is never shown
  as a color and is carried by MacCormack's corrected step. ByteDance's
  US 12,182,906 moves a fluid by a tracked object: nothing is tracked.
  Huawei's pending US 2025/0117558 takes the velocity from the carried
  density: the lift comes from the picture's warmth.

Also: Showreel 0.1.32 with twenty-nine effects (idle 0 of 8585, flash
within each setting; its page keeps 0.1.31's cost, since this session's
GPU times ran high), the README row, status and hero (2, nine rows of
three, one of two), the prototype (dev/protos/schlieren) and two probes
(count_pars, schlieren_look).
@kevinraymond
kevinraymond merged commit fe2e154 into main Oct 30, 2026
1 check passed
@kevinraymond
kevinraymond deleted the schlieren-0.1.0 branch October 30, 2026 07:51
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant