| FazBrowse GitHub Viewer | Trending | | Home |
| Tools: [Download Repo ZIP] [Original HTTPS Page] |
| Name | Name | Last commit date | ||
|---|---|---|---|---|
This project is a port of standard easing equations, CSS easings and many ported GLSL transitions for use in tandem with easing or alone. The easing expressions can be used for other filters besides xfade.
There are 2 variants:
Xfade is a FFmpeg video transition filter with many built-in transitions and an expression evaluator for custom transitions. However the progress rate is linear, starting and stopping abruptly and proceeding at constant speed, therefore transitions lack interest. Easing inserts a progress envelope to smooth transitions in a natural way.
Example usage:
custom ffmpeg:
set the new easing option to the easing name, with optional CSS-easing arguments,
and the transition option to the transition name, with optional transition arguments,
and the reverse option to reverse the easing and/or transition effect
(see reversing).
Example (quartic-out, radial):
xfade=duration=3:offset=10:easing=quartic-out:transition=radial
Example (CSS-easing, ported-GLSL transition, reversed):
xfade=duration=3:offset=10:easing='cubic-bezier(0.12,0.57,0.63,0.21)'
:transition='gl_cube(floating=5,unzoom=0.8,background=SlateGray)':reverse=1
custom expression:
set the xfade transition option to custom and the expr option to the concatenation of a standard easing expression and a transition expression
(this variant does not support CSS easings or reversed effects).
Example (quartic-out, radial):
xfade=duration=3:offset=10:transition=custom:expr='st(0,P^4);
st(1,atan2(X-W/2,Y-H/2)-(ld(0)-0.5)*PI*2.5); st(1,st(1,clip(ld(1),0,1))*ld(1)*(3-2*ld(1))); B*ld(1)+A*(1-ld(1))'
Pre-generated expressions can be copied verbatim from supplied files.
A CLI wrapper script is provided to generate custom expressions, test videos, slideshows and more. It also facilitates generic easing of ffmpeg filters – see Easing other filters.
The custom ffmpeg variant has backward compatible xfade arguments, is fast with a simple C API and no restrictions. Installation involves a few patches to a single ffmpeg C source file, with no dependencies. The custom expression variant is convenient but clunky – see performance – and runs on plain vanilla ffmpeg but with restrictions: it doesn’t support CSS easings, certain transitions, the reverse feature, full colour or textures.
ffmpeg -i first.mp4 -i second.mp4 -filter_complex "
xfade=duration=3:offset=1:easing=cubic-in-out:transition=wipedown
" output.mp4Easing mode in-out is the default mode; the above is equivalent to easing=cubic.
The default easing is linear (none).
ffmpeg -i first.mp4 -i second.mp4 -filter_complex_threads 1 -filter_complex "
xfade=duration=3:offset=1:transition=custom:expr='
st(0, if(lt(P, 0.5), 4 * P^3, 1 - 4 * (1-P)^3)) ;
if(gt(Y, H*(1-ld(0))), A, B)
'" output.mp4Here, the expr parameter is shown on two lines for clarity.
The first line is the easing expression $e(P)$ (cubic in-out) which stores its calculated progress value in st(0).
The second line is the transition expression $t(e(P))$ (wipedown) which loads its eased progress value from ld(0) instead of P.
The semicolon token combines expressions.
Note
ffmpeg option -filter_complex_threads 1 is required because ffmpeg expression state variables (the st() & ld() functions) are shared between slice processing jobs and therefore not thread-safe, consequently processing is slower
In this example you can copy the easing expression from file xfade-easings-inline.txt and the transition expression from eased-transitions-rgb24-inline.txt or eased-transitions-yuv420p-inline.txt. Those contain inline expressions for CLI use.
Alternatively use the CLI script:
xfade-easing.sh -t wipedown -e cubic -x -dumps the xfade expr parameter:
'st(0,if(lt(P,0.5),4*P^3,1-4*(1-P)^3));if(gt(Y,H*(1-ld(0))),A,B)'
Some expressions are very long, so using a filtergraph script keeps things manageable and readable.
For this same example you can copy the easing expression from file xfade-easings-script.txt and the transition expression from eased-transitions-rgb24-script.txt or eased-transitions-yuv420p-script.txt. Those contain multiline expressions for script use (but the inline expressions work too).
Alternatively use xfade-easing.sh with expansion specifiers expr='%n%X' (see Usage):
xfade-easing.sh -t wipedown -e cubic -s "xfade=offset=10:duration=5:transition=custom:expr='%n%X'" -x script.txtwrites the complete xfade filter description to file script.txt:
xfade=offset=10:duration=5:transition=custom:expr=' st(0, if(lt(P, 0.5), 4 * P^3, 1 - 4 * (1-P)^3)) ; if(gt(Y, H * (1 - ld(0))), A, B)'
and the command becomes
ffmpeg -i first.mp4 -i second.mp4 -filter_complex_threads 1 -/filter_complex script.txt output.mp4`Note
this option syntax has changed but is undocumented:
For simplicity, native xfade-easing is a header-only implementation in xfade-easing.h patched into vf_xfade.c at an optimal place. It comprises static functions only, sharing internal linkage with the vf_xfade.c compilation unit, so no Makefile changes are necessary.
My repo ffmpeg-makexe has a Bash script to build ffmpeg easily with (or without) xfade-easing under MSYS2 in two dispositions:
minimal static build (x264 + zlib) using
larger dynamic build using
these use dynamically-linked external components, creating a 7-Zip archive of all non-Windows binaries.
Native Msys2 Windows build using media-autobuild_suite is also possible but it broke for me. Both that and ffmpeg-windows-build-helpers are complex scripts promoted by the FFmpeg team.
I have not explored WSL or Cygwin.
Cross compiling using ffmpeg-windows-build-helpers on Linux is the method I have had most success with. It built a static feature-rich ffmpeg.exe with xfade-easing on a VirtualBox Ubuntu client running on Macos, but attempting the same process natively on Macos failed – needs investigation. You need to follow steps 2-4 above first and use the --ffmpeg-source-dir option.
Please see the FFmpeg Compilation Guide.
FFmpeg contains x264 and other components which require compliance with the GPL, therefore I am unable to distribute binary executables of ffmpeg with xfade-easing.
The custom FFmpeg version has been built and tested on Macos with clang, Ubuntu Linux and Msys2 with gcc and clang, and Windows with MSVC and clang-cl.
Pre-generated easing and transition expressions are in the expr/ subdirectory for mix and match use. The CLI script can produce combined expressions in any syntax using expansion specifiers (like printf).
This format is condensed into a single line stripped of whitespace.
Example: elastic out easing (leaves progress in st(0))
st(0,cos((1-P)*20.944)/2^(10*(1-P)))
This format is best for expressions that are too unwieldy for inline ffmpeg commands.
Example: gl_Swirl transition (expects eased progress in ld(0)) (cf. Swirl.glsl)
st(1, 0.5); st(2, 1); st(3, X / W - 0.5); st(4, 0.5 - Y / H); st(5, hypot(ld(3), ld(4))); if(lt(ld(5), ld(1)), st(1, (ld(1) - ld(5)) / ld(1)); st(5, 1 - 2 * abs(ld(0) - 0.5)); st(1, ld(1) * ld(1) * ld(5) * 8 * PI); ifnot(ld(2), st(1, -ld(1))); st(5, sin(ld(1))); st(6, cos(ld(1))); st(1, ld(3) * ld(6) - ld(4) * ld(5)); st(4, ld(3) * ld(5) + ld(4) * ld(6)); st(3, (ld(1) + 0.5) * W); st(4, (0.5 - ld(4)) * H); st(5, ifnot(PLANE, a0(ld(3),ld(4)), ifnot(1-PLANE, a1(ld(3),ld(4)), ifnot(2-PLANE, a2(ld(3),ld(4)), a3(ld(3),ld(4)))))); st(6, ifnot(PLANE, b0(ld(3),ld(4)), ifnot(1-PLANE, b1(ld(3),ld(4)), ifnot(2-PLANE, b2(ld(3),ld(4)), b3(ld(3),ld(4)))))); ld(6) * (1 - ld(0)) + ld(5) * ld(0), B * (1 - ld(0)) + A * ld(0) )
These use P directly for progress instead of ld(0).
Example: gl_WaterDrop transition (cf. WaterDrop.glsl)
st(1, 30); st(2, 30); st(3, 1 - P); st(4, X / W - 0.5); st(5, 0.5 - Y / H); st(6, hypot(ld(4), ld(5))); st(7, if(lte(ld(6), ld(3)), st(1, sin(ld(6) * ld(1) - ld(3) * ld(2))); st(4, X + ld(4) * ld(1) * W); st(5, Y - ld(5) * ld(1) * H); ifnot(PLANE, a0(ld(4),ld(5)), ifnot(1-PLANE, a1(ld(4),ld(5)), ifnot(2-PLANE, a2(ld(4),ld(5)), a3(ld(4),ld(5))))), A )); ld(7) * (1 - ld(3)) + B * ld(3)
These ease ld(0) instead of P - see Easing other filters.
This implementation derives from Michael Pohoreski’s single argument version of Robert Penner’s easing functions, further optimised by me for the peculiarities of xfade.
The squareroot and cuberoot easings focus more on the middle regions and less on the extremes, opposite to quadratic and cubic respectively:
The flipelastic and flipback easings invert overshoots to keep the progress value within limits, like faster and slower forms of bounce respectively, see Overshoots.
Here are all the above easings superimposed using the Desmos Graphing Calculator:
The custom ffmpeg variant supports CSS Easing Functions Level 2 which are too complex for custom expressions:
The new CSS linear() function can approximate any progress contour by interpolating between adjacent points, documented at W3C here. There’s a CSS Linear() Generator online by its pioneer Jake Archibald to convert easings expressed in JavaScript or SVG to linear().
There are 4 fixed CSS smoothing curves and a general cubic-bezier() easing function documented at W3C here. See also the CSS Cubic Bezier Generator to craft your own. The implementation used here is transpiled from Apple’s open-source Webkit.
The CSS steps() staircase function is for transitions that jump a constant amount, documented at W3C here.
Standard easings have ease-in and ease-out modes but not CSS easings. See Reversing Xfade effects to achieve this with the custom ffmpeg build.
The elastic and back easings overshoot out of range for xfade. CSS linear() and cubic-bezier() easings can also overshoot. This causes unpredictable behaviour and even access violation exceptions, so needs to be prevented (but see Easing other filters). The resolution depends on the variant:
Easings flipelastic and flipback flip elastic and back overshoots into range, creating more interest than clipping, see Supplementary easings.
Examples: elastic-in and back overshoot handling for slideup transition
The easing expressions are useful for filters other than xfade, e.g. blend, drawtext, geq, overlay, rotate, zoompan, etc. – anywhere an ffmpeg expr is used to calculate filter options.
For this purpose the CLI script includes text expansion codes %g & %G to generate generic easing expressions for the value in ld(0) (instead of P for xfade), leaving the result in ld(0). You can also copy generic easing expressions from file generic-easings-inline.txt for inline -filter_complex use, or generic-easings-script.txt for -/filter_complex scripts.
To ease other filters, store a normalised input value in st(0,…), append the easing expression, then scale the eased result left in ld(0).
Example: zoompan filter with elastic-out zooming
Here’s the zoom option expression for the zoompan filter:
zoom='st(0, clip((time - 1) / 3, 0, 1));
st(0, 1 - cos(ld(0) * 20.944) / 2^(10 * ld(0)));
lerp(1, 3, ld(0))'
The first line stores a 3 second duration delayed by 1 second normalised to a value between 0 and 1.
The last line scales the result to zoom between 1x and 3x.
The middle line performs elastic-out easing, obtained from generic-easings-script.txt, or
xfade-easing.sh -e elastic-out -s %G -x -
The zoompan filter can produce impressive Ken Burns effects when zoom, x, y are all dynamic. Adding easing can take the illusion of motion even further.
Example: zoompan with back zooming and drawtext with squareroot scrolling
The initial zoom here is 1.2x to accommodate the 10% undershoot that back easing produces. So the zoompan zoom expression, with back expr from generic-easings-inline.txt, is:
z='st(0, clip((time - 1) / 3, 0, 1));
st(0,if(lt(ld(0),0.5),(ld(0)*7.18982-2.59491)*ld(0)*ld(0)*2,(ld(0)*7.18982-4.59491)*(1-ld(0))^2*2+1))
lerp(1.2, 3.1, ld(0))'
And the drawtext y expression with squareroot easing is:
y='st(0, clip((t - 1) / 3, 0, 1));
st(0, if(lt(ld(0), 0.5), sqrt(ld(0) / 2), 1 - sqrt((1 - ld(0)) / 2)));
lerp(line_h - 10, h - line_h * 2 - 5, ld(0))'
For the custom expression variant, xfade transitions are provided as custom expressions for use with easing, converted from C-code in vf_xfade.c. The custom ffmpeg variant eases the built-in xfade transitions. Omitted transitions are distance and hblur which perform aggregation, so cannot be processed efficiently on a per plane-pixel basis.
※ native ffmpeg only
Here are the xfade transitions processed using custom expressions instead of the built-in transitions (for testing), without easing – see also the FFmpeg Wiki Xfade page:
The open collection of GL Transitions initiative lead by Gaëtan Renaudeau (gre) “aims to establish an universal collection of transitions that various softwares can use” released under a Free License.
Other GLSL transition sources here are from shadertoy and the Vegas Forum.
Most of the transitions at gl-transitions and many from elsewhere have been transpiled into native C transitions (for custom ffmpeg variant) and custom expressions (for custom expression variant) for use with or without easing.
All GLSL transitions adapted to the GL Transition Specification are in glsl/.
The following list shows the transition names, customisation parameters and defaults, and authors:
| transition | parameters (=default) | author |
|---|---|---|
| gl_angular | startingAngle=90 clockwise=0 |
Fernando Kuteken |
| gl_Bars | vertical=0 | Mark Craig |
| gl_blend※ | mode=0 | scriptituk |
| gl_BookFlip | hong | |
| gl_Bounce | bounces=3 direction=0 shadowAlpha=0.6 shadowHeight=0.075 shadowColor=0 |
Adrian Purser |
| gl_BowTie※ | vertical=0 | huynx |
| gl_ButterflyWaveScrawler※ | amplitude=1 waves=30 colorSeparation=0.3 |
mandubian |
| gl_cannabisleaf | Flexi23 | |
| gl_chessboard | grid=8 | lql |
| gl_CornerVanish | Mark Craig | |
| gl_CrazyParametricFun | a=4 b=1 amplitude=120 smoothness=0.1 |
mandubian |
| gl_crosshatch | center.x=0.5 center.y=0.5 threshold=3 fadeEdge=0.1 |
pthrasher |
| gl_CrossOut | smoothness=0.05 | Mark Craig |
| gl_crosswarp | Eke Péter | |
| gl_CrossZoom※ | strength=0.4 centerFrom.x=0.25 centerFrom.y=0.5 centerTo.x=0.75 centerTo.y=0.5 |
rectalogic |
| gl_cube | persp=0.7 unzoom=0.3 reflection=0.4 floating=3 background=0 |
gre |
| gl_Diamond | smoothness=0.05 | Mark Craig |
| gl_DirectionalScaled | direction.x=0 direction.y=1 scale=0.7 background=0 |
Thibaut Foussard |
| gl_directionalwarp | smoothness=0.1 direction.x=-1 direction.y=1 |
pschroen |
| gl_doorway | reflection=0.4 perspective=0.4 depth=3 background=0 |
gre |
| gl_DoubleDiamond | smoothness=0.05 | Mark Craig |
| gl_Dreamy | mikolalysenko | |
| gl_EdgeTransition※ | edgeThickness=0.001 edgeBrightness=8 |
Woohyun Kim |
| gl_Exponential_Swish※ | zoom=0.8 angle=0 offset.x=0 offset.y=0 exponent=4 wrap.x=2 wrap.y=2 blur=0 background=0 |
Boundless |
| gl_fadecolor※ | color=0 colorPhase=0.4 |
gre |
| gl_FanIn | smoothness=0.05 | Mark Craig |
| gl_FanOut | smoothness=0.05 | Mark Craig |
| gl_FanUp | smoothness=0.05 | Mark Craig |
| gl_Flower | smoothness=0.05 rotation=360 |
Mark Craig |
| gl_GridFlip※ | size.x=4 size.y=4 pause=0.1 dividerWidth=0.05 randomness=0.1 background=0 |
TimDonselaar |
| gl_heart | gre | |
| gl_hexagonalize | steps=50 horizontalHexagons=20 |
Fernando Kuteken |
| gl_InvertedPageCurl | angle=100 radius=0.159 reverseEffect=0 |
Hewlett-Packard |
| gl_kaleidoscope | speed=1 angle=1 power=1.5 |
nwoeanhinnogaehr |
| gl_LinearBlur | intensity=0.1 | gre |
| gl_Lissajous_Tiles※ | grid.x=10 grid.y=10 speed=0.5 freq.x=2 freq.y=3 offset=2 zoom=0.8 fade=3 power=3 background=0 |
Boundless |
| gl_morph※ | strength=0.1 | paniq |
| gl_Mosaic | endx=2 endy=-1 |
Xaychru |
| gl_perlin | scale=4 smoothness=0.01 |
Rich Harris |
| gl_pinwheel | speed=2 | Mr Speaker |
| gl_polar_function | segments=5 | Fernando Kuteken |
| gl_PolkaDotsCurtain | dots=20 centre.x=0 centre.y=0 |
bobylito |
| gl_powerKaleido | scale=2 z=1.5 speed=5 |
Boundless |
| gl_randomNoisex | towrabbit | |
| gl_randomsquares | size.x=10 size.y=10 smoothness=0.5 |
gre |
| gl_ripple | amplitude=100 speed=50 |
gre |
| gl_Rolls | type=0 rotDown=0 |
Mark Craig |
| gl_RotateScaleVanish | fadeInSecond=1 reverseEffect=0 reverseRotation=0 background=0 trkMat=0 |
Mark Craig |
| gl_rotateTransition | haiyoucuv | |
| gl_rotate_scale_fade | centre.x=0.5 centre.y=0.5 rotations=1 scale=8 background=0.15 |
Fernando Kuteken |
| gl_SimpleBookCurl※ | angle=150 radius=0.1 shadow=0.2 |
scriptituk |
| gl_SimplePageCurl | angle=80 radius=0.15 roll=0 reverseEffect=0 greyBack=0 opacity=0.8 shadow=0.2 |
Andrew Hung |
| gl_Slides | type=0 slideIn=0 |
Mark Craig |
| gl_squareswire | squares.x=10 squares.y=10 direction.x=1.0 direction.y=-0.5 smoothness=1.6 |
gre |
| gl_StageCurtains※ | color=0xCC1A33FF bumps=15 drop=0.1 |
scriptituk |
| gl_StarWipe | borderThickness=0.01 starRotation=0.75 borderColor=1 |
Ben Lucas |
| gl_static_wipe | upToDown=1 maxSpan=0.5 |
Ben Lucas |
| gl_StereoViewer※ | zoom=0.9 radius=0.25 flip=0 background=0 trkMat=0 |
Ted Schundler |
| gl_Stripe_Wipe※ | nlayers=3 layerSpread=0.5 color1=0x3319CCFF color2=0x66CCFFFF shadowIntensity=0.7 shadowSpread=0 angle=0 |
Boundless |
| gl_swap | reflection=0.4 perspective=0.2 depth=3 background=0 |
gre |
| gl_Swirl | radius=1 clockwise=1 |
Sergey Kosarevsky |
| gl_WaterDrop | amplitude=30 speed=30 |
Paweł Płóciennik |
| gl_windowblinds | Fabien Benetou | |
| gl_windowslice | count=10 smoothness=0.5 |
gre |
※ native build only
Here are the ported GLSL transitions with default parameters and no easing. They are all supported by the custom ffmpeg variant but check above for the dozen or so that are not supported by the custom expression variant.
See also the GL Transitions Gallery.
GLSL transitions can also be eased, although easing is integral with some:
Example: Swirl transition with bounce easing
Many GLSL transitions accept parameters to customise the transition effect. The parameters and default values are shown above.
Example: two pinwheel speeds: 'gl_pinwheel(0.5)' and 'gl_pinwheel(10)'
Parameters are appended to the transition name as CSVs within parenthesis.
For the custom ffmpeg variant the parameters may be name=value pairs in any order, e.g. gl_WaterDrop(speed=20,amplitude=50), or they may be indexed values, as follows.
For the custom expression variant the parameters must be indexed values only but empty values assume defaults, e.g. gl_GridFlip(5,3,,0.1,,1) arguments are size.x=5,size.y=3,dividerWidth=0.1,background=1 with default values for other parameters.
Custom expressions can also be amended directly: parameters are specified using store functions st(p,v) where p is the parameter number and v its value. So for gl_pinwheel with a speed value 10, change the first line of its expr below to st(1, 10);.
st(1, 2); st(2, 1 - ld(0)); st(1, atan2(0.5 - Y / H, X / W - 0.5) + ld(2) * ld(1)); st(1, mod(ld(1), PI / 4)); if(lte(ld(2), ld(1)), A, B)
Similarly, gl_directionalwarp takes 3 parameters: smoothness, direction.x, direction.y (from xfade-easing.sh -L) and its expr starts with 3 corresponding st() (store) functions which may be changed from their default values:
st(1, 0.1); st(2, -1); st(3, 1); st(4, hypot(ld(2), ld(3))); etc.
Example: gl_InvertedPageCurl 30° with uncurl
(useful for sheet music with repeats)
'gl_InvertedPageCurl(30,0.15,0)' and 'gl_InvertedPageCurl(30,0.15,1)' concatenated
GLSL shader code runs on the GPU in real time, unlike ffmpeg. However GL Transition and Xfade APIs are broadly similar and non-complex algorithms are easily ported using simple vector resolution.
| context | GL Transitions | Xfade filter | notes |
|---|---|---|---|
| progress | uniform float progress moves from 0 to 1 |
P moves from 1 to 0 |
progress ≡ 1 - P |
| ratio | uniform float ratio | W / H | |
| coordinates | vec2 uv uv.y == 0 is bottom uv == vec2(1.0) is top-right |
X, Y Y == 0 is top (X,Y) == (W,H) is bottom-right |
GL width and height are normalised uv.x ≡ X / W uv.y ≡ 1 - Y / H |
| texture | vec4 getFromColor(vec2 uv) vec4 getToColor(vec2 uv) |
a0(x,y) to a3(x,y) or A for first input b0(x,y) to b3(x,y) or B for second input |
GL colour values are normalised GL function runs for every pixel xfade expr runs for every component (plane) of every pixel |
| plane data | normalised RGBA | GBRA or YUVA unsigned integer | xfade bit depth depends on pixel format |
| precision | single | double | (float type) |
The custom ffmpeg variant, like GL Transitions, operates on single precision unit interval coordinate and colour data, processing all planes together. GL Transitions have been altered to work with YUV plane data as well as RGB.
To make the transpiled code easier to follow, original variable names from the GLSL and xfade source code are retained in xfade-easing.sh and xfade-easing.h. The CLI script uses pseudo functions to emulate real functions, expanding them inline later.
See xfade-easing.h for the C code transpiled from GLSL that helped to optimize the custom expressions. See the files in glsl/ refactored from other GLSL transition sources that were used for intermediate testing in the GL Transition Editor.
Example: porting transition gl_randomsquares
uniform ivec2 size; // = ivec2(10, 10)
uniform float smoothness; // = 0.5
float rand (vec2 co) {
return fract(sin(dot(co.xy, vec2(12.9898, 78.233))) * 43758.5453);
}
vec4 transition(vec2 p) {
float r = rand(floor(vec2(size) * p));
float m = smoothstep(0.0, -smoothness, r - (progress * (1.0 + smoothness)));
return mix(getFromColor(p), getToColor(p), m);
}xfade-easing.sh (custom expression variant):
gl_randomsquares) # (case)
_make "st(1, ${a[0]-10});" # size.x
_make "st(2, ${a[1]-10});" # size.y
_make "st(3, ${a[2]-0.5});" # smoothness
_make 'st(1, floor(ld(1) * X / W));'
_make 'st(2, floor(ld(2) * (1 - Y / H)));'
_make 'st(4, frand(ld(1), ld(2)));' # r
_make 'st(4, ld(4) - (1 - P) * (1 + ld(3)));'
_make 'st(4, smoothstep(0, -ld(3), ld(4), 4));' # m
_make 'mix(A, B, ld(4))'
;;Here, frand(), smoothstep() and mix() are pseudo functions. Customizable parameters are generally stored first. _make is just an expression string builder function.
xfade-easing.h (custom ffmpeg variant):
static vec4 gl_randomsquares(const XTransition *e)
{
INIT_BEGIN
ARG2(ivec2, size, 10, 10)
ARG1(float, smoothness, 0.5)
INIT_END
float r = frand2(floor2(mul2(vec2i(size), e->p)));
float m = smoothstep(0, -smoothness, r - e->progress * (1 + smoothness));
return mix4(e->a, e->b, m);
}Here, vec4 and ivec2 simulate GLSL vector types and XTransition encapsulates data pertaining to a transition:
typedef struct XTransition {
float ratio; // frame width / height (cf. W / H)
float progress; // transition progress, 0.0 to 1.0 (cf. P)
vec2 p; // pixel position, .y==0 is bottom (cf. X, Y)
vec4 a, b; // plane data at p (cf. A, B)
...
} XTransition;And the extended transitions delegate is:
static void xtransition_transition(AVFilterContext *ctx,
const AVFrame *a, const AVFrame *b,
AVFrame *out,
float progress,
int slice_start, int slice_end,
int jobnr)
{
const XFadeContext *s = ctx->priv;
const XFadeEasingContext *k = s->k;
const float sw = 1.f / k->mw, sh = 1.f / k->mh, sv = 1.f / k->mv; // scale
XTransition e = { // slice data
.xf = {a, b}, // input frame data
.ratio = k->r, // pixel ratio
.progress = 1 - progress, // 0 to 1 for xtransitions
.k = k // common context
};
// pixel iterator and unit interval conversions
for (int y = slice_start; y < slice_end; y++) {
e.p.y = 1 - y * sh; // y=0 is bottom
for (int x = 0, p = 0; x <= k->mw; x++) {
e.p.x = x * sw;
e.a = PLANED, e.b = PLANED; // plane defaults
do {
e.a.p[p] = *pix(a, p, x, y) * sv; // from colour
e.b.p[p] = *pix(b, p, x, y) * sv; // to colour
} while (++p < k->n);
vec4 c = k->xtransitionf(&e); // transition colour
do {
--p;
*pix(out, p, x, y) = scaleUI(c.p[p], k->mv); // clips
} while (p > 0);
}
}
}This is transpiled from the InvertedPageCurl GL Transition which originated from the WebVfx WebGL pagecurl shader which is itself based on code by Calyptus Life AB which seems no longer available. The Hewlett-Packard accreditation by Sergey Kosarevsky is obscure but preserved here.
This is adapted from the elegant
simple page curl effect by Andrew Hung
who also provides an excellent shader breakdown
to demystify the deformation effect.
It is more versatile than gl_InvertedPageCurl and takes the following parameters:
The main differences from gl_InvertedPageCurl are:
Example: using gl_SimplePageCurl to emulate gl_InvertedPageCurl
'gl_InvertedPageCurl(30)' vs 'gl_SimplePageCurl(24.8,0.159,1,0,1,0.8,0.1)'
these parameters factor in the aspect ratio (5:4 here), 1/2π radius, roll effect, greyscale overleaf, and shadowing.
There is barely any noticeable difference, which confirms Mr Hung’s remark that complex mathematics isn’t needed: just simple trigonometry and scalar product projections. However gl_SimplePageCurl takes longer to process than gl_InvertedPageCurl.
Example: gl_SimplePageCurl with various angle and roll options (Abstract and Renaissance art by Kandinsky and Titian)
A by-product effect is a wipe transition in any direction, achieved by setting radius=0,roll=1.
Example: 'gl_SimplePageCurl(160,0,1)' 160° wipe
(custom ffmpeg only)
This is adapted from gl_SimplePageCurl to clamp the curl to the virtual ‘spine’ at the horizontal centre,
then flatten the radius to zero, using built-in easing to appear more realistic.
It takes the following parameters:
Example: gl_SimpleBookCurl with various angle and radius values paging forwards and backwards, rotated and overlaid onto a desk texture (credit: Ethical Corporation Magazine)
The CLI script can generate random GLSL-style transitions using the pseudo transition name gl_random which shuffles all the available transition names then cycles through them. Random transitions are particularly useful for slideshows. As the random transition is initially unknown, the expression options will not work in this case.
For the custom expression variant which cannot take named parameters, customisation parameters are ignored and resort to defaults.
For the custom ffmpeg variant which can take named parameters,
a common background colour or transparency or texture can be set using background, e.g.:
'gl_random(background=black)' renders a black background (the default)
'gl_random(background=white)' renders a white background
'gl_random(background=gray@0.5)' renders a semi-transparent grey background
'gl_random(background=-1)' renders a transparent background
'gl_random(background=-10)' renders the texture -10 (a rainbow effect)
and for even more randomness:
'gl_random(background=random)' renders a random background colour picked by ffmpeg
These affect transitions that take a background parameter and have no effect on those that do not.
Example: (7 random transitions)
xfade-easing.sh -X -t gl_random -v left.mp4 gl*.png (left, custom expression)
xfade-easing.sh -t 'gl_random(background=random)' -v right.mp4 gl*.png (right, custom ffmpeg)
(custom ffmpeg only)
The generic xfade reverse option reverses the transition and/or easing effects. It takes a bitmapped number:
It is necessary to swap the inputs during the reversed transition to match the inputs before and after the transition, i.e. during the offset time and after the offset+duration time. Note that elastic and back easings implicitly swap the inputs and add ±1 to the progress value during overshoots, see Overshoots.
Reversing is tricky but not impossible to implement in the xfade-easing.sh script for custom expressions. The script has a -b option to pass reverse values through to xfade for the custom ffmpeg variant.
Most standard xfade transitions have reversed equivalents, e.g. wipeleft and wiperight, but few GLSL transitions do. Unlike standard easings, CSS easings also have no mirror-image reversal mode.
Example: using the same CSS Linear coefficients as above
easing: 'linear(0, 0.5 30%, 0.2 60% 80%, 1)', transition: gl_FanUp, reverse: 0–3
There is no gl_FanDown transition but reversing gl_FanUp provides one. Reversing is also particurly useful for squeezeh, squeezev, gl_BookFlip, gl_BowTie, gl_cube, gl_doorway, gl_heart, gl_pinwheel, gl_rotateTransition, gl_Slides, gl_Swirl, gl_swap, gl_windowslice.
This is a powerful feature that considerably increases the number of transitions available.
Creative effects can result from using the same media source for both transition inputs.
Examples: some effect-only transitions with and without easing or reversing
Transitions that affect colour components work differently for RGB than non-RGB colour spaces and for different bit depths. For the custom expression variant, xfade-easing.sh emulates vf_xfade.c function config_output() by deducing the RGB signal type AV_PIX_FMT_FLAG_RGB from the -f option format name (rgb/bgr/etc. see pixdesc.c) and the bit depth from ffmpeg -pix_fmts data. It can then set the maximum and mid plane values correctly. See How does FFmpeg identify color spaces? for details.
The expression files in expr/ cater for RGB and YUV formats with 8-bit component depth. For faster processing of greyscale media use xfade-easing.sh -f gray. Greyscale is not RGB therefore it is processed like a luma plane.
If in doubt, check with ffmpeg -pix_fmts or use the xfade-easing.sh -f option.
These conventions are adopted:
Consequently a value of exactly 1 is rendered white but 2 (RGBA #00000002) is almost transparent black. To get R=0,G=0,B=1 specify the colour using hexadecimal notation, #000001.
The custom expression variant only suports transparent white and opaque grey values, -1 and 0.0 to 1.0,
it does not support colour or textures.
e.g. gl_swap(, , , 0.67) for 67% grey background (other parameters take default values).
The custom ffmpeg variant supports the full Color syntax
including named colours and variable alpha,
e.g. gl_Stripe_Wipe(color1=DeepSkyBlue, color2=ffd700)
Colour value examples:
Avoid decimal numbers above 1, e.g. 255 is not blue but opaque black (RGB #000000FF).
Example: gl_StarWipe transitions with Lime and -15 (still diamond pattern) border colours
gl_StarWipe(borderThickness=0.1, borderColor=Lime) (left)
gl_StarWipe(borderThickness=0.1, borderColor=-15) (right)
Many transitions reveal areas which are not painted, exposing a background instead. Unlike GL Transitions which show a black background, this implementation provides a background parameter which takes a colour value – grey or RGBA colour or texture or transparent (the custom expression variant only suports grey and transparent).
Transparent background transitions make for good overlay effects. Alternatively, use opaque chroma-key compositing backgrounds, commonly #00B140 for green-screen and #0047BB or #0827F5 for blue-screen.
(custom ffmpeg only)
Note
this features is experimental – for generic background effects use the ffmpeg overlay filter in conjunction with transparent xfade-easing.
Textures are shader effects ported from Shadertoy, mainly for transition backgrounds selected with the background parameter but any colour parameter can select a texture – see Colour parameters. They are referenced by a negative index, where
-2,-3: Natural vignetting by ApoorvaJ
-4,-5: glowingMarblingBlack by nasana
-6,-7: Monochrome Hyperbola by MichaelPohoreski
-8,-9: Skyline in 132 chars by GregRostami
-10,-11: simple rainbow formula by Jodie
-12,-13: simple plasma by Kastor
-14,-15: diamond pattern by rcread
-16,-17: Glowing thing by denzen
-18,-19: cinetunnel by tomviolin
-20,-21: spring time by bergi
-22,-23: Skyline4 by FabriceNeyret2
-24,-25: Water Ripple by liucc09
Example: gl_DirectionalScaled transition with glowingMarblingBlack background texture and sinusoidal easing
gl_DirectionalScaled(direction.x=-1,direction.y=1,scale=0.1,background=-4)
Textures are useful for all transitions that take a background parameter and also for gl_fadecolor, gl_StarWipe.
Blending involves compositing image layers according to a blending function and opacity formula, whereas normal transition colour mixing performs simple linear interpolation in all planes determined by a weight factor.
One exception is the gl_blend transition (custom ffmpeg only) which blends the inputs according to its numerical mode parameter. The first input acts as the backdrop layer and the second as the source layer. At half the duration, the source is blended over the backdrop according to the blend mode algorithm. Blend modes available are the CSS blend modes which use the standard PDF 1.7 section 11.3.3 compositing formula and blending functions, which are quite ubiquitous:
To blend transitions over a backdrop layer use the blend filter.
It is poorly documented and the implementation in
vf_blend.c,
vf_blend_init.h,
blend_modes.c
is quite basic: it does not render standard blending as above because it seems to be designed for premultiplied alpha,
achieved for RGB using
geq="r='r(X,Y)*alpha(X,Y)/255':g='g(X,Y)*alpha(X,Y)/255':b='b(X,Y)*alpha(X,Y)/255'"
before blend.
Example: four all_mode blend options for gl_StereoViewer transitions with 50% grey transparency
gl_StereoViewer(zoom=0.6,radius=0.3,flip=1,background=-0.5)
The expression files in expr/ cater for RGBA and YUVA transparency formats in 4 planes as well as opaque RGB and YUV in 3 planes and mono/gray in 1 plane.
For lossless intermediate video content with alpha channel support use the xfade-easing.sh -v -f options with an alpha format, e.g. rgba or yuva420p, and .mkv filename extension.
