¶Terminology

¶Activation Node

Activation Node or AdnActivation is an Adonis node that allows you to perform operations on input activation values to produce a final activation value that can be used to drive the activation of a muscle. This node allows you to override, add, subtract, multiply, divide, etc. multiple input activations to produce one single output. The input activations can be used by ingesting Adonis sensor data (AdnSensorPosition, AdnSensorDistance, AdnSensorRotation). This node is recommended to be used when on-demand activations are required or when multiple activations from several sensors have to be merged into one value.

¶Adonis Model (.adnm)

An Adonis Model or .adnm file is the trained model file generated by the AdonisML training process. It stores the learned relationship between the input data and the output deformation data so it can later be used by AdonisML workflows during inference. The .adnm file is generated from extracted training data and can be used to enhance deformation results without running the original full simulation.

¶Anatomy Transfer

Anatomy Transfer is the workflow used to transfer anatomical layers from one character to another while preserving the spatial relationships between the different layers. In this workflow, the skin and mummy geometries are usually transferred first, then the resulting deformation is propagated to the remaining anatomy layers such as skin cut, fat, muscles, and fascia. Anatomy Transfer can also make it possible to reuse an existing Adonis simulation rig authored for a source character on the transferred anatomy of a new character.

¶Closest Fit

Closest Fit or AdnClosestFit is an Adonis deformer that projects an input geometry onto one or more target geometries using a closest-surface projection model. For each point of the deformed geometry, it finds the nearest target surface and moves the point to that position, making the input geometry conform to the target shape.

Closest Fit is useful for fitting meshes onto anatomical models, conforming accessories to character surfaces, and creating surface-matching workflows between unrelated geometries. Its influence can be controlled globally with the deformer envelope and locally with paintable weights.

¶Constraints

Constraints are rules that an Adonis solver applies during simulation to ensure that the relationship between elements involved in the simulation is maintained, such as distance between geometry points, distance between geometry points and external attachments, rig joints or external meshes, etc. The catalog of constraints is presented below.

Attachment To Transform. An attachment to transform constraint defines the relationship between a geometry point and an external transform object. During simulation, an attachment constraint will try to keep the geometry point at a constant location relative to the transform.

Attachment To Geometry. An attachment to geometry constraint defines the relationship between a geometry point and an external mesh object. During simulation, an attachment constraint will try to keep the geometry point at a constant location relative to the closest point to the target geometry at rest.

Distance. A distance constraint defines the relationship between two points of a geometry. During simulation, distance constraints will try to keep the edge lengths of the mesh at rest.

Fiber. A fiber constraint defines the relationship between two points of a muscle geometry. It is a specialization of the distance constraint where the muscle fibers flow is taken into consideration to emulate the behavior of a real fiber contraction.

Glue. A glue constraint defines the relationship between a geometry point and another point on an external geometry. These constraints aim to keep the point at a certain distance to the external geometry point without any restrictions of relative orientation. The constraint behavior aligns with Soft Constraints that are used to glue muscles together.

Hard. A hard constraint defines the location of a geometry point in the tangent space of a polygon. This constraint type is similar to the attachment constraint but in this case the transformation used is the one given by the tangent space at the closest point on an external geometry.

Self-Collision. Applies corrections to the points that are intersecting with the geometry itself. There are two modes of self-collisions correction in Adonis: point-to-point and triangle-to-triangle. In point-to-point mode, an intersection occurs if the volume around a point intersects with the volume of another point. This volume can be a perfect sphere (uniform, defined by a radius) or a spheroid (when there is thickness and differs with the radius). In triangle-to-triangle mode, an intersection occurs if two triangles are intersecting each other, which can happen if one or two edges of one triangle cross the area of the other triangle.

Shape Preservation. A shape preservation constraint defines the state of the shape formed by a vertex with its adjacents on initialization. During simulation, a shape preservation constraint will try to maintain the rest shape of the geometry with its neighboring vertices.

Slide. A slide constraint defines the distance between a geometry point and a surface. This constraint allows the point to travel along the surface. During simulation, this constraint will try to keep the point at a constant distance to the surface in a given radius.

Slide Collision. A slide collision constraint is an extension of a slide constraint that includes information about the relative orientation of the point against the surface (inside/outside) plus the ability to allow the point to be closer to the surface up to a given threshold.

Slide On Segment. A slide on segment constraint defines the relationship between a geometry point and a segment defined by two transform objects (i.e. two joints of a rig). This constraint aims to keep the point at a certain distance to the segment but with the possibility of traveling along the segment. It also includes a restriction by angle, preventing the simulated point to twist around the segment.

Slide On Geometry. A slide on geometry constraint defines the relationship between a geometry point and a surface of an external mesh object. This constraint allows the point to travel along the surface. During simulation, this constraint aims to keep the point at a constant distance to the surface in a given radius.

Soft. A soft constraint defines the relationship between a geometry point and another point on an external geometry. This constraint aims to keep the point at a certain distance to the external geometry point without any restrictions of relative orientation.

Uber. A uber constraint is a compound of hard, slide and soft constraints.

Volume. A volume constraint defines the volume at rest of a geometry. During simulation, this constraint will try to preserve the volume of the geometry with the ability to introduce volume gain or loss modulated by the volume ratio parameter.

Volume Shape Preservation. A volume shape preservation constraint defines the state of the shape of a unit of volume of a volumetric geometry on initialization. This constraint is used by the fat solver to preserve the shape of every piece of volume existing in the structure generated between the inner and outer layers (fascia and fat geometries respectively) during simulation.

¶Data Extraction

Data Extraction is the process of generating the input and output datasets required for AdonisML training. During this process, pose, joint, and deformation information is sampled from a prepared animation or simulation setup and written to files that can be used by the training process. The extracted data usually includes the input data used to describe the character state and the output data used as the target deformation.

¶Epoch

An epoch is one complete pass through the training dataset during the training process. During each epoch, the model updates its internal parameters based on the training samples and reports metrics such as loss. Reviewing the loss values across epochs helps evaluate whether the model is learning correctly, improving, or overfitting.

¶Fat

Fat or AdnFat is an Adonis solver for fat simulation. This solver allows simulating a mesh surface as if it were a closed volume of fat tissue. The volume is constructed procedurally using two compatible geometries: a base mesh to drive the simulation (i.e. the fascia) and a destination mesh (i.e. the fat) on which to apply the simulation. Thanks to that internal volume structure, the solver is able to produce realistic dynamics typical of fat tissues.

¶Glue

Glue or AdnGlue is an Adonis solver for gluing muscles together after simulation. This solver allows you to glue muscles together giving the simulation of the muscles layer a more compact look. The gluing is achieved by ingesting the muscles into the solver and generating one combined output mesh with Glue Constraints applied. Given a maximum glue distance the gluing can be modulated and controlled to reduce the gluing effect against unwanted muscles. Glue can be useful for improving the simulation of the fascia layer by creating a more compact version of the muscles layer avoiding big gaps.

¶Inference

Inference is the process of evaluating a trained Adonis Model to produce deformation results from new input data. In an AdonisML workflow, inference uses the trained .adnm model to predict deformation or tissue-related output from the current character state. Inference can be used to enhance deformed skin results, drive ML-based deformation, or modulate solver behavior depending on the workflow.

Inference can run on the CPU, but GPU inference may be available when the required ML dependencies and compatible hardware are installed.

¶License Bundles

License Bundles define which groups of Adonis features are available under an activated license. They are applied automatically according to the product purchased, so users do not need to manually select or configure a bundle during activation.

Adonis currently distinguishes between the FX Bundle and the ML Bundle. The FX Bundle enables users to author and run Adonis rigs with all solvers and deformers, including runtime use of existing AdonisML models. The ML Bundle includes the FX functionality and additionally enables the tools and workflows required to extract data, prepare neural clusters, and train new AdonisML models.

¶Locator

Locators are intended to visualize the output of an Adonis sensor. There are three types of locators that require a specific number of inputs and adopt custom shapes in the viewport: AdnLocatorPosition (a squared box at the location of a node), AdnLocatorDistance (a parallelepiped with a line connecting two nodes) and AdnLocatorRotation (an angle with two segments connecting three nodes). Each type is associated with its homologous sensor.

¶ML Deformer

ML Deformer or AdnMLDeformer is an AdonisML deformer used to enhance the result of a deformed skin using learned simulation data. It evaluates a trained Adonis Model and applies the inferred deformation to improve or approximate simulation details on top of an already deformed character skin.

The ML Deformer is useful when a high-quality simulation result has been learned from extracted training data and needs to be reproduced without running the full original simulation during playback or evaluation.

¶Morphed Mummy

Morphed Mummy refers to the mummy geometry after it has been reshaped and reposed during the first stage of the anatomy transfer workflow, usually using Radial Wrap. It acts as a transferred body-volume reference and can be used as one of the targets when propagating deformation to internal anatomy layers such as muscles.

¶Morphed Skin

Morphed Skin refers to the skin geometry after it has been reshaped and reposed during the first stage of the anatomy transfer workflow, usually using Radial Wrap. It acts as a target for transferring the skin cut and helps propagate the final transferred character shape through the rest of the anatomy hierarchy.

¶Mummy

The Mummy is a simplified, unified proxy that represents the overall body volume and serves as the foundation for deformation and simulation across the entire Adonis stack. It is typically built as a wrapped mesh around the individual bone geometries, which improves reliability when computing attachments and allows for stable sliding against target geometry, helping to avoid artifacts such as popping, creasing, and stretching. The mummy must generally be a single, continuous mesh with no separated or floating pieces, although it can optionally be split into multiple pieces when more control over attachment regions is required. Its vertex count should adapt to the proportions and complexity of the creature without introducing unnecessary density. The geometry must also be fully sealed and watertight, especially around critical areas like the rib cage and torso, while maintaining clean, optimized topology with minimal unnecessary subdivisions. For best results, it should be modeled at real-world scale to ensure consistent and reliable simulation behavior.

¶Muscle

AdnMuscle is an Adonis solver for muscle simulation including volume preservation. It allows applying dynamics such as fibers contraction and volume gain to a geometry.

¶Muscle anisotropy

Muscle anisotropy refers to the property of muscle tissue that causes it to behave differently depending on the direction of the applied forces. In Adonis, this property affects the edge stiffness based on the alignment with the fiber flow:

• Edges aligned with the fiber direction have stiffness equal to the object's overall stiffness.
• Edges orthogonal to the fiber direction have reduced stiffness.

This behavior is modulated with the anisotropy parameter where a value of 0.0 (default) makes the object fully isotropic, and a value of 1.0 makes the object fully anisotropic.

¶Muscle Patches

A muscle patch is a group of connected geometry vertices that represent an internal muscle projected onto the skin. A muscle patch is not an actual modelled geometry. Muscle patches are used by AdnSimshape solver for facial simulation. The muscle patches of a facial geometry can be generated with the Learn Muscle Patches Tool which applies Machine Learning techniques to estimate the distribution of muscles based on the set of facial expressions that a facial rig can reproduce.

Adonis Muscle Patches File. The Adonis Muscle Patches File is a proprietary file format that stores the distribution of muscle patches generated by the Learn Muscle Patches tool. The file extension is amp.

¶Mush

Mush or AdnMush is an Adonis deformer used to smooth deformation artifacts while preserving the overall shape of the geometry. It can be used as a standalone deformer or as part of Adonis solver workflows to improve the final deformation quality.

¶Neural Cluster

A Neural Cluster is a localized region of geometry used to guide AdonisML training. Neural clusters associate groups of vertices with relevant joints or deformation areas, helping the training process focus on local relationships between the input pose and the output deformation.

Neural clusters are commonly used to improve locality during training and can also be used by augmentation workflows to generate synthetic training samples from localized regions of the character.

¶Push

Push or AdnPush is a deformer that applies a displacement to the vertices of a mesh in the direction of the normals. This effect is modulated with a scalar value to define the global displacement and a paintable map to provide control per vertex. If the sign of the scalar value is positive the displacement is applied in the direction of the normal (outwards), while if the sign is negative the displacement is applied in the opposite direction (inwards).

¶Radial Wrap

Radial Wrap or AdnRadialWrap is an Adonis deformer used to reshape and repose an input geometry using pairs of corresponding landmarks. The deformation is driven by two landmark sets: input landmarks placed on the source geometry and goal landmarks placed on the desired target shape.

Each input landmark is matched with a goal landmark, and those correspondences define how the source geometry should deform toward the target. Because the correspondence is defined by landmarks instead of point order, Radial Wrap does not require the input and goal geometries to have matching topology.

Radial Wrap is commonly used for character reshaping, anatomy transfer, pose transfer, and fitting operations between related but structurally different meshes. In anatomy transfer workflows, it is often used as the first stage to reshape and repose the skin and mummy geometries before transferring the deformation to the remaining anatomy layers.

¶Relax

Relax or AdnRelax is a deformer designed to smooth creases and correct over-compression or over-stretching on geometry surfaces. This deformer can help refine different types of meshes, like the fascia and skin resulting from the simulation by computing an iterative algorithm that combines smoothing, relaxation, and volume corrections.

