How to Bind a Skeleton to a Character Mesh in Houdini

A skeleton and a mesh can occupy the same space without being connected. This lesson adds the capture information that lets joint movement deform a character, then tests and adjusts the result. The goal is a working bind and a first weight-painting pass, not finished production skinning.

The scene begins with a procedural character mesh and an existing skeleton. The instructor explains that their construction belongs to earlier lessons. For this workflow, the relevant starting condition is that the mesh and skeleton share the intended capture pose; the recording uses a T-pose.

Capture the rest geometry

The instructor adds Joint Capture Biharmonic, connecting the rest mesh to the first input and its matching skeleton to the capture-pose input. The node produces a weight distribution for the character. The demonstration also shows a tetrahedral mesh associated with the capture process.

The instructor uses nearby fingers to explain why volume connectivity matters: parts may be close in space without being connected directly through the character’s interior. Treat that discussion as an intuition for the method, not a full mathematical definition or a guarantee that neighboring fingers cannot receive unwanted influence.

SideFX’s Joint Capture Biharmonic reference confirms that the node creates capture weights for Joint Deform and builds a tetrahedral mesh internally. It also specifies closed rest geometry. That prerequisite comes from the documentation; the recording does not demonstrate how to diagnose or repair an open mesh.

At this stage, the expected result is captured geometry with weights, not visible animation. The source’s third pose input is connected for later convenience even though no animation is present. A static result is therefore expected until a pose is changed downstream.

Test capture through deformation

Next, add Joint Deform and connect the rest geometry, capture pose and animated pose to the corresponding inputs. The instructor adds Rig Pose on the pose stream feeding the deform node’s third input, then selects and rotates a joint.

Now the surface moves with the skeleton. This distinguishes capture from deformation: capture establishes the relationship, while the deform operation uses a changed pose to show what that relationship does.

The instructor initially inspects a proxy mesh, then switches to his final geometry. He notices that the elbow’s shape and segmentation affect the result. A weight problem and a topology limitation are not automatically the same problem. Before painting, confirm which geometry branch you are evaluating and whether it has enough structure for the bend you are inspecting.

Use a revealing pose to judge the weights

The first automatic result is usable, but the instructor wants less of one region to follow the bend and more to remain with the upper arm. He introduces Joint Capture Paint and uses a posed view to evaluate changes.

The recording rearranges the painting and pose setup during this exploration. Its captions do not establish every final connection clearly enough for a precise wiring diagram. Use the video to check the displayed branch before painting; the supported lesson is the comparison of a deformation before and after weight adjustments.

The practical sequence is:

1. Pose the joint so the unwanted deformation is visible.
2. Select the relevant joint influence—in the example, the left shoulder, then the elbow.
3. Inspect which area of the mesh receives that influence.
4. Paint a small change and compare the same posed region again.

Keeping the pose stable makes the comparison useful. If both the pose and weights change together, it is harder to tell what the paint stroke improved.

Distinguish setting, adding and smoothing

The instructor notes that setting a value of one gives the selected influence full weight in the painted region. He prefers adding influence gradually for this demonstration rather than guessing the final value immediately.

He paints too much in one area and discusses undoing or increasing the neighboring joint’s influence. When he uses smoothing, the redistribution is not the result he wants, so he reduces the brush and increment to work more gradually. The spoken numeric correction is ambiguous in the captions; no exact brush-strength recommendation is reproduced here.

The useful comparison is behavioral. A set operation moves toward a chosen value, an additive edit builds influence in increments, and smoothing redistributes the weight pattern. The recording shows why the instructor changes his approach after looking at the deformation. None of these operations makes every elbow correct automatically.

The joint list and the character’s naming convention help him find the intended influence. Check that selection before a stroke: painting the wrong joint cannot be judged solely by the brush’s screen position.

Remove the test pose and check a short motion

After the paint demonstration, the instructor bypasses or removes the temporary pose and creates a new Rig Pose for an animation test. He attempts to set keys, notices that one key was missed, sets it again, and then sees the deformation change over time. He also tests a small bowing motion.

This is a useful final inspection because a single still pose cannot show the whole transition. The missed key is an animation-test problem, separate from whether the character is bound. Confirm that the pose actually changes before diagnosing a static playback result as bad capture.

The instructor explicitly says he is not fully satisfied with the paint pass and is demonstrating the process. The finished result is a bound character with minimum deformation and adjustment steps, not a certified final rig.

Revisit and evaluate

Caption-supported ranges: 00:00–03:01 establishes the mesh, skeleton and capture; 03:01–05:02 explains the volume intuition and deformation setup; 05:02–10:02 tests the elbow and explores painting; 10:02 onward removes temporary posing and tests animation. These are editorial reading ranges, not official chapters.

The exact Houdini build and modern compatibility remain unverified. No scene was reproduced for this article. The most useful takeaway is to evaluate capture through a deliberate pose, inspect the correct geometry and influence, make a bounded weight change, then check motion before calling the bind finished. The related FBX export lesson follows this work with a save-and-reimport check.

Keep exploring

• Rigging Revolution | KineFX Procedural Skeleton for APEX ↗