For lossy video with alpha use an alpha format and the .webm extension. Note: webm encoding is extremely slow and webm alpha is not widely supported.
To specify alpha in transition parameters, see Colour parameters.
Example: overlaid transparent gl_RotateScaleVanish transition with quadratic-in easing
ffmpeg -i skaro.png -i tardis.png -i gallifrey.png -filter_complex "
[0]loop=-1:1, fps=25, format=rgba, scale=250:-2[a];
[1]loop=-1:1, fps=25, format=rgba, scale=250:-2[b];
[2]loop=-1:1, fps=25, format=rgb24, scale=250:-2[bg];
[a][b]xfade=offset=1:duration=3:easing=quadratic-in:transition='gl_RotateScaleVanish(FadeInSecond=0,ReverseEffect=1,trkMat=1)'[fg];
[bg][fg]overlay" -t 5 -c:v libx264 -pix_fmt yuv420p drwho.mp4or using the CLI script
xfade-easing.sh -f rgba -e quadratic-in -t 'gl_RotateScaleVanish(FadeInSecond=0,ReverseEffect=1,trkMat=1)' -v alpha.mkv -z 250x skaro.png tardis.png
ffmpeg -i gallifrey.png -i alpha.mkv -filter_complex '[0]scale=250:-2[b]; [b][1]overlay' drwho.mp4This demonstrates the additional trkMat parameter which tracks the Tardis alpha value to expose Skaro behind,
then Gallifrey’s Citadel when the transition ends, both planets being opaque images.
(trkMat is only availble in the custom ffmpeg variant)
Transition gl_StereoViewer also has a trkMat parameter for clean cutout effects.
See also the example under Transition gl_SimpleBookCurl which overlays a transparent transition.
FFmpeg incorporates a simple arithmetic expression evaluator implemented as a LL(1) recursive descent parser. Custom expr strings initially get parsed into an expression tree of AVExpr nodes in libavutil/eval.c which is then executed for every pixel in each plane. Unfortunately threading must be disabled (-filter_complex_threads 1) in order to use the st() and ld() state variables shared between slices (a slice is a range of frame lines processed by a thread job). But modern CPU speeds make xfade custom transitions a viable option, if somewhat challenging.
The following plots show empirical processing times for a 3-second transition of 3-plane frames (rgb24) through a null muxer in SD (720x480) and HD (1280x720) sizes on an otherwise idle 2018 Core i7 3.2GHz Mac mini, taking the minimum times of several runs. Times include the CLI script overhead.
Note
this is an old machine now, so below each plot is a table with estimated times for later PCs
Based on benchmark scores (Geekbench Mac Benchmark Chart) these times need scaling as follows:
Mac model performance varies enormously so timing vintage Macs is very approximate. Windows performance has not been measured.
These are only needed when easing standard Xfade transitions.
plot data| Transition | SD 2017–20 |
HD 2017–20 |
HD÷1.8 2020–23 |
HD÷2.2 2023–25 |
HD÷2.7 2025– |
|---|---|---|---|---|---|
| circleclose | 17 | 44 | 24 | 20 | 16 |
| circlecrop | 12 | 30 | 17 | 14 | 11 |
| circleopen | 16 | 41 | 23 | 19 | 15 |
| coverdown | 9 | 24 | 13 | 11 | 9 |
| coverleft | 11 | 28 | 15 | 13 | 10 |
| coverright | 9 | 24 | 13 | 11 | 9 |
| coverup | 11 | 28 | 15 | 13 | 10 |
| diagbl | 11 | 30 | 16 | 13 | 11 |
| diagbr | 12 | 32 | 18 | 15 | 12 |
| diagtl | 11 | 28 | 15 | 13 | 10 |
| diagtr | 12 | 30 | 17 | 14 | 11 |
| dissolve | 8 | 20 | 11 | 9 | 8 |
| fade | 2 | 6 | 3 | 3 | 2 |
| fadeblack | 24 | 64 | 35 | 29 | 24 |
| fadefast | 8 | 21 | 12 | 10 | 8 |
| fadegrays | 33 | 88 | 49 | 40 | 33 |
| fadeslow | 8 | 21 | 12 | 10 | 8 |
| fadewhite | 24 | 62 | 34 | 28 | 23 |
| hlslice | 14 | 36 | 20 | 16 | 13 |
| hlwind | 17 | 45 | 25 | 20 | 17 |
| horzclose | 11 | 28 | 16 | 13 | 10 |
| horzopen | 11 | 28 | 16 | 13 | 10 |
| hrslice | 14 | 37 | 21 | 17 | 14 |
| hrwind | 17 | 44 | 24 | 20 | 16 |
| pixelize | 27 | 72 | 40 | 33 | 27 |
| radial | 13 | 35 | 19 | 16 | 13 |
| rectcrop | 8 | 21 | 12 | 10 | 8 |
| revealdown | 9 | 24 | 13 | 11 | 9 |
| revealleft | 11 | 28 | 16 | 13 | 10 |
| revealright | 9 | 24 | 13 | 11 | 9 |
| revealup | 11 | 28 | 15 | 13 | 10 |
| slidedown | 10 | 26 | 15 | 12 | 10 |
| slideleft | 12 | 30 | 17 | 14 | 11 |
| slideright | 10 | 26 | 15 | 12 | 10 |
| slideup | 12 | 31 | 17 | 14 | 11 |
| smoothdown | 11 | 28 | 15 | 13 | 10 |
| smoothleft | 10 | 26 | 14 | 12 | 9 |
| smoothright | 11 | 28 | 15 | 13 | 10 |
| smoothup | 10 | 25 | 14 | 12 | 9 |
| squeezeh | 7 | 17 | 9 | 8 | 6 |
| squeezev | 7 | 17 | 9 | 8 | 6 |
| vdslice | 14 | 37 | 21 | 17 | 14 |
| vdwind | 17 | 44 | 25 | 20 | 16 |
| vertclose | 11 | 28 | 16 | 13 | 10 |
| vertopen | 11 | 28 | 16 | 13 | 10 |
| vuslice | 13 | 35 | 19 | 16 | 13 |
| vuwind | 17 | 45 | 25 | 20 | 17 |
| wipebl | 4 | 9 | 5 | 4 | 3 |
| wipebr | 4 | 10 | 6 | 5 | 4 |
| wipedown | 3 | 6 | 3 | 3 | 2 |
| wipeleft | 2 | 5 | 3 | 2 | 2 |
| wiperight | 2 | 6 | 3 | 3 | 2 |
| wipetl | 3 | 8 | 5 | 4 | 3 |
| wipetr | 4 | 9 | 5 | 4 | 3 |
| wipeup | 2 | 5 | 3 | 2 | 2 |
| zoomin | 28 | 74 | 41 | 34 | 28 |
| Transition | SD 2017–20 |
HD 2017–20 |
HD÷1.8 2020–23 |
HD÷2.2 2023–25 |
HD÷2.7 2025– |
|---|---|---|---|---|---|
| gl_Bars | 9 | 23 | 13 | 11 | 9 |
| gl_BookFlip | 14 | 36 | 20 | 16 | 13 |
| gl_Bounce | 28 | 72 | 40 | 33 | 27 |
| gl_CornerVanish | 7 | 17 | 10 | 8 | 6 |
| gl_CrazyParametricFun | 59 | 156 | 87 | 71 | 58 |
| gl_CrossOut | 16 | 42 | 23 | 19 | 16 |
| gl_Diamond | 8 | 22 | 12 | 10 | 8 |
| gl_DirectionalScaled | 39 | 105 | 58 | 48 | 39 |
| gl_DoubleDiamond | 11 | 28 | 16 | 13 | 10 |
| gl_Dreamy | 31 | 82 | 45 | 37 | 30 |
| gl_FanIn | 13 | 33 | 18 | 15 | 12 |
| gl_FanOut | 13 | 33 | 18 | 15 | 12 |
| gl_FanUp | 10 | 26 | 14 | 12 | 10 |
| gl_Flower | 29 | 78 | 43 | 35 | 29 |
| gl_InvertedPageCurl | 48 | 126 | 70 | 57 | 47 |
| gl_LinearBlur | 344 | 923 | 513 | 420 | 342 |
| gl_Mosaic | 74 | 197 | 109 | 90 | 73 |
| gl_PolkaDotsCurtain | 20 | 52 | 29 | 24 | 19 |
| gl_Rolls | 35 | 91 | 51 | 42 | 34 |
| gl_RotateScaleVanish | 42 | 112 | 62 | 51 | 41 |
| gl_SimplePageCurl | 122 | 323 | 180 | 147 | 120 |
| gl_Slides | 20 | 52 | 29 | 24 | 19 |
| gl_StarWipe | 46 | 122 | 68 | 55 | 45 |
| gl_Swirl | 41 | 110 | 61 | 50 | 41 |
| gl_WaterDrop | 21 | 56 | 31 | 25 | 21 |
| gl_angular | 15 | 38 | 21 | 17 | 14 |
| gl_cannabisleaf | 28 | 73 | 41 | 33 | 27 |
| gl_chessboard | 17 | 43 | 24 | 20 | 16 |
| gl_crosshatch | 51 | 135 | 75 | 61 | 50 |
| gl_crosswarp | 35 | 93 | 51 | 42 | 34 |
| gl_cube | 68 | 178 | 99 | 81 | 66 |
| gl_directionalwarp | 62 | 164 | 91 | 75 | 61 |
| gl_doorway | 40 | 106 | 59 | 48 | 39 |
| gl_heart | 18 | 47 | 26 | 21 | 17 |
| gl_hexagonalize | 67 | 179 | 99 | 81 | 66 |
| gl_kaleidoscope | 171 | 460 | 256 | 209 | 170 |
| gl_perlin | 86 | 223 | 124 | 102 | 83 |
| gl_pinwheel | 12 | 30 | 17 | 14 | 11 |
| gl_polar_function | 17 | 46 | 25 | 21 | 17 |
| gl_powerKaleido | 488 | 1306 | 725 | 593 | 484 |
| gl_randomNoisex | 8 | 22 | 12 | 10 | 8 |
| gl_randomsquares | 24 | 63 | 35 | 29 | 23 |
| gl_ripple | 36 | 96 | 53 | 44 | 35 |
| gl_rotateTransition | 34 | 89 | 49 | 40 | 33 |
| gl_rotate_scale_fade | 59 | 156 | 86 | 71 | 58 |
| gl_squareswire | 49 | 130 | 72 | 59 | 48 |
| gl_static_wipe | 52 | 137 | 76 | 62 | 51 |
| gl_swap | 75 | 200 | 111 | 91 | 74 |
| gl_windowblinds | 21 | 56 | 31 | 25 | 21 |
| gl_windowslice | 16 | 43 | 24 | 19 | 16 |
This shows most GL transitions take around ½ minute for SD and 1½ for HD, however some take considerably longer (off-scale transitions shown in italics).
