BD AutoRig (Make-It-Animatable):
BD_AutoRigMIA automatically rigs humanoid character meshes using Make-It-Animatable. Rigging adds a bone structure and assigns parts of the mesh to those bones so the character can be posed or animated. The node predicts joint positions and skinning weights, then exports a rigged FBX with a Mixamo-compatible skeleton. It can also rename the exported bones for a UEFN-oriented workflow. The inference code is included with this node pack, so you do not need to install the ComfyUI-UniRig pack separately. Model weights are downloaded on first use and reused within the same process on later runs.
BD AutoRig (Make-It-Animatable) Input Parameters:
mesh
Required TRIMESH input. Supply the humanoid mesh you want to rig. The node passes this mesh to Make-It-Animatable, which predicts the skeleton and how the mesh should follow its bones. There is no default, minimum, or maximum specified for this input; you must connect a compatible mesh.
fbx_name
Optional STRING input that sets the output filename stem, without an extension. The default is an empty string, which makes the node generate a timestamped stem such as rigged_mia_20260101_120000. With a stem such as character, the node prepares a Mixamo-stage filename of character_mixamo.fbx and, when inline renaming is enabled, requests the stem character_uefn for the renamed file. No length limit or allowed-character rules are specified in the interface.
device
Optional choice of auto, cuda, or cpu; the default is auto. This setting is passed to model loading and identifies the process-level cache entry, so repeated calls with the same choice can reuse a loaded model. In the shown loader, auto and cuda enable GPU caching of model weights, while cpu does not. The source does not establish that every inference operation will run exclusively on the selected device, so treat this as a model-loading setting rather than a guarantee of execution speed or hardware use.
no_fingers
Optional BOOLEAN; the default is true. When enabled, finger weights are merged into the hand bone, simplifying the rig for characters that do not need separately animated fingers. The node description associates this setting with a 22-bone output instead of the 52-bone output described when fingers are retained. Set it to false when separate finger bones matter to your animation workflow. Its choices are true and false; no numeric range applies.
use_normal
Optional BOOLEAN; the default is false. Enabling this asks the inference process to use the mesh’s surface normals when calculating skinning. The interface notes that this can help produce tighter skinning where limbs touch other parts of the body, such as arms resting against the torso. Leave it off for the default behavior, or try true when those contact areas deform poorly. Its choices are true and false; no numeric range applies.
reset_to_rest
Optional BOOLEAN; the default is true. This setting is described as transforming the output mesh toward a T-pose rest position for downstream retargeting. However, the interface explicitly states that it is currently a legacy no-op on its own: it only takes effect when apply_predicted is also true. With both enabled, the source describes an additional use of predicted pose transforms to place the skeleton in the input mesh’s own pose, including an A-pose route for non-T-posed inputs. Its choices are true and false; no numeric range applies.
apply_predicted
Optional BOOLEAN; the default is false. Turn this on to apply Make-It-Animatable’s predicted joint positions instead of exporting the fixed T-pose template armature unchanged. When reset_to_rest is also on, the interface says predicted pose transforms additionally pose the skeleton toward the input mesh’s pose. This is an opt-in change to the default rigging behavior, so compare the resulting FBX with your intended animation or retargeting workflow before adopting it for a batch. Its choices are true and false; no numeric range applies.
remap_to_uefn
Optional BOOLEAN; the default is true. When enabled, the node runs the Mixamo-to-UEFN bone-name remap after rigging and returns the renamed FBX path. The interface gives names such as pelvis, spine_01 through spine_05, and upperarm_l as examples. Set it to false when you want the Mixamo-compatible rigged FBX directly or intend to run a separate renaming step. Its choices are true and false; no numeric range applies.
BD AutoRig (Make-It-Animatable) Output Parameters:
fbx_path
STRING containing the filesystem path to the exported rigged FBX, not the mesh or FBX file contents. With the default remap_to_uefn=true, it points to the result returned by the inline Mixamo-to-UEFN renaming step. With remap_to_uefn=false, it points to the Mixamo-stage FBX returned by inference. Use this path in a downstream node that accepts an FBX path, or locate the file in ComfyUI’s output directory. The node checks that the inference-stage file exists before returning or attempting the inline remap.
BD AutoRig (Make-It-Animatable) Usage Tips:
- Connect a humanoid
TRIMESHand start with the defaults to produce a rigged FBX with inline UEFN-oriented bone renaming. - Keep
no_fingers=truefor characters whose fingers will not be animated separately. Set it tofalsewhen your downstream animation requires individual finger bones. - Try
use_normal=truewhen skinning looks poor where body parts touch, particularly around arms against the torso. - For a workflow that needs predicted joint placement, enable
apply_predicteddeliberately. Remember that changingreset_to_restalone has no effect according to the interface. - Set
remap_to_uefn=falseif a later node will handle bone renaming, or if you need the Mixamo-compatible FBX as the returned result. Avoid applying the same renaming step twice. - Allow time and an internet connection for the first run: model weights are downloaded on first use. Later calls in the same process can reuse the loaded model for the same
devicechoice.
BD AutoRig (Make-It-Animatable) Common Errors and Solutions:
Failed to download Make-It-Animatable model weights from HuggingFace. Check internet connectivity + write permissions on <ComfyUI>/models/autorig/.
- Explanation: The node could not obtain the required model weights during model loading.
- Solution: Confirm that ComfyUI can reach Hugging Face and can write to
<ComfyUI>/models/autorig/. Check available disk space and any network restrictions, then run the node again.
MIA inference returned <path> but file missing
- Explanation: The inference function returned an FBX path, but the node could not find a file at that path.
- Solution: Check that ComfyUI’s output directory exists, is writable, and has free space. Inspect the ComfyUI console for an earlier export or inference error; that earlier message may identify why the FBX was not created.
Import or missing-module error while loading the vendored inference code
- Explanation: The node imports its bundled Make-It-Animatable inference code and its dependencies when it runs. If that code or a required dependency is unavailable, loading can fail. The supplied source does not specify a single exact error message for this case.
- Solution: Check the full ComfyUI console error to identify the missing module. Verify that this node pack’s
lib/autorig/files are present and that the pack’s required dependencies are installed in the Python environment running ComfyUI.
Error during the inline Mixamo-to-UEFN remap
- Explanation: With
remap_to_uefn=true, rigging is followed by a separate bone-name remap. An error in that step can prevent the node from returning a path even after the inference-stage FBX was created. The supplied source does not define an exact remap error message. - Solution: Check the console for the specific remap error. To isolate the step, run again with
remap_to_uefn=falseand inspect the Mixamo-stage FBX. If that succeeds, troubleshoot the remap separately.
