Easily create and animate math expressions in manim. Reactive-manim reduces the programming overhead of creating animations that animate individual parts of an equation, through the use of reusable math components that are reactive to changes in state.
https://philip-murray.github.io/reactive-manim
Reusable Math Components
Component Composition
Automatic Animation
➡️ Examples ⬅️ Click for more examples.
pip install reactive-manimreactive_manim overwrites the definition for MathTex, it must be imported after importing manim.
from manim import *
from reactive_manim import *Note: The stuttering in the animations is a result of using gif files for the README. It does not occur when rendering mp4 files.
class ExponentScene(Scene):
def construct(self):
term1, term2 = Term("2"), Term("2", "4")
tex = MathTex(term1, Parentheses(term2))
self.add(tex).wait(1)
term2.term.save_y()
term1.term.target_id = term2.term.id
term2.superscript.set_tex_string("5")
tex.terms = [ term2 ]
term2.term.restore_y()
self.play(TransformInStages.progress(tex, lag_ratio=0.4))
self.wait(1)
class NaturalLogScene(Scene):
def construct(self):
tex = MathTex("a", "=", "b")
self.add(tex).wait(0.5)
tex[0] = Term("e", tex[0])
tex[2] = Term("e", tex[2])
self.play(TransformInStages.progress(tex, lag_ratio=0.6))
tex[0] = Function(r"\ln", tex[0])
tex[2] = Function(r"\ln", tex[2])
self.play(TransformInStages.progress(tex, lag_ratio=0.6))
tex[0] = MathTex(tex[0].input.superscript.pop(), tex[0])
tex[2] = MathTex(tex[2].input.superscript.pop(), tex[2])
self.play(TransformInStages.progress(tex))
tex[0] = tex[0][0]
tex[2] = tex[2][0]
self.play(TransformInStages.progress(tex, lag_ratio=0.8))
class CasesScene(Scene):
def construct(self):
tex = MathTex("f(x) =", MathCases(
CaseLine(r"b, ", r"\text{even}(x)"),
CaseLine(r"\neg b,", r"\text{odd}(x)"),
CaseLine(r"?,", r"\text{otherwise}"),
))
self.add(tex).wait(1)
cases = tex[1]
cases[1].condition.set_color(ORANGE)
cases[2].output.set_color(ORANGE)
self.play(TransformInStages.progress(tex))
cases[1].condition = cases[2].condition
cases.lines = [ cases[0], cases[1] ]
self.play(TransformInStages.progress(tex, lag_ratio=0.4, run_time=1.5))
self.wait(2)
class MatrixScene(Scene):
def construct(self):
y0, y1 = Term("y", subscript=0), Term("y", subscript=1)
x0, x1 = Term("x", subscript=0), Term("x", subscript=1)
w00, w01 = Term("w", subscript="00"), Term("w", subscript="01")
w10, w11 = Term("w", subscript="10"), Term("w", subscript="11")
tex1 = MathTex(y0, "=", w00, x0, "+", w01, x1)
tex2 = MathTex(y1, "=", w10, x0, "+", w11, x1)
group = DGroup(tex1, tex2).arrange(DOWN).shift(UP)
self.add(group).wait(2)
tex3 = MathTex(
MathMatrix([
[ y0 ],
[ y1 ]
]),
"=",
MathMatrix([
[ w00, w01 ],
[ w10, w11 ]
]),
MathMatrix([
[ x0 ],
[ x1 ]
])
)
tex3.shift(DOWN)
self.play(TransformInStages.from_copy(group, tex3[0], lag_ratio=0.4, run_time=2.5))
self.play(TransformInStages.from_copy(group, tex3[1], lag_ratio=0.4, run_time=1.4))
self.play(TransformInStages.from_copy(group, tex3[2], lag_ratio=0.4, run_time=2.5))
self.play(TransformInStages.from_copy(group, tex3[3] - x0, lag_ratio=0.4, run_time=2.5))
self.play(TransformInStages.from_copy(group, x0, lag_ratio=0.4, run_time=2.5))
self.wait(2)This example makes careful use of auto-disconnect.