If easing then these times must be added to the Xfade or GL transition times above.
plot data| Easing | SD 2017–20 |
HD 2017–20 |
HD÷1.8 2020–23 |
HD÷2.2 2023–25 |
HD÷2.7 2025– |
|---|---|---|---|---|---|
| back | 9 | 23 | 13 | 10 | 8 |
| bounce | 14 | 36 | 20 | 16 | 13 |
| circular | 6 | 16 | 9 | 7 | 6 |
| cuberoot | 7 | 17 | 9 | 8 | 6 |
| cubic | 7 | 17 | 9 | 8 | 6 |
| elastic | 13 | 34 | 19 | 16 | 13 |
| exponential | 6 | 16 | 9 | 7 | 6 |
| flipback | 10 | 25 | 14 | 11 | 9 |
| flipelastic | 14 | 35 | 20 | 16 | 13 |
| linear | 3 | 6 | 3 | 3 | 2 |
| quadratic | 4 | 12 | 6 | 5 | 4 |
| quartic | 7 | 17 | 9 | 8 | 6 |
| quintic | 6 | 17 | 9 | 8 | 6 |
| sinusoidal | 5 | 13 | 7 | 6 | 5 |
| squareroot | 5 | 12 | 7 | 5 | 4 |
The custom ffmpeg C code in xfade-easing.h deploys a single pixel iterator for all extended transition functions which in turn operate on all planes at once, threading enabled. Use is made of aligned variables and inlined code snippets as compiler hints for fast SIMD vector and sincos instructions.
Easing is calculated once per frame and presents no discernable performance hit.
This plot combines both Xfade and GL transitions.
plot data| Transition | SD 2017–20 |
HD 2017–20 |
HD÷1.8 2020–23 |
HD÷2.2 2023–25 |
HD÷2.7 2025– |
|---|---|---|---|---|---|
| circleclose | 1 | 1.7 | 0.9 | 0.8 | 0.6 |
| circlecrop | 0.7 | 1.2 | 0.7 | 0.5 | 0.4 |
| circleopen | 1 | 1.7 | 0.9 | 0.8 | 0.6 |
| coverdown | 0.5 | 0.6 | 0.3 | 0.3 | 0.2 |
| coverleft | 0.8 | 1.3 | 0.7 | 0.6 | 0.5 |
| coverright | 0.8 | 1.3 | 0.7 | 0.6 | 0.5 |
| coverup | 0.5 | 0.6 | 0.3 | 0.3 | 0.2 |
| diagbl | 0.8 | 1.4 | 0.8 | 0.6 | 0.5 |
| diagbr | 0.9 | 1.4 | 0.8 | 0.7 | 0.5 |
| diagtl | 0.8 | 1.4 | 0.8 | 0.6 | 0.5 |
| diagtr | 0.9 | 1.5 | 0.8 | 0.7 | 0.5 |
| dissolve | 1 | 1.9 | 1.1 | 0.9 | 0.7 |
| distance | 1 | 2 | 1.1 | 0.9 | 0.7 |
| fade | 0.5 | 0.6 | 0.3 | 0.3 | 0.2 |
| fadeblack | 0.6 | 0.7 | 0.4 | 0.3 | 0.3 |
| fadefast | 2 | 4.6 | 2.6 | 2.1 | 1.7 |
| fadegrays | 1 | 1.8 | 1 | 0.8 | 0.7 |
| fadeslow | 2 | 4.7 | 2.6 | 2.1 | 1.7 |
| fadewhite | 0.6 | 0.7 | 0.4 | 0.3 | 0.3 |
| hblur | 0.7 | 1.2 | 0.7 | 0.5 | 0.4 |
| hlslice | 0.8 | 1.4 | 0.8 | 0.6 | 0.5 |
| hlwind | 0.9 | 1.5 | 0.8 | 0.7 | 0.6 |
| horzclose | 0.8 | 1.2 | 0.7 | 0.5 | 0.4 |
| horzopen | 0.8 | 1.2 | 0.7 | 0.5 | 0.4 |
| hrslice | 0.9 | 1.4 | 0.8 | 0.6 | 0.5 |
| hrwind | 0.9 | 1.5 | 0.8 | 0.7 | 0.5 |
| pixelize | 0.8 | 1.4 | 0.8 | 0.6 | 0.5 |
| radial | 1.1 | 2 | 1.1 | 0.9 | 0.8 |
| rectcrop | 0.6 | 0.8 | 0.4 | 0.3 | 0.3 |
| revealdown | 0.5 | 0.6 | 0.3 | 0.3 | 0.2 |
| revealleft | 0.8 | 1.2 | 0.6 | 0.5 | 0.4 |
| revealright | 0.8 | 1.2 | 0.7 | 0.5 | 0.4 |
| revealup | 0.5 | 0.6 | 0.3 | 0.3 | 0.2 |
| slidedown | 0.5 | 0.6 | 0.3 | 0.3 | 0.2 |
| slideleft | 0.8 | 1.2 | 0.6 | 0.5 | 0.4 |
| slideright | 0.7 | 1.1 | 0.6 | 0.5 | 0.4 |
| slideup | 0.5 | 0.6 | 0.3 | 0.3 | 0.2 |
| smoothdown | 0.8 | 1.2 | 0.7 | 0.5 | 0.4 |
| smoothleft | 0.8 | 1.4 | 0.8 | 0.6 | 0.5 |
| smoothright | 0.8 | 1.4 | 0.8 | 0.6 | 0.5 |
| smoothup | 0.8 | 1.2 | 0.7 | 0.5 | 0.4 |
| squeezeh | 0.5 | 0.6 | 0.3 | 0.2 | 0.2 |
| squeezev | 0.7 | 1.2 | 0.6 | 0.5 | 0.4 |
| vdslice | 0.7 | 1.1 | 0.6 | 0.5 | 0.4 |
| vdwind | 1.2 | 2.5 | 1.4 | 1.1 | 0.9 |
| vertclose | 0.8 | 1.3 | 0.7 | 0.6 | 0.5 |
| vertopen | 0.8 | 1.4 | 0.8 | 0.6 | 0.5 |
| vuslice | 0.8 | 1.2 | 0.6 | 0.5 | 0.4 |
| vuwind | 1.2 | 2.5 | 1.4 | 1.1 | 0.9 |
| wipebl | 0.5 | 0.7 | 0.4 | 0.3 | 0.2 |
| wipebr | 0.5 | 0.7 | 0.4 | 0.3 | 0.2 |
| wipedown | 0.5 | 0.5 | 0.3 | 0.2 | 0.2 |
| wipeleft | 0.5 | 0.6 | 0.3 | 0.3 | 0.2 |
| wiperight | 0.5 | 0.7 | 0.4 | 0.3 | 0.2 |
| wipetl | 0.5 | 0.7 | 0.4 | 0.3 | 0.2 |
| wipetr | 0.5 | 0.7 | 0.4 | 0.3 | 0.2 |
| wipeup | 0.5 | 0.5 | 0.3 | 0.2 | 0.2 |
| zoomin | 0.9 | 1.6 | 0.9 | 0.7 | 0.6 |
| gl_Bars | 1.3 | 2.8 | 1.5 | 1.2 | 1 |
| gl_BookFlip | 1.3 | 2.7 | 1.5 | 1.2 | 1 |
| gl_Bounce | 1.6 | 3.5 | 2 | 1.6 | 1.3 |
| gl_BowTie | 1.4 | 3.3 | 1.8 | 1.5 | 1.2 |
| gl_ButterflyWaveScrawler | 5 | 14.5 | 8.1 | 6.6 | 5.4 |
| gl_CornerVanish | 1.1 | 2.1 | 1.2 | 1 | 0.8 |
| gl_CrazyParametricFun | 2.8 | 7.1 | 4 | 3.2 | 2.6 |
| gl_CrossOut | 1.1 | 2.2 | 1.2 | 1 | 0.8 |
| gl_CrossZoom | 42.8 | 122.9 | 68.3 | 55.9 | 45.5 |
| gl_Diamond | 1.1 | 2.2 | 1.2 | 1 | 0.8 |
| gl_DirectionalScaled | 2.5 | 6 | 3.3 | 2.7 | 2.2 |
| gl_DoubleDiamond | 1.1 | 2.3 | 1.3 | 1 | 0.8 |
| gl_Dreamy | 2.4 | 5.9 | 3.3 | 2.7 | 2.2 |
| gl_EdgeTransition | 9.5 | 30.5 | 17 | 13.9 | 11.3 |
| gl_Exponential_Swish | 5 | 14.2 | 7.9 | 6.5 | 5.3 |
| gl_FanIn | 1.3 | 2.6 | 1.5 | 1.2 | 1 |
| gl_FanOut | 1.3 | 2.7 | 1.5 | 1.2 | 1 |
| gl_FanUp | 1.2 | 2.6 | 1.4 | 1.2 | 1 |
| gl_Flower | 1.8 | 3.9 | 2.2 | 1.8 | 1.5 |
| gl_GridFlip | 2.1 | 4.9 | 2.7 | 2.2 | 1.8 |
| gl_InvertedPageCurl | 1.9 | 4.3 | 2.4 | 1.9 | 1.6 |
| gl_LinearBlur | 18.9 | 58.9 | 32.7 | 26.8 | 21.8 |
| gl_Lissajous_Tiles | 84.5 | 230.9 | 128.3 | 104.9 | 85.5 |
| gl_Mosaic | 2.8 | 6.9 | 3.8 | 3.1 | 2.6 |
| gl_PolkaDotsCurtain | 1.4 | 3 | 1.7 | 1.4 | 1.1 |
| gl_Rolls | 1.5 | 3.3 | 1.8 | 1.5 | 1.2 |
| gl_RotateScaleVanish | 1.9 | 4.5 | 2.5 | 2 | 1.7 |
