# Quick keyframe frame insertion for Shape Keys with foreach_set from data

I would like to animate an objects' shape keys from a data set. Imagine a default cube that will get longer (side face vertices will move in +X direction) according to a data.

The date looks like this {'2000':1, '2001':2.5, '2003':3.4, ... '2020':11.9} As you can see, we have time versus accumulative values. In reality, the data set has 4000 data points.

In the function below, I can receive a new value, add a shape key for it and move my target face in the desired direction. Frame by frame, this works just fine. However the code execution gets really slow after about 100 keyframes insertion.

I am looking for a way to create this animation from the data set quickly. After research, I found about foreach_set function of Blender API. However I could not find a way to use it for my case.

Could anybody please explain me how to use foreach_set function for FAST shape key animation from a dataset?

Relevant questions I could find. However they were not helpful for my case:
Fast keyframe insertion
Editing fcurve.keyframe points in FAST mode?

def resize_bar(obj_name, newValue = 0, active_face = 2, start_frame = 1, end_frame = 2):
# Get the active mesh
obj = bpy.data.objects[obj_name]
obj_mesh = bpy.data.objects[obj_name].data
obj_mesh_matrix = bpy.data.objects[obj_name].matrix_world
# get base shape key ready
if start_frame == 1:
basis_key.keyframe_insert('value', frame=start_frame)
# Get a BMesh representation
bm = bmesh.new()   # create an empty BMesh
bm.from_mesh(obj_mesh)   # fill it in from a Mesh
# get top face of mesh
bm.faces.ensure_lookup_table()
top_face = bm.faces[active_face]
top_normal = bm.faces[active_face].normal
# add shape key and insert a key frame before transformation
new_shape_key.value = 1.0
new_shape_key.keyframe_insert(data_path='value', frame=start_frame)
# apply transformations
bmesh.ops.translate(bm, vec=newValue*top_normal, verts=top_face.verts)
bm.to_mesh(obj_mesh)
bm.free()
obj_mesh.update()
# insert a new key frame after transformation
new_shape_key.value = 0.0
new_shape_key.keyframe_insert(data_path='value', frame=start_frame+end_frame)

• What really is the question here? Rather than shapekeys could simply keyframe z scale, or an array mod count.. (move origin to bottom to only expand one way) Not sure why the linked questions re using foreach get and set on an fcurve are not helpful either???? Aug 7, 2020 at 6:39
• In the initial state of the cube, it is dimensionless (squashed down to a flat object). In the animation, it should be clear that the Cube starts growing from 'a flat surface'. Also, how can I describe the scale in Z direction, when the values are accumulative? This gets complex already. Besides, to my understanding, foreach_set method creates and assigns all the keyfames from a data set at once. This is the efficient way when you have thousands of data points. But I just could not figure out a way to use it in shape keys. Aug 7, 2020 at 7:56

Could be simplified to one shape key.

In the initial state of the cube, it is dimensionless (squashed down to a flat object). In the animation, it should be clear that the Cube starts growing from 'a flat surface'.

Also, how can I describe the scale in Z direction, when the values are accumulative? This gets complex already.

IMO you have totally overcomplicated it, by adding a shapekey for each data point. For this case even Using foreach methods isn't going to help particularly, ie for 4000 points of data creating as many shapekeys and keyframing 2 points per.

Am also of the opinion this could be reduced to simply keyframing object scale, a single shapekey, or solidify thickness to name but 3.

Using solidify thickness on faces defined by vert group "Group"

Shapekey emulating how scale works.

• Transform such that it is unit size, origin at base
• Copy the coords
• Scale it flat, make that the basis shape,
• Make the other shape key unit.

Ok now have a shapekey that if zero has zero height Basis, and if 1 unit height. Changing the range allows scaling outside the (0, 1) range. As illustrated, but appears to be hardcoded to only 10

There is really no need for the shapekeys, since the result is same as scaling in z axis, So could instead create no shapekeys and simply keyframe scale.

Test script. Creates the unit cube illustrated above

import bpy
from mathutils import Matrix
ob = bpy.context.object
me = ob.data
me.transform(
Matrix.Translation((0, 0, .5)) @
Matrix.Diagonal((.5, .5, .5, 1))
)

coords = [None] * len(me.vertices) * 3
me.vertices.foreach_get("co", coords)
me.transform(
Matrix.Diagonal((1, 1, 0, 1))
)
unit.data.foreach_set("co", coords)


Some animation basics

When you add a keyframe to a shapekey, it creates an Action if there is not already one, an fcurve for the data path being animated (if there already isn't one) and inserts a 2D point for (frame, value)

It is in populating the fcurves keyframe points where foreach_set is then used.

>>> C.object.data.shape_keys.animation_data.action
bpy.data.actions['Key.018Action']

>>> action = C.object.data.shape_keys.animation_data.action
>>> for fc in action.fcurves:
...     fc.data_path
...
'key_blocks["Unit"].value'


Besides, to my understanding, foreach_set method creates and assigns all the keyfames from a data set at once. This is the efficient way when you have thousands of data points. But I just could not figure out a way to use it in shape keys

Yeah?, thanks for the tip.

Knowing where the action belongs and what the data path looks like, is the key to using the methods outlined in my answers to links posted

Fast keyframe insertion

Editing fcurve.keyframe points in FAST mode?

Example

Using data from question

• Created mesh with shapekey as above,
• Make the shapekey maximum (as big as it gets from data)
• Have offset data such that 2000 is frame 1, and taken one from value for example sake, to get zero at frame 1.
• Created an action for the "Unit" shapekey and populated it.

Test script.

import numpy as np
import bpy
from mathutils import Matrix

data = (
(2000, 1),
(2001, 2.5),
(2003, 3.4),

(2020, 11.9),
)

ob = bpy.context.object
me = ob.data

d = np.array(data)
d -= (1999, 1)
frames, values = d.T

# max range of 10 ..
#unit.slider_max = values.max()

me.transform(
Matrix.Translation((0, 0, .5)) @
Matrix.Diagonal((.5, .5, .5, 1))
)
me.transform(
Matrix.Diagonal((1, 1, values.max(), 1))
)

coords = [None] * len(me.vertices) * 3
me.vertices.foreach_get("co", coords)
me.transform(
Matrix.Diagonal((1, 1, 0, 1))
)
unit.data.foreach_set("co", coords)
# animation
me.shape_keys.animation_data_create()
action = bpy.data.actions.new("Grow")
action.id_root = 'KEY'
fc = action.fcurves.new(f'key_blocks["{unit.name}"].value')

values /= values.max()

• Very helpful @batFinger, thank you very much. I apologize if I explained my question in a very complicated way. I was already trying to explain that adding a keyframe for each point is overcomplicating things and that I need a different aproach. Anyways, you perfectly provided the aproach that solves the efficiency problem. Just a few question tho: - Could you further explain, in the lines with me.transform, how do you specify the direction in which the cube should grow? Which values specify +Z or +X drections? - What does it actually mean "co" in foreach_set functions?? Aug 7, 2020 at 22:02
• The foreach methods instead of looping over and setting each vert coordinate, can pass a flat list to the vertices collection to set all coordinates. A vertex is a singular member of the vertices collection. Setting its "co" ie getattr(v, "co") or v.co Put the origin on the bottom, hence scaling positive in Z moves in Z. Could just as easily used the bmesh example in question, but only made the one shape at max. Aug 8, 2020 at 19:24