Color Modes

The FX Special Effects Image Operator

The FX Special Effects Image OperatorThe Anatomy of an FX Expression

The FX special effects image operator applies a mathematical expression to each pixel channel of an image. The FX expression language provides a powerful and flexible way to manipulate images, allowing you to perform a wide range of operations and transformations on your images. Use FX to:

  • create canvases, gradients, mathematical colormaps
  • move color values between images and channels
  • translate, flip, mirror, rotate, scale, shear and generally distort images
  • merge or composite multiple images together
  • convolve or merge neighboring pixels together
  • generate image metrics or 'fingerprints'

The operator loops through all the pixels of an image and all the channels of each pixel and returns a new image with the results. The expression can reference any image in the image sequence, but only a copy of the first image is returned with suitable updates per your expression.

The expression can be simple:

magick -size 64x64 canvas:black -channel blue -fx "1/2" fx_navy.png

Here, we convert a black to a navy blue image:

    black ==> navy

Or the expression can be complex:

magick rose: \
  -fx "(1.0/(1.0+exp(10.0*(0.5-u)))-0.006693)*1.0092503" \

This expression results in a high contrast version of the source image:

    rose ==> rose-sigmoidal

The expression can include variable assignments. Assignments, in most cases, reduce the complexity of an expression and permit some operations that might not be possible any other way. For example, lets create a radial gradient:

magick -size 70x70 canvas: \
  -fx "Xi=i-w/2; Yj=j-h/2; 1.2*(0.5-hypot(Xi,Yj)/70.0)+0.5" \

The command above returns this image:


This FX expression adds random noise to an image:

magick photo.jpg -fx 'iso=32; rone=rand(); rtwo=rand(); \
  myn=sqrt(-2*ln(rone))*cos(2*Pi*rtwo); myntwo=sqrt(-2*ln(rtwo))* \
  cos(2*Pi*rone); pnoise=sqrt(p)*myn*sqrt(iso)* \
  channel(4.28,3.86,6.68)/255; max(0,p+pnoise)' noisy.png

This FX script utilizes a loop to create a Julia set:

magick -size 400x400 xc:gray -fx " \
  Xi=2.4*i/w-1.2; \
  Yj=2.4*j/h-1.2; \
  for (pixel=0.0, (hypot(Xi,Yj) < 2.0) && (pixel < 1.0), \
    delta=Xi^2-Yj^2; \
    Yj=2.0*Xi*Yj+0.2; \
    Xi=delta+0.4; \
    pixel+=0.00390625 \
  ); \
  pixel == 1.0 ? 0.0 : pixel" \
  \( -size 1x1 xc:white xc:red xc:orange xc:yellow xc:green1 xc:cyan xc:blue \
     xc:blueviolet xc:white -reverse +append -filter Cubic -resize 1024x1! \) \
  -clut -rotate -90 julia-set.png
    Julia Fractals

This FX script prints the first 10 prime numbers:

magick xc:gray -fx " \
  for (prime=2, prime < 30, composite=0; \
    for (nn=2, nn < (prime/2+1), if ((prime % nn) == 0, composite++, ); nn++); \
      if (composite <= 0, debug(prime), ); prime++)" null:

See Using FX, The Special Effects Image Operator for more examples.

The -fx option replaces any image sequence with a clone of the first image updated with the results of the expression. If you prefer to apply the expression to each image in the sequence, use +fx instead.

The FX expression is interpreted in a single-thread, however, it is executed in multiple threads unless the expression includes the debug() function.

The next section discusses the FX expression language.

The Anatomy of an FX Expression

The FX Expression Language

The formal FX expression language is defined here:

integer, floating point, scientific notation (+/- required, e.g. 3.81469e-06), International System number postfixes (.e.g KB, Mib, GB, etc.)
E (Euler's number), Epsilon, Opaque, Phi (golden ratio), Pi, QuantumRange, QuantumScale, Transparent
FX operators (in order of precedence):
^ (power), unary -, *, /, % (modulo), +, -, <<, >>, <, <=, >, >=, +=, -=, *=, /=, %=, <<= , >>=, &=, |=, ++, --, ==, !=, & (bitwise AND), | (bitwise OR), && (logical AND), || (logical OR), ~ (logical NOT), ?: (ternary conditional)
an image offers array storage (e.g. p[-1,-1].r) bounded by its width and height. An image sequence represents multiple arrays (e.g. u.p[0,0].r, v.p[0,0].r). Storage is limited to Quantum values, e.g. [0..65535] for Q16 builds and floating point for HDRI-enabled builds.
math functions:
abs(), acos(), acosh(), airy(), alt(), asin(), asinh(), atan(), atanh(), atan2(), ceil(), clamp(), cos(), cosh(), debug(), drc(), erf(), exp(), floor(), gauss(), gcd(), hypot(), int(), isnan(), j0(), j1(), jinc(), ln(), log(), logtwo(), max(), min(), mod(), not(), pow(), rand(), round(), sign(), sin(), sinc(), sinh(), sqrt(), squish(), tan(), tanh(), trunc()
channel functions:
define up to 5 pixel channels
color names:
red, cyan, black, etc.
color functions:
srgb(), srgba(), rgb(), rgba(), cmyk(), cmyka(), hsl(), hsla(), etc.
color hex values:
#ccc, #cbfed0, #b9e1cc00, etc.
u: first image in list
v: second image in list
s: current image in list (for %[fx:] otherwise = u)
t: index of current image (s) in list
n: number of images in list
i: column offset
j: row offset
p: pixel to use (absolute or relative to current pixel)
w: width of this image
h: height of this image
z: channel depth
r: red value (from RGBA), of a specific or current pixel
g: green
b: blue
a: alpha
o: opacity
c: cyan value of CMYK color of pixel
y: yellow
m: magenta
k: black
all: all channels
this: this channel
intensity: pixel intensity
hue: pixel hue
saturation: pixel saturation
lightness: pixel lightness
luma: pixel luma
page.width: page width
page.height: page height
page.x: page x offset
page.y: page y offset
printsize.x: x printsize
printsize.y: y printsize
resolution.x: x resolution
resolution.y: y resolution
depth: image depth
extent: image extent
minima: image minima
maxima: image maxima
mean: image mean
median: image median
standard_deviation: image standard deviation
kurtosis: image kurtosis
skewness: image skewness (add a channel specifier to compute a statistic for that channel, e.g. depth.r)
do(), for(), while()
image attributes:
s.depth, s.kurtosis, s.maxima, s.mean, s.minima, s.resolution.x, s.resolution.y, s.skewness, s.standard_deviation
user settings:
define Fx symbols as user settings, e.g.
magick ... -set option:wd1 "%[fx:w/2]" -resize "%[fx:wd1-5]" ...

The FX Expression

An FX expression may include any combination of the following:

x ^ y
exponentiation (xy)
( ... )
x * y
x / y
x % y
x + y
x - y
x << y
left shift
x >> y
right shift
x < y
boolean relation, return value 1.0 if x < y, otherwise 0.0
x <= y
boolean relation, return value 1.0 if x <= y, otherwise 0.0
x > y
boolean relation, return value 1.0 if x > y, otherwise 0.0
x >= y
boolean relation, return value 1.0 if x >= y, otherwise 0.0
x == y
boolean relation, return value 1.0 if x == y, otherwise 0.0
x != y
boolean relation, return value 1.0 if x != y, otherwise 0.0
x & y
binary AND
x | y
binary OR
x && y
logical AND connective, return value 1.0 if x > 0 and y > 0, otherwise 0.0
x || y
logical OR connective (inclusive), return value 1.0 if x > 0 or y > 0 (or both), otherwise 0.0
logical NOT operator, return value 1.0 if not x > 0, otherwise 0.0
unary plus, return 1.0*value
unary minus, return -1.0*value
condition ? true-statements : false-statements
ternary conditional expression, return value true-statements if condition != 0, otherwise false-statements
x = y
assignment; single character variables are reserved, instead use 2 or more characters, letter combinations only (e.g. Xi not X1)
x ; y
statement separator
constant (1.618034...)
constant (3.14159265359...)
constant (2.71828...)
constant maximum pixel value (255 for Q8, 65535 for Q16)
constant 1.0/QuantumRange
pixel intensity whose value respects the -intensity option.
pixel hue
pixel saturation
pixel lightness; equivalent to 0.5*max(red,green,blue) + 0.5*min(red,green,blue)
pixel luminance; equivalent to 0.212656*red + 0.715158*green + 0.072186*blue
red, green, blue, etc.
color names
#ccc, #cbfed0, #b9e1cc00, etc.
color hex values
rgb(), rgba(), cmyk(), cmyka(), hsl(), hsla()
color functions
s, t, u, v, n, i, j, w, h, z, r, g, b, a, o, c, y, m, k
absolute value function
arc cosine function
inverse hyperbolic cosine function
Airy function (max=1, min=0); airy(x)=[jinc(x)]2=[2*j1(pi*x)/(pi*x)]2
sign alternation function (return 1.0 if int(x) is even, -1.0 if int(x) is odd)
arc sine function
inverse hyperbolic sine function
arc tangent function
inverse hyperbolic tangent function
arc tangent function of two variables
smallest integral value not less than argument
supports from zero to five arguments, e.g., channel(0.1) sets the first channel to 0.1 and zero's the other channels.
clamp value
cosine function
hyperbolic cosine function
print x (useful for debugging your expression)
do(statements, condition)
iterate while the condition is not equal to 0
dynamic range compression (knee curve); drc(x,y)=(x)/(y*(x-1)+1); -1<y<1
error function
natural exponential function (ex)
largest integral value not greater than argument
for(initialize, condition, statements)
iterate while the condition is not equal to 0
gaussian function; gauss(x)=exp(-x*x/2)/sqrt(2*pi)
greatest common denominator
the square root of x2+y2
if(condition, nonzero-statements, zero-statements)
interpret expression depending on condition
greatest integer function (return greatest integer less than or equal to x)
return 1.0 if x is NAN, 0.0 otherwise
Bessel functions of x of the first kind of order 0
Bessel functions of x of the first kind of order 1
jinc function (max=1, min=-0.1323); jinc(x)=2*j1(pi*x)/(pi**x)
natural logarithm function
logarithm base 10
logarithm base 2
natural logarithm
max(x, y)
maximum of x and y
min(x, y)
minimum of x and y
mod(x, y)
floating-point remainder function
return 1.0 if x is zero, 0.0 otherwise
power function (xy)
value uniformly distributed over the interval [0.0, 1.0) with a 2 to the 128th-1 period
round to integral value, regardless of rounding direction
return -1.0 if x is less than 0.0 otherwise 1.0
sine function
sinc function (max=1, min=-0.21); sinc(x)=sin(pi*x)/(pi*x)
squish function; squish(x)=1.0/(1.0+exp(-x))
hyperbolic sine function
square root function
tangent function
hyperbolic tangent function
round to integer, towards zero
while(condition, statements)
iterate while the condition is not equal to 0
image.depth, image.kurtosis, image.maxima, image.mean, image.median, image.minima, image.resolution.x, image.resolution.y, image.skewness, image.standard_deviation
image attributes

The expression semantics include these rules:

  • symbols are case insensitive
  • only one ternary conditional (e.g. x ? y : z) per statement
  • statements are assignments or the final expression to return
  • an assignment starts a statement, it is not an operator
  • single character variables are reserved. Assignments to reserved built-ins throw an exception; e.g. r=3.0; r returns Attempted assignment to non-UserSymbol 'r' at '3.0'.
  • unary operators have a lower priority than binary operators, that is, the unary minus (negation) has lower precedence than exponentiation, so -3^2 is interpreted as -(3^2) = -9. Use parentheses to clarify your intent (e.g. (-3)^2 = 9).
  • care must be exercised when using the slash ('/') symbol. The string of characters 1/2x is interpreted as (1/2)x. The contrary interpretation should be written explicitly as 1/(2x). Again, the use of parentheses helps clarify the meaning and should be used whenever there is any chance of misinterpretation.
  • As -- is the variable decrement operator, use parenthesis to subtract a negative number, e.g, -4-(-5).

Source Images

The symbols u and v refer to the first and second images, respectively, in the current image sequence. Refer to a particular image in a sequence by appending its index to any image reference (usually u), with a zero index for the beginning of the sequence. A negative index counts from the end. For example, u[0] is the first image in the sequence, u[2] is the third, u[-1] is the last image, and u[t] is the current image. The current image can also be referenced by s. If the sequence number exceeds the length of the sequence, the count is wrapped. Thus in a 3-image sequence, u[-1], u[2], and u[5] all refer to the same (third) image.