| gl_SimpleBookCurl | 2.2 | 5.3 | 2.9 | 2.4 | 2 |
| gl_SimplePageCurl | 1.7 | 3.8 | 2.1 | 1.7 | 1.4 |
| gl_Slides | 1.2 | 2.5 | 1.4 | 1.1 | 0.9 |
| gl_StageCurtains | 1.9 | 4.6 | 2.5 | 2.1 | 1.7 |
| gl_StarWipe | 2.3 | 5.7 | 3.1 | 2.6 | 2.1 |
| gl_StereoViewer | 2.3 | 5.7 | 3.2 | 2.6 | 2.1 |
| gl_Stripe_Wipe | 2.7 | 6.7 | 3.7 | 3 | 2.5 |
| gl_Swirl | 3 | 7.3 | 4.1 | 3.3 | 2.7 |
| gl_WaterDrop | 1.9 | 4.6 | 2.5 | 2.1 | 1.7 |
| gl_angular | 1.5 | 3.2 | 1.8 | 1.5 | 1.2 |
| gl_blend | 1.8 | 4.2 | 2.3 | 1.9 | 1.6 |
| gl_cannabisleaf | 3.2 | 8.3 | 4.6 | 3.8 | 3.1 |
| gl_chessboard | 1.1 | 2.3 | 1.3 | 1 | 0.8 |
| gl_crosshatch | 2.1 | 4.8 | 2.6 | 2.2 | 1.8 |
| gl_crosswarp | 1.9 | 4.5 | 2.5 | 2 | 1.7 |
| gl_cube | 1.8 | 4 | 2.2 | 1.8 | 1.5 |
| gl_directionalwarp | 2.6 | 6.4 | 3.5 | 2.9 | 2.4 |
| gl_doorway | 1.6 | 3.5 | 1.9 | 1.6 | 1.3 |
| gl_fadecolor | 1.4 | 3.1 | 1.7 | 1.4 | 1.1 |
| gl_heart | 1.1 | 2.2 | 1.2 | 1 | 0.8 |
| gl_hexagonalize | 2.4 | 5.6 | 3.1 | 2.5 | 2.1 |
| gl_kaleidoscope | 5.3 | 14.8 | 8.2 | 6.7 | 5.5 |
| gl_morph | 1.8 | 4.3 | 2.4 | 1.9 | 1.6 |
| gl_perlin | 2.9 | 6.9 | 3.8 | 3.1 | 2.5 |
| gl_pinwheel | 1.5 | 3.3 | 1.9 | 1.5 | 1.2 |
| gl_polar_function | 1.9 | 4.4 | 2.5 | 2 | 1.6 |
| gl_powerKaleido | 6 | 17.4 | 9.7 | 7.9 | 6.5 |
| gl_randomNoisex | 1.7 | 3.9 | 2.1 | 1.8 | 1.4 |
| gl_randomsquares | 2 | 4.5 | 2.5 | 2.1 | 1.7 |
| gl_ripple | 2.4 | 5.7 | 3.2 | 2.6 | 2.1 |
| gl_rotateTransition | 2.2 | 5.3 | 2.9 | 2.4 | 2 |
| gl_rotate_scale_fade | 2.5 | 6.2 | 3.4 | 2.8 | 2.3 |
| gl_squareswire | 1.3 | 2.8 | 1.6 | 1.3 | 1 |
| gl_static_wipe | 2 | 4.6 | 2.5 | 2.1 | 1.7 |
| gl_swap | 1.6 | 3.5 | 2 | 1.6 | 1.3 |
| gl_windowblinds | 1.3 | 2.7 | 1.5 | 1.2 | 1 |
| gl_windowslice | 1.2 | 2.4 | 1.3 | 1.1 | 0.9 |
This shows most transitions take around 1 second for SD and 2½ for HD but again some take considerably longer (off-scale transitions shown in italics).
Other, faster ways to use GL Transitions with FFmpeg are:
xfade-easing.sh is a Bash 4 shell wrapper for ffmpeg. It can:
FFmpeg Xfade easing and extensions version 3.6.6 by Raymond Luckhurst, https://scriptit.uk
Wrapper script to render eased Xfade/GLSL transitions natively or with custom expressions.
Generates easing and transition expressions for xfade and for easing other filters.
Also creates easing graphs, demo videos, presentations and slideshows.
See https://github.com/scriptituk/xfade-easing
Usage: xfade-easing.sh [options] [image/video inputs]
Options:
-t transition name and arguments, if any (default: fade); use -L for list
args in parenthesis as CSV, e.g.: gl_perlin(5,0.1) (both variants)
or key=value pairs, e.g.: gl_perlin(smoothness=0.1, scale=5) (custom ffmpeg only)
use gl_random to cycle through shuffled transitions ported from GLSL
-e easing function and arguments, if any (default: linear)
CSS args in parenthesis as CSV, e.g.: cubic-bezier(0.17,0.67,0.83,0.67)
-b reverse transition and/or easing effect (custom ffmpeg only) (default: 0)
1 reverses the inputs and transition effect; 2 reverses the easing; 3 reverses both
-x expr output filename (default: no expr), accepts expansions, - for stdout
-a append to expr output file
-s expr output format string with text expansion (default: '%x')
%f expands to pixel format, %F to format in upper case
%e expands to the easing name
%t expands to the transition name
%E, %T upper case expansions of %e, %t
%c expands to the CSS easing arguments
%a expands to the GL transition arguments; %A to the default arguments (if any)
%x expands to the generated expr, condensed, intended for inline filterchains
%X uncondensed version of %x, intended for -/filter_complex script files
%p expands to the progress easing expression, condensed, for inline filterchains
%g expands to the generic easing expression (for other filters), condensed
%z expands to the eased transition expression only, condensed
for the uneased transition expression only, omit -e option and use %x or %X
%P, %G, %Z, uncondensed versions of %p, %g, %z, for -/filter_complex script files
%n inserts a newline
-p easing plot filename (default: no plot), accepts expansions
formats: gif, jpg, png, svg, pdf, eps, html <canvas>, from file extension
-m multiple easings to plot on one graph (default: the -e easing)
CSV easings with optional legend prefix, e.g. in=cubic-in,out=cubic-out,in-out=cubic
-q plot title (default: easing name, or Easings for multiple plots)
-c canvas size for easing plot (default: 640x480, scaled to inches for PDF/EPS)
format: WxH; omitting W or H keeps aspect ratio, e.g. -z x300 scales W
-v video output filename (default: no video), accepts expansions
formats: mkv (FFV1), mp4 (H264), webm (VP9), raw
animated formats: webp (VP8), png (APNG), gif, y4m (yuv4mpeg)
inferred from filename; if filename is - then format is the null muxer (no output)
if -f format has alpha then mkv,webm,raw generate transparent video output
if gifski is installed then gifs will be optimised
raw decode: ffmpeg -f rawvideo -pixel_format f -framerate r -video_size s -i ...