¶Remap

In terms of functionality, AdnRemap nodes are typically used in Adonis rigs to remap the output of a sensor into a value that adjusts better to the range expected at the destination attribute of a solver, like the activation or the volume ratio gain of a muscle. In terms of workflow, these nodes aim to enhance the portability of the whole Adonis rig, including not only the input and output connections but also the configuration of the remap ramp attribute.

¶Ribbon Muscle

The ribbon muscle or AdnRibbonMuscle is an Adonis solver for muscle simulation. It allows applying dynamics such as fibers contraction to a planar geometry.

¶Rigid Wrap

Rigid Wrap or AdnRigidWrap is an Adonis deformer that transfers deformation from one or more target geometries to an input geometry using a closest-surface attachment model. For each point of the deformed geometry, it finds the nearest target surface and stores a relative attachment to that surface. As the target geometry deforms, the input geometry follows while preserving its relative position to the target.

Rigid Wrap is useful for attaching secondary geometry to deforming surfaces, transferring deformation between unrelated meshes, and creating rigid surface-following setups. In anatomy transfer workflows, it is commonly used to transfer the skin cut using the morphed skin as target, and to transfer the fascia using the transferred muscles as targets.

¶Sensor

Sensors are nodes to measure positions, distances, angles, velocities and accelerations. There are three types of sensors that require different number of input transform objects: AdnSensorPosition (one single input to compute its velocity and acceleration), AdnSensorDistance (two inputs to compute the distance between them and their relative velocity and acceleration) and AdnSensorRotation (three inputs to compute the angle between them and the angular velocity and acceleration). Each type is associated with its homologous locator that will allow visualizing the output values.

¶Simshape

Simshape or AdnSimshape is an Adonis solver for facial simulation. It allows applying dynamics on top of the deformation driven by a facial rig. Also, it has the ability to mimic the change in rigidity of the skin due to the activation of the internal facial muscle patches.

¶Skin

Skin or AdnSkin is an Adonis solver for skin and fascia simulation. It allows applying dynamics to the skin of a character to produce realistic effects like wrinkles.

¶Skin Cut

Skin Cut is the simulation skin layer used in the anatomy transfer workflow. It is transferred after the morphed skin has been created, usually by applying a Rigid Wrap using the morphed skin as target. The transferred skin cut can then be used as a target for transferring deeper anatomy layers such as fat.

¶Skin Merge

Skin Merge or AdnSkinMerge is a deformer to merge simulation and animation meshes into a single final mesh. It allows selecting multiple animated and simulated skin geometries and dynamically blending their results.

¶Skip Frames

Skip Frames is a data extraction setting that defines how many frames are skipped between recorded samples. It can be used to reduce redundant training poses and control the density of the extracted dataset. Higher values record fewer samples, while lower values record more frames from the animation.

¶Smart Tissue

Smart Tissue or AdnSmartTissue is an AdonisML solver that uses inference data to modulate tissue properties and add a simulation layer on top of an already deformed geometry. It is intended to enhance the deformation result without requiring a full muscle simulation.

Smart Tissue can be used when the skin or tissue deformation already exists, but an additional simulation layer is needed to improve the final motion, secondary behavior, or tissue response using learned data.

¶Soft Wrap

Soft Wrap or AdnSoftWrap is an Adonis deformer that transfers deformation from one or more target geometries to an input geometry using a proximity-based influence model. For each point of the deformed geometry, it searches for nearby points on the target geometries and computes the deformation from the contribution of those neighboring target points.

Soft Wrap does not require topological correspondence between the input and target geometries, which makes it useful for transferring complex deformations between unrelated meshes, driving secondary geometry, and creating flexible surface-following setups. Its behavior is controlled by settings such as radius, which defines the search distance for influencing target points, and maximum points, which limits how many neighboring points can contribute to each deformed point.

In anatomy transfer workflows, Soft Wrap is commonly used to propagate transferred deformation through soft anatomy layers such as fat and muscles.

¶Stabilization Frames

Stabilization Frames is a data extraction setting that defines how many stabilization cooks are performed before recording each frame. It is used to let the simulation settle before writing training data. Increasing the number of stabilization frames can improve the stability of extracted samples, but it also increases extraction time.

¶Target Faces

Target Faces are selected faces on a target geometry that are included or excluded from the computation of external constraints. This allows Adonis solvers to restrict which parts of a target surface are considered when evaluating attachments, sliding, or other external relationships.

¶Training

Training is the process of generating an Adonis Model from extracted input and output data. During training, the ML system learns how the input data relates to the target deformation data and writes the resulting model to an .adnm file. Training can use optional settings such as standardization, neural clusters, and augmentation depending on the dataset and the desired result.