Eye Control Bone Stabilization Problem in the Face Rig

cihpargx

Hello,

I'm having trouble with the character's face. I understand the logic behind how a face is boned and how a mesh system is created. When I animate the face, I have a very difficult time moving the eyes in line with the facial movement.

I'd like to briefly explain the bone system...

Simple bone creation for the eye:

_+Neck Bone
____+Eye Control Bone
________+Eyelid Bone (Upper)
____________Eye Control Bone
________+Eyelid Bone (Lower)
___________Eye Control Bone
______Eye Bone

I'm not having any problems so far. Opening and closing the eyes, eye movements, etc., are not a problem so far.

I've created two bones for the face.

_+Neck Bone
___Frontal Face Bone
___Posterior Face Bone

The anterior face bone contains the nose, eyebrows, and eyes.
The posterior face bone contains the ears and the back of the head.

I constrained the Posterior Facial Bone to the Frontal Facial Bone. Then, I adjusted the color values and added volume using "weights." This allowed me to move the face up, down, right, and left. So far, everything looks normal.

Then, I constrained the "Eye Control Bone" to the "Frontal Facial Bone" based on facial movement and entered the necessary values. This is where I can't quite get it right. The eyes are protruding or moving out of their sockets. In other words, the "Eye Control Bone" is protruding from the eye socket. Even though I set different values for the constraint, I still get the "Eye Control Bone" problem with up-and-down facial movement, but the "left-and-right movement" is disrupted, and vice versa.

I can't solve this problem with "weights." This is because the "Eyelid Bone (Upper)" and "Eyelid Control Bone" are tied to the image through weights. If I also bind the "Frontal Facial Bone" to the weights, I'll have trouble opening and closing the eyes with the "Eyelid Control Bone."

Can you give me some ideas on how to keep the eyes in harmony with the facial movements according to this system I have created?

Misaki

cihpargx This is actually quite a challenging task. While our blog does have a tutorial on setting up eye movement restrictions using IK, and it explains the basic information needed to prevent eyes from popping out of their sockets, it becomes extremely complex when perspective rendering comes into play. I've retested it myself, and I still believe it fundamentally requires painstaking adjustments.

I have created an example that incorporates eyelid images and control bones for the entire face's orientation, while building upon the content of the eye movement distance restriction tutorial:

eye-control-example.zip

61kB

(The reason the entire eye moves is because it is assumed that the entire face is changing direction, and the eye is following that movement.)

I'm not sure if this can be directly referenced for your skeleton, but I think a setup like this would generally be easier to control.

Spinebot

Based on your description, it sounds like you're experiencing issues with the eye control bone not following the face properly when using transform constraints. Here are some approaches to help solve this problem:

Understanding the Core Issue

When you constrain the Eye Control Bone to the Frontal Facial Bone, the constraint applies transformations that may not perfectly match the distortion happening in your weighted face mesh. This causes the eyes to appear to move out of their sockets.

Solution 1: Using Weights Instead of Constraints

Instead of using a transform constraint on the Eye Control Bone, consider binding it to the face bones using weights:

Bind the Eye Control Bone (and related eye bones) to both the Frontal Face Bone and Posterior Face Bone in setup mode
Use weight painting to control how much each bone influences the eye position
This way, the eyes will follow the mesh deformation naturally

For the eyelids, you can maintain their current setup since they need independent control. The key is that the Eye Control Bone itself follows the face through weights, while the eyelid bones remain children of the Eye Control Bone for independent animation.

Solution 2: Adjusting Transform Constraint Settings

If you prefer using constraints, try these adjustments:

Experiment with different mix values for translation and rotation separately in the transform constraint
The translation mix might need to be lower (like 20-60%) to create a subtle "drag" effect that adds depth
Consider setting up the constraint with different values for different directions of movement

You might find that a translate mix around 60-80% works better than 100%, allowing some independence while still keeping the eyes roughly in place.