class EquationVariablesScene(Scene):
def construct(self):
y_pred = MathString("6.75")
y_act = MathString("9.0")
tex = MathTex(r"E = \frac{1}{2}(", "y", "-", "a", ")^2")
y_pred.move_to(np.array([ -1.0, 1.25, 0 ]))
y_act.move_to(np.array([ 1.0, 1.25, 0 ]))
tex.move_to(np.array([ 0, -0.5, 0 ]))
self.add(y_pred, y_act, tex).wait(1)
tex[1] = y_pred.copy()
anim = TransformInStages.progress(tex, lag_ratio=0, run_time=1.5)
anim.intercept(y_pred).set_source(y_pred)
self.play(anim)
self.wait(1)
tex[0].save_center()
tex[3] = y_act.copy()
tex[0].restore_center()
anim = TransformInStages.progress(tex, lag_ratio=0, run_time=1.5)
anim.intercept(y_act).set_source(y_act)
self.play(anim)
self.wait(1)This uses animation.intercept(), which is not discussed in the guide section because it is still in development.
However, the same result could be accomplished without animation.intercept() using the from_copy() constructor of TransformInStages.
class MathListScene(Scene):
def construct(self):
tex = MathList("4", "7", "1", "2")
self.add(tex).wait(1)
tex.remove(tex[1])
self.play(TransformInStages.progress(tex, lag_ratio=0.8))
self.wait(1)
tex.append("3")
self.play(TransformInStages.progress(tex, lag_ratio=0.8))
self.wait(1)
four = tex[0]
tex.remove(four)
tex.append(four)
animation = TransformInStages.progress(tex, lag_ratio=0.5, track_run_time=1.5)
animation.intercept(four).set_animation(
lambda source, target: Transform(source, target, path_arc=(3*PI)/4)
)
self.play(animation)This uses MathList, which is not discussed in the guide section because it is still in development.
This uses animation.intercept(), which is not discussed in the guide section because it is still in development.
class QuadraticScene(Scene):
def construct(self):
a1, p1, b1 = MathTex("a", "+", "b")
tex1 = MathTex(Parentheses([ a1, p1, b1 ]), Parentheses([ "a", "-", "b" ])).shift(UP)
self.add(tex1).wait(1)
tex1[1].set_color(BLUE)
self.play(TransformInStages.progress(tex1))
paren_2 = tex1[1]
tex2 = MathTex([ a1, paren_2 ], [ [ p1, b1 ], paren_2 ])
self.play(TransformInStages.from_copy(tex1, tex2[0], run_time=1.5))
self.play(TransformInStages.from_copy(tex1, tex2[1], run_time=1.5))
self.wait(1)
a2, m2, b2 = tex2[0][1].inner
a3, m3, b3 = tex2[1][1].inner
part1 = Term(a1, "2")
a2.target_id = a1.id
part2 = MathTex(m2, a1.clone(), b2)
part3 = MathTex(p1, a3, b1.clone())
part4 = MathTex(m3, Term(b2, "2"))
b3.target_id = b2.id
tex3 = MathTex(part1, [ part2, part3 ], part4).shift(DOWN).set_color(WHITE)
self.play(TransformInStages.from_copy(tex2, part1, lag_ratio=0.4, track_run_time=1.5))
self.play(TransformInStages.from_copy(tex2, part2, lag_ratio=0.4, track_run_time=1.5))
self.play(TransformInStages.from_copy(tex2, part3, lag_ratio=0.4, track_run_time=1.5))
self.play(TransformInStages.from_copy(tex2, part4, lag_ratio=0.4, track_run_time=1.5))
point_mobject = tex3[1].get_point_dynamic_mobject().shift(DOWN*0.05)
tex3.terms = [ tex3[0], tex3[2] ]
anim = TransformInStages.progress(tex3, lag_ratio=0)
anim.intercept(point_mobject).set_target(point_mobject)
self.play(anim)
self.wait(1)
Mobjects including tex, x, and y are components of an mobject-graph where tex is the root. Methods that change a component's state, for example z.set_tex_string(tex_string) or tex.remove(term), trigger a rerender that propagates to the root. The root's position is unaffected by state changes.
I use the phrases mobject and mobject-graph interchangeably, because in certain contexts, an mobject can be thought to implicitly include its children. We will observe in the section on partial progression, that progress(tex) is interpreted as a subgraph of tex.graph that includes tex.id and the IDs of its recursive descendants.
In nested constructions, a string can be used as a shorthand syntax to denote a MathString.
The tex_string of a MathTex is generated by concatenating the tex_strings of its child components or "terms". The list of terms can be changed programmatically, via
remove(term)append(term)insert(index: int, term)__setitem__(self, index, term)- tex[0] = term
@terms.setter- tex.terms = [ term1, term2 ]
MathTex does not currently support kwargs. However, default kwargs such as tex_template can be set using SingleStringMathTex.set_default(tex_template=...)
Every mobject derived from DynamicMobject has an ID. An mobject-graph-transform is an animation between between a source_graph and target_graph, guided by IDs. For each ID, an mobject representing that ID is interpolated between source_graph[id] and target_graph[id], if both exist.
In this example:
-
The
source_graphis taken atscene.add(tex)- Refreshed by
scene.wait() - Refreshed by changes in state following a progress transform.
- Refreshed by
-
The
target_graphis taken atTransformInStages.progress(tex)
attach_progress_interceptors overwrites scene.add() with a function that saves a copy of the mobject-graph, if the mobject is of type DynamicMobject.
TransformInStages is currently the only available automatic animation.
-
If an ID is only found in the source_graph,
TransformInStagesconstructs a FadeOut target_mobject. These mobjects are removers. -
If an ID is only found in the target_graph,
TransformInStagesconstructs a FadeIn source_mobject. These mobjects are introducers. -
lag_ratio: float = 1- lag_ratio for the remover, transformer, and introducer tracks.
-
run_time: float | None = None- The run_time for the entire animation. If
None, track_run_time is used instead.
- The run_time for the entire animation. If
-
track_run_time: float | None = 1- The run_time provided for the remover, transformer and introducer tracks individually. The default value of 1 yields a 3-second animation if all 3 remover, transformer, and introducer types are present.
tex1 = MathTex("a", "b", "c")
tex2 = MathTex(tex1[1].copy(), tex1[2].copy(), "d")
self.add(tex1)
self.wait(1)
self.play(TransformInStages.replacement_transform(tex1, tex2, lag_ratio=0.5))
tex1.set_color(RED)
tex2.set_color(GREEN)
self.wait(2)replacement_transform removes the source_mobject from the scene, and adds the target_mobject. Mobject.copy() copies the ID of a DynamicMobject and the IDs of its children, the copied IDs are recognized by the animation generator. An mobject-graph cannot have two mobjects with the same ID, so MathTex(a, a.copy()) will raise an error, whereas MathTex(a, a.clone()) does not.
tex1 = MathTex("a", "b", "c").shift(UP)
tex2 = MathTex(tex1[1].copy(), tex1[2].copy(), "d")
self.add(tex1)
self.wait(1)
self.play(TransformInStages.from_copy(tex1, tex2, lag_ratio=0.5))
tex1.set_color(RED)
tex2.set_color(GREEN)
self.wait(2)
class Scene1(Scene):
def construct(self):
attach_progress_interceptors(self)
tex1 = MathTex("a", r"\\", "b")
tex2 = MathString("x + y")
group = DGroup(tex1, tex2).arrange(DOWN, buff=LARGE_BUFF)
self.add(group).wait(1)
tex1.terms = [ tex1[0], tex1[2] ]
self.play(TransformInStages.progress(group))
self.wait(1)DGroup is a reactive version of VGroup. It also enables multiple equations to be placed in the same mobject-graph, where the group is the root-mobject. Using DGroup, making an animation that moves a term from one equation to another equation, is handled the same way as moving a term within the same equation. That is, as a transformer type and not an introducer or remover type.
@mobjects.setteradd(*mobjects)remove(*mobjects)arrange(self, direction = RIGHT, buff = DEFAULT_MOBJECT_TO_MOBJECT_BUFFER, center = True, **kwargs)
Currently, arrange is the only arrangement method supported.
class TermComponent(Scene):
def construct(self):
attach_progress_interceptors(self)
term = Term("a", "b", "c")
term.term.set_color(RED)
term.subscript.set_color(YELLOW)
term.superscript.set_color(BLUE)
self.add(term).wait(2)
term.subscript = None
term.superscript.set_tex_string("2")
self.play(TransformInStages.progress(term))
self.wait(2)Term can have None for superscript, and subscript, and Term(term, superscript=None, subscript=None) renders equivalent to MathTex(term).
@term.setter@subscript.setterNonesupported
@superscript.setterNonesupported
class FractionComponent(Scene):
def construct(self):
attach_progress_interceptors(self)
fraction = Fraction("a", "b")
fraction.numerator.set_color(RED)
fraction.denominator.set_color(BLUE)
fraction.vinculum.set_color(LIGHT_BROWN)
self.add(fraction).wait(2)
fraction.numerator, fraction.denominator = fraction.denominator, fraction.numerator
self.play(TransformInStages.progress(fraction))The numerator and denominator cannot be None.
The vinculum has no setter.
@numerator.setter@denominator.setter@property vinculum
class FunctionComponent(Scene):
def construct(self):
attach_progress_interceptors(self)
function = Function(r"\text{ln}", "x")
function.function_name.set_color(YELLOW)
function.input.set_color(LIGHT_BROWN)
function.shift(UP)
parentheses = Parentheses("x")
parentheses.inner.set_color(LIGHT_BROWN)
self.add(function, parentheses)The property name is used by Mobject, so I made the function's name property function_name.
There is function.bracket_l and function.bracket_r, but I might change that in the future to a use the parentheses component. Parentheses also has parentheses.bracket_l and parentheses.bracket_r.
Let's see how the RHS could be implemented:
In nested constructions, an array can be used as a shorthand syntax to denote a MathTex. In the below code sample, the first green box will be interpreted as Term("e", MathTex("-", "x")). The second green box will be interpreted as Fraction(MathTex("d", tex[0]), "dx").
class Scene1(Scene):
def construct(self):
attach_progress_interceptors(self)
cases = MathCases("a", "b", "c").set_color(GREEN)
self.add(cases).wait(1)
cases.remove(cases[0])
cases.append("d")
self.play(TransformInStages.progress(cases, lag_ratio=0.5))
cases.lines = [ cases[0], cases[2], cases[1] ]
cases.shift(LEFT)
self.play(TransformInStages.progress(cases, lag_ratio=0.5))
self.wait(1)MathCases is similar to MathTex as it uses a list to store it's children. Instead of tex.terms we have cases.lines.
cases = MathCases(
"a, & b",
CaseLine("aaa,", "b")
)
self.add(cases)
cases.set_color(GREEN)
cases[1].output.set_color(RED)
cases[1].condition.set_color(BLUE)At first, I wanted to have CaseLine(output, condition) be "{output}, & if {condition}", but I found examples without commas or without if statements. So currently it separates the output and condition with an alignment character, or "{output} & {condition}".
MathMatrix is a component for a LaTeX bmatrix, which is visually different from manim's Matrix implementation.
class Scene1(Scene):
def construct(self):
attach_progress_interceptors(self)
root = DGroup()
mat = MathMatrix([[ "x", "z" ], [ "y", "w" ]]).set_color(LIGHT_PINK)
self.add(root.add(mat)).wait(1)
mat.matrix = [
[ "a", mat[0][0] ],
[ "b", mat[1][0] ]
]
self.play(TransformInStages.progress(root, lag_ratio=0.4, run_time=1.5))
self.wait(1)
root.mobjects = [ Term("e", mat) ]
self.play(TransformInStages.progress(root, lag_ratio=0.4, run_time=1.5))
self.wait(1)Getting an element via matrix[0][0] is supported, but editing via matrix[0][0] = elem is not supported. Edits must be made through the matrix.matrix setter providing a 2D list.
In this example, we see it's possible for a matrix to be a superscript. However, the default LaTeX rendering is not ideal, the matrix should be smaller in the subscript position. The transformation of the brackets is also not ideal, you will notice a wobble in brackets as it moves into the superscript position. This is because the brackets acquire more submobjects as they get taller, or when they are placed in an superscript position. This mismatch in the submobject count causes the wobble during the render.
An mobject-graph cannot have two mobjects with the same id. Therefore, clone() must be used instead of copy() when placing a copy of an mobject within the same graph.
TransformInStages recognizes that an mobject B is a clone of another mobject A, if B's source_id matches A's id.
class Scene1(Scene):
def construct(self):
attach_progress_interceptors(self)
tex = MathTex(Term("x", "n"))
self.add(tex).wait(1)
tex.terms = [ tex[0], tex[0].clone() ]
self.play(TransformInStages.progress(tex))In this example, the Term, MathString("x") and MathString("n") in tex[0].clone() each have a source_id corresponding to a component in tex[0].
Auto-disconnect enables notation such as A[0] = Component(A[0]), by disconnecting child from its parent without changing the list size of the parent component.