-o additional ffmpeg options, e.g. -o '-movflags +faststart' for MP4 Faststart
-r video framerate (default: 25fps)
-f pixel format (default: rgb24): use ffmpeg -pix_fmts for list
-z video size (default: input 1 size)
format: WxH; omitting W or H keeps aspect ratio, e.g. -z 400x scales H
-d video transition duration (default: 3s, minimum: 0) (see note for -l)
-i time between video transitions (default: 1s, minimum: 0) (see note for -l)
-l video length (default: 5s)
note: options -d, -i, -l are interdependent: l=ni+(n-1)d for n inputs
given -i & -l, d is calculated; else given -l, i is calculated; else l is calculated
-j allow input videos to play within transitions (default: no)
normally videos only play during the -i time but this sets them playing throughout
-n show effect name on video as text (requires the libfreetype library)
-u video text font size multiplier (default: 1.0)
-k video stack orientation,gap,colour,padding (default: ,0,white,0), e.g. h,2,red,1
stacks uneased and eased videos horizontally (h), vertically (v) or auto (a)
auto selects the orientation that displays easing to best effect
also stacks transitions with default and custom parameters, eased or not
videos are only stacked if they are different (nonlinear-eased or customised)
unstacked videos can be padded using orientation=1, e.g. 1,0,blue,5
-L list all transitions and easings
-H show this usage text
-V show this script version
-X use custom expressions, not the xfade API that supports xfade-easing natively
by default native support is detected automatically using ffmpeg --help filter=xfade
the native API adds easing and reverse options and runs much faster
e.g. xfade=duration=4:offset=1:easing=quintic-out:transition=wiperight
e.g. xfade=duration=5:offset=2:easing='cubic-bezier(.17,.67,.83,.67)' \
:transition='gl_swap(depth=5,reflection=0.7,perspective=0.6)' (see repo README)
-I set ffmpeg loglevel to info for -v (default: warning), also dumps ffmpeg command
-D dump debug messages to stderr and set ffmpeg loglevel to debug for -v
-P log xfade progress percentage using custom expression print() function (implies -I)
-T temporary file directory (default: /tmp)
-K keep temporary files if temporary directory is not /tmp
Notes:
1. this script requires Bash 4 (2009), ffmpeg, ffprobe, gawk, gsed, seq
also gnuplot for plots, gifski for optimised animated gifs
2. use ffmpeg option -filter_complex_threads 1 (slower) because xfade expression
vars used by st() & ld() are shared across slices, therefore not thread-safe
(the custom ffmpeg build works without -filter_complex_threads 1)
3. CSS easings are supported in the custom ffmpeg build but not as custom expressions
4. certain xfade transitions are not implemented as custom expressions because
they perform aggregation (distance, hblur)
5. many GLSL transitions are also ported, some of which take customisation parameters
to override defaults append parameters in parenthesis (see -t option)
6. certain GLSL transitions are only available in the custom ffmpeg build
7. many transitions do not lend themselves well to easing, others have built-in easing
easings that overshoot (back & elastic) may cause weird effects
Expr code is generated using the -x option and customised with the -s,-a,-f options.
$expr['rectcrop+exponential-out'] = '
st(0, if(lte(P, 0), 0, 2^(10 * P - 10)))
;
st(1, abs(ld(0) - 0.5));
if(lt(abs(X - W / 2), ld(1) * W) * lt(abs(Y - H / 2), ld(1) * H),
if(lt(ld(0), 0.5), B, A),
ifnot(3-PLANE, 255)
)';expr='
st(1, 5);
st(2, X / W - 0.5);
st(3, 0.5 - Y / H);
st(4, atan2(ld(3), ld(2)) - PI / 2);
st(4, cos(ld(1) * ld(4)) / 4 + 1);
st(1, hypot(ld(2), ld(3)));
if(gt(ld(1), ld(4) * (1 - P)), A, B)'Plots are generated using the -p option and customised with the -m,-q,-c options.
Plot data is logged using the print function of the ffmpeg expression evaluator for the first plane and first pixel as xfade progress P goes from 1 to 0 at 100fps. Specifically, if the xfade transition option is custom and an easing option is provided, then P is the constrained eased progress (0–1) and variables ld(0) and ld(1) contain preloaded plot data: ld(0) is the uneased progress (linear P value) and ld(1) is the unclipped eased progress (may overshoot). Plotting y=ld(1) against x=ld(0) with gnuplot creates the graphs.
The plots above in Standard easings show test plots for all standard easings and all three modes (in, out and in-out).
Videos are generated using the -v option and customised with the -b,-r,-f,-z,-d,-i,-l,-j,-n,-u,-k,-o options.
Note
all transition effect demos on this page are animated GIFs regardless of the commands shown
Input media is serialised according to the expression $L=NI+(N-1)D$ where $L$ is the total video length (option -l); $I$ is the individual display time (option -i); $D$ is the transition duration (option -d); $N$ is the number of inputs. Transition offsets are spaced accordingly. Depending on option -j and the input media length, pre and post padding is added by frame cloning to ensure enough frames are available for transition processing. See Usage for the precedence of options -l, -i, -d.
xfade-easing.sh -t hlwind -e quintic-in -v windy.gif
creates an animated GIF image of the hlwind transition with quintic-in easing using default built-in images

xfade-easing.sh -t fadeblack -e circular -v maths.mp4 dot.png cross.png
creates a MP4 video of the fadeblack transition with circular easing using specified inputs
(credit: Math & Science Tutor)

xfade-easing.sh -t coverdown -e bounce-out -v %t-%e.mp4 wallace.png shaun.png
creates a video of the coverdown transition with bounce-out easing using expansion specifiers for the output file name

xfade-easing.sh -t 'gl_polar_function(25)' -v paradise.mkv -n -u 1.2 islands.png rainbow.png
creates a lossless (FFV1) video (e.g. for further processing) of an uneased polar_function GL transition with 25 segments annotated in enlarged text

xfade-easing.sh -t 'gl_Lissajous_Tiles(16,20,0.3,9,3,1,0.8,3,2,Lavender)' -e quadratic -v lissajous.mp4 titian.png kandinsky.png
creates a stunning Lissajous effect quadratic-eased against a Lavender background demonstrating extensive use of transition parameters

xfade-easing.sh -t 'gl_cube(,,,,-23)' -e 'cubic-bezier(0.5,0.9,0.5,0.1)' -v cube.mp4 -b 1 BBC_Test_Card_C.png BBC_Test_Card_J.png
creates a reversed cube GL transition against a texture background with cubic-bezier easing that slows down the middle movement

xfade-easing.sh -t 'gl_StageCurtains(Purple,30,0.05)' -v stage.mp4 -i 1 -d 5 hamlet-players.png hamlet-yorick.png hamlet-fight.png
creates a slow stage curtain effect GL transition with default linear easing showcasing three scenes from Hamlet

xfade-easing.sh -t 'gl_angular(270,1)' -e exponential -v multiple.mp4 -n -k h -l 20 street.png road.png flowers.png bird.png barley.png
creates a video of the angular GL transition with parameter startingAngle=270 (south) and clockwise=1 (an added parameter) for 5 inputs with fast exponential easing

xfade-easing.sh -t gl_BookFlip -e quartic-out -v book.mp4 -f gray -z 248x -n -k h,2,black,1 alice12.png alice34.png
creates a simple greyscale page turn with quartic-out easing for a more realistic effect.

xfade-easing.sh -t circlecrop -e sinusoidal -v home-away.mp4 -l 10 -d 8 -z 246x -k h,4,LightSkyBlue,2 -n phone.png beach.png
creates a 10s video with a slow 8s circlecrop xfade transition with sinusoidal easing, horizontally stacked with a 4px LightSkyBlue gap (see Color) and 2px padding

xfade-easing.sh -t gl_InvertedPageCurl -e cubic-in -v score.mp4 -f gray -i 2 -d 3 -z 480x -k 1,0,0xD8D8D8,10 fugue1.png fugue2.png fugue3.png
a 3s page curl effect, static for 2s, with cubic-in easing using greyscale format (-k 1,0,colour,padding creates a border)
🎹 I play this Bach fugue on my YouTube channel digitallegro but the GL InvertedPageCurl there was generated by ffmpeg-concat
xfade-easing.sh -t 'gl_PolkaDotsCurtain(10,0.5,0)' -e 'cubic-bezier(0.4,1.2,0.6,-1.1)' -v life.mp4 -l 7 -d 5 -z 500x -f yuv420p -r 30 balloons.png fruits.png
a GL transition with arguments and cubic-bezier easing, running at 30fps for 7 seconds, processing in YUV (Y'CbCr) colour space throughout

| Back | FazBrowse Home | New Git URL |