Solution 3: Additional Control Bones

Create an intermediate bone hierarchy:

Eye Socket Bone (weighted to face bones)
- Eye Control Bone (child of socket bone)
- Eyelid bones and eye bones

This way, the Eye Socket Bone follows the face deformation through weights, and you maintain full control over eye direction through the Eye Control Bone.

Important Consideration About Binding

Make sure all eye-related bones were bound to the face bones in the same pose (preferably setup mode). If bones are bound at different times when the face is in different positions, this can cause alignment issues. If this is the case, you can use Update bindings (https://esotericsoftware.com/spine-weights#Update-bindings) to rebind all bones in a consistent state.

The weight-based approach is generally more reliable for this type of setup, as it allows the eyes to follow the organic deformation of the face mesh naturally, rather than trying to approximate it with constraint values.

cihpargx

Misaki Thank you very much. I also have two questions.

The style of a design is certainly important, but if a character needs to be standardized, is it most logical to make the eye socket empty (transparent) or fill it in, as shown below? For the character below, the eye socket could be left empty and the skin of the eyelids and eyes could be added to the background. Which method do you think is best for perspective?

When breaking a character into objects, I'm undecided about whether to leave the eye socket empty (transparent) or filled.

My second question is the order of the scale and rotator...

+Bone Scale
___+Bone Rotator
_____+Bone Holder
________Visual

Why is Scale at the top and Rotator at the bottom here? What's the difference if the order is reversed, with the Rotator (parent) being first and Scale (child) being the child? Is there a specific logic to it? Is there a difference in performing the same operation by swapping the positions?

Misaki

cihpargx Whether to leave the eye sockets empty depends less on whether you use perspective projection and more on whether you want to create smooth eyelid opening/closing animations using meshes. The “chibi-stickers” skeleton you mentioned achieves eye appearance changes solely by switching region attachments and doesn't use mesh attachments for the eyes. If you want this kind of expression switching, creating eye sockets probably isn't necessary.

On the other hand, if you want to create smooth eyelid opening/closing animations, or if you're considering things like partially closing the eyelids to create a puzzled expression, then the type that creates a hollow, like the elf in the “Snowglobe” skeleton, is preferable. The example I shared in my previous reply used images without shading. However, in real cases, there might be shadows on the whites of the eyes or hair shadows falling on the eyelids. In such situations, having the eyes positioned deeper within the face, as if sunken, makes it easier to animate these shadows along with the eyelids.

Placing the Scale below and the Rotator above yields the same result. However, generally speaking, placing the Scale above and the Rotator below is easier to understand. This becomes clear when you try using a Rotator on a bone with length rather than a bone with no length:

In the left setup, Scale is above and Rotator is below. In the right setup, Rotator is above and Scale is below.
In the right setup, Rotator is not scaled, so the Rotator bone's length remains unchanged. In the left setup, Rotator is scaled, so the bone shortens according to the set scale. (Scale X/Y 0.1 is set.)
The left side visually matches the Child bone's movement distance with the Rotator's length, giving it a more logical appearance.

cihpargx

I absolutely couldn’t manage to get the eyes to focus on a face. I watched videos all day, but every time I got to the eyes, I got stuck. I think I tried five or six times a day… Each attempt took three hours. Hours passed… Even when I positioned the eyes correctly, they kept breaking as I adjusted other parts. I just couldn’t get past this.
If I tried to adjust the eyes one by one, they would pop out of their sockets. If I tried to set them using weights, I ended up working on the eyelids, and then the pupils would jump out. The whole process was really painful. I’m disappointed that I couldn’t achieve anything after all this time.
I checked everything: constraints, weights, the bone system, the parent–child hierarchy. And in the end, I gave up. Maybe I’ll continue later.