As an example, we form an image by averaging the first image and third images (the second (index 1) image is ignored and just junked):

magick image1.jpg image2.jpg image3.jpg -fx "(u+u[2])/2" image.jpg

By default, the image to which p, r, g, b, a, etc., are applied is the current image s in the image list. This is equivalent to u except when used in an escape sequence %[fx:...].

It is important to note the special role played by the first image. This is the only image in the image sequence that is modified, other images are used only for their data. As an illustrative example, consider the following, and note that the setting -channel red instructs -fx to modify only the green channel; nothing in the red or blue channels will change. It is instructive to ponder why the result is not symmetric.

magick logo: -flop logo: -resize "20%" -channel green -fx "(u+v)/2" image.jpg
    logo-sm-flop.png logo-sm.png ==> logo-sm-fx.png

Accessing Pixels

All color values are normalized to the range of 0.0 to 1.0. The alpha channel ranges from 0.0 (fully transparent) to 1.0 (fully opaque).

The pixels are processed one at a time, but a different pixel of an image can be specified using a pixel index represented by p. For example,

p[-1].g      green value of pixel to the immediate left of the current pixel
p[-1,-1].r   red value of the pixel diagonally left and up from current pixel

To specify an absolute position, use braces, rather than brackets.

p{0,0}.r     red value of the pixel in the upper left corner of the image
p{12,34}.b   blue pixel value at column number 12, row 34 of the image

Integer values of the position retrieve the color of the pixel referenced, while non-integer position values return a blended color according to the current -interpolate setting.

A position outside the boundary of the image retrieves a value dictated by the -virtual-pixel option setting.

Specify u.r to specify the red channel of the current image. If you don't specify a channel qualifier, you get the current channel. Use mean.this to set the output channel to the mean of just the input channel. Use mean.all to set the overall mean of the input channels.

Apply an Expression to Select Image Channels

Use the -channel setting to specify the output channel of the result. If no output channel is given, the result is set over all channels except the opacity channel. For example, to replace the red channel of alpha.png with the average of the green channels from the images alpha.png and beta.png, use:

magick alpha.png beta.png -channel red -fx "(u.g+v.g)/2" gamma.png


The -fx operator evaluates the given expression for each channel (set by -channel) of each pixel in the first image (u) in the sequence. The computed values are temporarily stored in a copy (clone) of that first image until all the pixels have been processed, after which this single new image replaces the list of images in the current image sequence. As such, in the previous example the updated version of alpha.png replaces both of the original images, alpha.png and beta.png, before being saved as gamma.png.

The current image s is set to the first image in the sequence (u), and t to its index, 0. The symbols i and j reference the current pixel being processed.

For use with -format, the value-escape %[fx:] is evaluated just once for each image in the current image sequence. As each image in the sequence is being evaluated, s and t successively refer to the current image and its index, while i and j are set to zero, and the current channel set to red (-channel is ignored). An example:

$ magick canvas:'rgb(25%,50%,75%)' rose: -colorspace gray  \
  -format 'Red channel of NW corner of image #%[fx:t] is %[fx:s]\n' info:
Red channel of NW corner of image #0 is 0.464883
Red channel of NW corner of image #1 is 0.184582

Here we use the image indexes to rotate each image differently, and use -set with the image index to set a different pause delay on the first image in the animation:

magick rose: -duplicate 29 -virtual-pixel Gray -distort SRT '%[fx:360.0*t/n]' \
  -set delay '%[fx:t == 0 ? 240 : 10]' -loop 0 rose.gif

This example tests the difference between two images, measured by RMSE. If the difference is greater than 0.1 it returns 1; otherwise it returns 0:

magick water.png reference.png -metric RMSE -compare -format "%[fx:%[distortion]>0.1]" info:

The color-escape %[pixel:] or %[hex:] is evaluated once per image and per color channel in that image (-channel is ignored), The values generated are then converted into a color string (a named color or hex color value). The symbols i and j are set to zero, and s and t refer to each successively current image and index.