Most failed character rigs do not fail at the skeleton. They fail at the shoulder fold, the hip crease, or a handful of vertices that follow the wrong limb. This guide treats auto rig as a testable workflow: load a neutral humanoid mesh, place anatomical landmarks, build the skeleton, compute weights, bend the difficult joints, and export only after the deformation passes. The Sorceress controls and behavior described here were verified against the live product source on July 19, 2026.

What auto rig must produce for a game character
An auto rig has two deliverables. The first is a skeleton: an ordered hierarchy of bones whose joints sit at the character’s real rotation centers. The second is skin weights: values that tell every mesh vertex how strongly it follows each bone. The glTF 2.0 skin specification represents that relationship with joints, inverse bind matrices, and a mesh reference to the skin.
Those deliverables can fail independently. A correctly placed elbow can still pinch if the forearm and upper-arm influences switch too sharply. Smooth weights cannot rescue a knee joint placed above the actual knee. A clean viewport pose does not prove the downloaded file carries the same bone and skin data. Keep the checks separate so each defect points to the stage that created it.
The exact primary phrase auto rig has 720 monthly searches and KD 12 in the DataForSEO pull generated July 19, 2026. The search intent is practical, but “automatic” should not mean invisible. The safest automation handles repetitive skeleton construction and weight calculation while leaving enough evidence for you to reject a bad result before it reaches animation production.
A passing rig should preserve the character’s silhouette in a raised-arm pose, a deep crouch, a bent elbow, a bent knee, and a rotated wrist. It should also reload from the exported file at the intended scale and orientation. If one of those checks fails, the character is not ready merely because an idle animation happens to look acceptable.
Check the mesh before automatic rigging
Start with a humanoid in a neutral pose. The arms need visible separation from the torso, and the legs need enough space for the hip, knee, and ankle centers to be identified. Crossed limbs, a deep action pose, a cape fused to the arm, or a weapon merged into the hand can make a valid anatomical fit impossible. Fix the source mesh instead of asking skin weights to disguise unclear geometry.
Inspect the model from front, side, and three-quarter views. Confirm that the character is upright, that both feet share a sensible floor, and that the body faces the expected direction. Sorceress Auto-Rigging accepts OBJ, FBX, and GLB files. The file is selected through the browser, parsed into mesh data, and displayed in a Three.js viewport. The W3C File API defines the browser file-selection boundary behind that local upload interaction.
Topology also matters. Look for disconnected fragments, accidental interior surfaces, holes around joints, and extreme triangle density. A solver can assign weights to flawed geometry, but the output remains flawed geometry with a skeleton attached. Remove pieces that should not deform with the body, and keep intentional accessories organized so their motion contract is clear.
Do not judge the mesh only in the rest pose. Imagine the shoulder rotating overhead and the thigh moving forward into a crouch. There must be enough surface structure around those joints to fold rather than collapse. Rigging changes transforms; it does not create missing volume or repair a badly fused sleeve.
Run the Sorceress auto rig workflow
Open Sorceress Auto-Rigging. It is marked as a Pro tool in the current product catalog. Use this focused page when you already have a humanoid mesh. Use 3D Studio when the rig belongs to a larger image-to-3D, refinement, and animation workflow.
- Load the mesh. Drop an OBJ, FBX, or GLB into the upload area, or click to browse. Wait for the viewport and mesh statistics to appear before placing landmarks.
- Orient the character. Verify the front-facing direction and inspect more than one camera angle. A landmark that looks centered from the front can sit outside the body from the side.
- Place the landmarks. Guided mode starts enabled, Auto Mirror starts enabled, and Center Snap is available for center-line points. The source defines pelvis, neck, chin, paired hips, shoulders, elbows, wrists, knees, and ankles.
- Build the skeleton. Run the detection/build action only after every required visible landmark is present. Inspect the resulting bone chains inside the body volume.
- Compute weights. Click Compute Weights. The live implementation sends the mesh and generated deform skeleton to the project’s Blender auto-weight service, then maps the returned influences back onto the viewport mesh.
- Inspect and pose. Select bones, show the weight visualization, and use pose controls to expose deformation around major joints.
- Export after acceptance. The current export path builds a skinned GLB internally, converts it through the project’s FBX pipeline, and downloads a rigged FBX.
The compute step is deliberately not described as browser-only. Mesh parsing and interaction happen in the web app, while the current weight calculation calls the Blender service. That distinction matters when diagnosing a failed request: a missing viewport mesh is an import problem; a weight request error is a compute-service problem.