Misaki

cihpargx I'm sorry to hear that you're having trouble with eye rigging even after spending so much time on it. While I'm not sure exactly how much facial orientation you're trying to change, if you're making changes beyond a certain degree, you should consider using attachment switching as well. For example, if you want to change the orientation of a face from a front-facing position to a fully side-facing one, it's safe to say this is nearly impossible to achieve without switching attachments midway.
When I need to make significant orientation changes, I often prepare duplicate attachments that reference the same image but have different vertex and weight settings, switching between them mid-process. While this setup requires managing attachments alongside bone-based orientation changes during animation, I find it more reliable to preserve attachments with weight settings that worked well for specific orientations and prepare separate attachments for other orientations. This approach allows for steady progress.

In 4.3-beta, you can also use sliders to bring in poses from other animations. This allows you to specify something like “when a particular bone reaches a certain transform value, apply the pose from a specific frame of another animation.” However, a slider can reference only one type of transform for the target bone. For example, it can react to Translate X, but it cannot reference Translate Y at the same time. Because of this limitation, it isn’t very suitable for motions that require multiple coordinated directions, such as facial orientation.

We plan to add joystick functionality in the future. If your facial rigging is not time-sensitive and is more of an experimental attempt, waiting for that feature to be added may be a sensible approach.

cihpargx

Misaki Thank you for your interest. My response will be a bit long. I apologize for that. 🙂 Your answers were helpful and informative. I will try again because I love Spine and want to specialize in this area. I hope I can solve it easily. If you allow me, I have a suggestion regarding training. Spine has separate training videos for each topic, which I appreciate. Also, these videos are short, and I can revisit them whenever I want. Therefore, I think Spine is excellent in this regard because it provides the foundation. Generally speaking, there are other training videos available on Spine and YouTube.

I just want to talk about face rigging. For example, I open any face rigging video on YouTube (any video other than Spine training videos). In the "introductory video" of the training, the face moves left-right-up-down. When I open the training video, the face moving left-right-up-down is left in one corner. A copy of this is taken and only the head rigging that turns left and right is done. So the logic is explained very well, but the really important parts are left out. While I've learned how to move the face left and right, I can't seem to learn how to solve problems involving the up and down movement of the face with these values. Yes, the logic behind the system for turning the face left and right is the same as for turning it up and down, but because it's limited to just turning the face left and right, I'm lacking in understanding how to solve the problems. I've watched many videos, but something is missing, and I don't know why. Maybe the problem is with me.

For example, a classic character can be defined. The face of this character can be rigmed as follows:

A side view of the same character's face can be rigmed.
A view of the same character's face at a 45-degree angle can be rigmed.
A view of the same character's face at a 55-60-degree angle can be rigmed.
A front view of the same character's face can be rigmed.

For example, in this image:

Number 1: Looking at a 45-degree angle. The area I've marked in red on the left looks very thin (between the eye and the temple bone). The red area appears thin, which will cause problems if there's a gap in the eye socket because there's a thin layer of skin. At this angle, if we try to move it a little further to the side, the eyes and the limitations don't align.

Number 2: This is a face angle I've generally seen in many training videos. Because the skin layer between the eye and the temple bone is larger than in number 1, it will be easier to rig and move left, right, up, and down more easily. Easier than number 1.

Number 3: Rigping is easier than in images 1 and 2.

I think these could be basic...

With your permission, I have a suggestion regarding the Spine program; I don't know if it's mathematically correct and applicable. Perhaps it's an invalid suggestion. Weights are determined as percentages (100%). I don't know if this can be converted to a two-layer structure.

For a face rig (for only one image/layer), the first percentage section can be adjusted for right and left movement.

Separately, a second weight can be entered with a second percentage slice, up or down. In total, it could be a 200% slice. This would make it even more differentiated. But then I don't know how the transitions between them would be achieved. Maybe with a blender....

Also, while thinking about this, I noticed a problem in the "Weights" section. In the image below, with 3 weights, the total is 101. With 5 weights, the total is 102. Is there any logic to this?