Given a construction A(B), where B is a child of A, if you attempt to construct Component(B) without disconnecting the parent/child relationship between A and B, then B will be disconnected from A by means of rerendering A with B.clone() in place of B.
In this example, x is rerendered using n.clone() in place of n.
Given a construction A(B), where B is a child of A, the use of B.pop() is a shorthand for A.remove(B). This may throw a NotImplementedError if A does not override the abstract remove method declaration in DynamicMobject.
In this example, n.pop() executes x.remove(n). Term's implementation of remove recognizes that n is its superscript, and sets its superscript to None. Then n is available to be placed inside tex.
Enables a remover-mobject to merge onto another mobject during a transformation. Both source_id and target_id are set back to None after transformations.
class Scene1(Scene):
def construct(self):
attach_progress_interceptors(self)
term1 = Term("2")
term2 = Term("2", "4")
tex = MathTex(term1, Parentheses(term2))
self.add(tex).wait(1)
term1.term.target_id = term2.term.id
term2.superscript.set_tex_string("5")
tex.terms = [ term2 ]
self.play(TransformInStages.progress(tex, lag_ratio=0.4))In the above example, the vertical position of the 2 changes when the parentheses are removed. In the below example save_y and restore_y restore the vertical position of the 2, making it move only horizontally.
class Scene1(Scene):
def construct(self):
attach_progress_interceptors(self)
term1 = Term("2")
term2 = Term("2", "4")
tex = MathTex(term1, Parentheses(term2))
self.add(tex).wait(1)
term2.term.save_y()
term1.term.target_id = term2.term.id
term2.superscript.set_tex_string("5")
tex.terms = [ term2 ]
term2.term.restore_y()
self.play(TransformInStages.progress(tex, lag_ratio=0.4))In this example, the original 2 remains at the same position during the animation. This is due to save_y() andrestore_y() being called.
By default, restore shifts the root mobject by the difference of the before and after positions of the mobject that restore is being called on.
Without auto-disconnect
In setting up an animation using the from_copy() constructor, if the target mobject is generated using copies of the original mobject's components, then the copies will remain components of the original mobject.
class Scene1(Scene):
def construct(self):
x = MathString("x")
y = MathString("y")
tex1 = MathTex(x, y)
tex1.shift(UP)
self.add(tex1).wait(1)
tex2 = MathTex(y.copy(), x.copy())
self.play(TransformInStages.from_copy(tex1, tex2))
x.set_color(RED)
y.set_color(RED)
self.wait(3)Using auto-disconnect
However, using auto-disconnect, we can construct a target_mobject directly using the components of the original mobject. This is because replacement copies will be generated by auto-disconnect to fill in for the missing components of the original mobject. Then we can edit the components (that were formerly apart of the original mobject), to edit the new target mobject.
class Scene1(Scene):
def construct(self):
x = MathString("x")
y = MathString("y")
tex1 = MathTex(x, y)
tex1.shift(UP)
self.add(tex1).wait(1)
tex2 = MathTex(y, x)
self.play(TransformInStages.from_copy(tex1, tex2))
x.set_color(RED)
y.set_color(RED)
self.wait(1)
tex3 = MathTex(x, y)
tex3.shift(DOWN)
self.play(TransformInStages.from_copy(tex2, tex3))
x.set_color(BLUE)
y.set_color(BLUE)
self.wait(3)Partial transformation is still in development, it has unresolved problems.
All three transformation constructors support partial transformation.
When TransformInStages is provided a subcomponent of an mobject-graph, it uses the complete mobject-graph to construct the transformation, and uses the subcomponent as an ID selection to determine which IDs to animate.
When provided a subcomponent, the animation generator includes the subcomponent's ID, and recursively includes its descendants in the ID selection. For example, in tex = MathTex("y, "=", Term("e", ["-", "x"]), passing in the term subcomponent would include an ID selection for: Term, MathString("e"), MathTex("-", "x"), MathString("-") and MathString("x").
Using partial transforms requires multiple transforms to cover the total transformation of an mobject-graph from a source_graph to target_graph. Therefore, the progress constructor does not update the target state of the mobject-graph if it's a subsequent transform. This enables subsequent transforms to use the same target state that the first progress constructor uses. Wait calls between progress constructors do not update the target state. The first state change to the mobject-graph following a progress constructor is interpreted as the conclusion of the partial progressions.