Animcraft 6.0 Adds AI Rigging and Kimodo

For those who don’t know the tool: Basefount develops the animation and crowd-production software Animcraft, a collaborative animation library and productivity ecosystem. Animcraft ingests motion from Maya, 3ds Max, Blender, MMD, FBX, BVH and glTF into a reusable library, and its retargeting layer connects workflows across Maya, 3ds Max, Blender, Unreal Engine, Unity and MMD. The new AI features sit upstream of that reuse layer by helping create skin weights and generated motion before those assets move through the existing animation pipeline.

Animcraft 6.0 adds two AI-driven functions aimed at character setup and motion creation: automatic skinning and weight processing from Basefount’s own rigging model, and text-to-animation based on NVIDIA’s Kimodo motion-generation technology. Neither feature replaces Animcraft’s existing role as a motion library and retargeting hub. Instead, both add new ways of producing data that can then be managed, edited and reused inside the wider Animcraft workflow.

The rig gets a model

Basefount says it independently trained an AI character rigging foundation model for Animcraft 6.0. The concrete capabilities named in the announcement are automatic skinning and weight processing. That distinction matters. The release material does not establish that the new model automatically builds an entire production rig, generates a specific skeleton topology or replaces host-side rigging systems. What is explicitly documented is the deformation side of character setup: assigning a mesh to a skeleton and processing the weights that determine how the mesh follows joint motion.

For character teams, that places the new system in a familiar problem area. Initial skinning can be fast, but the cost usually appears in deformation review around shoulders, hips, elbows, knees, fingers and any costume or accessory geometry that does not behave like a clean anatomical surface. An AI-generated weight pass can therefore be useful even when it is not the final pass, provided artists can inspect and correct the result rather than treating automation as an approval stamp.

Kimodo moves in

The second addition brings NVIDIA’s Kimodo into Animcraft for text-to-animation and what Basefount calls intelligent animation. NVIDIA describes Kimodo as a kinematic motion diffusion model trained on 700 hours of optical motion-capture data. The model can generate human or humanoid motion from text and can also be controlled with kinematic constraints.

NVIDIA’s own Kimodo documentation goes considerably further than the short Animcraft announcement. Kimodo supports text prompts together with controls such as full-body keyframes, sparse joint positions and rotations, 2D waypoints and dense 2D paths. NVIDIA distributes several trained model variants for different skeleton representations and datasets. Those details are useful for understanding the underlying research project, but they should not be confused with the Animcraft user interface.

Basefount does not specify which Kimodo model variant Animcraft uses, which constraint types it exposes to artists, or whether every control in NVIDIA’s standalone implementation is available through Animcraft. The announcement confirms the integration and text-driven motion generation, not feature parity with NVIDIA’s research demo or command-line tools. “Intelligent animation” is likewise not defined…

Research formats are not product formats

The standalone Kimodo implementation also shows why integration details matter. NVIDIA’s command-line tools save Kimodo motion in NPZ and provide skeleton-dependent export paths. SOMA models can additionally write BVH, G1 models can write MuJoCo qpos CSV, and SMPL-X models can write AMASS NPZ. For SOMA BVH, NVIDIA documents a 77-joint hierarchy and a choice between the BONES-SEED rest pose and a standard T-pose.

None of that should be presented as an Animcraft export specification. Basefount has not documented which of those representations are exposed, converted or hidden by the 6.0 integration. Animcraft already has its own universal animation-library layer and existing support for ingesting BVH and other animation sources, so the product may abstract some of the research implementation’s skeleton and format plumbing. The public release material does not say how.

That gap is not academic. A text-generated motion becomes production-ready only after skeleton mapping, orientation, scale, rest pose, frame rate and contact behaviour survive the hand-off to the character rig. Kimodo’s research tooling exposes enough of that machinery to show where mismatches can occur. Animcraft’s value proposition has long been to reduce exactly this kind of retargeting friction, but studios should verify how the new generator enters that established pipeline rather than assuming the research defaults pass through unchanged.

NVIDIA’s standalone Kimodo can be installed in a virtual environment or Docker, and its model checkpoints are downloaded for generation. That still does not establish Animcraft’s deployment model. Basefount has not stated in the supplied 6.0 material whether its Kimodo integration runs through the same local stack, packages the models differently or introduces any service dependency.

The useful bit is where it lands

Text-to-motion systems are easy to demonstrate in isolation. Their production value depends on what happens immediately after a motion is generated. That is where Animcraft’s existing architecture makes the addition more interesting than another standalone prompt box.

Animcraft is built around a universal animation library, retargeting and cross-DCC reuse. Its current product page documents motion and rig transfer across Maya, 3ds Max, Blender, Unreal Engine, Unity and MMD, while the library can ingest animation from Maya, 3ds Max, Blender, MMD, FBX, BVH and glTF among other sources. The application also includes motion editing, rig conversion, facial tools, markerless mocap and engine-oriented asset workflows.

That means the generated motion arrives in an environment whose main job is already moving animation between characters and software. The likely production benefit is not simply generating a walk, turn or gesture from text. It is reducing the hand-off between motion generation and the retargeting, editing and reuse work that follows. The release material does not document every step of that path for Kimodo-generated clips, so studios should verify the exact data flow before assuming a one-click round trip.

The same caution applies to the new AI skinning. Animcraft’s broader interoperability does not prove that the 6.0 AI rigging functions are exposed identically in every connected DCC or engine. Basefount’s announcement explicitly tags Maya and 3ds Max, while the main Animcraft site describes a wider ecosystem that also includes Blender, Unreal Engine and Unity. Host coverage for the new AI functions should therefore be treated separately from Animcraft’s established retargeting and library connections.

Automation still needs knees

For rigging and animation departments, automatic weights and generated motion attack different kinds of repetitive work. One starts from geometry and skeleton relationships; the other starts from an intended action. Putting both inside the same animation-management environment could shorten the distance between character setup, first motion and downstream retargeting.

It also creates two new review points. Weighting needs deformation tests, not just a successful bind. Generated motion needs contact, balance, timing, trajectory and character-specific performance checks. A text prompt can produce a plausible motion and still be wrong for the shot, the rig or the gameplay requirement. AI does not remove animation direction. It merely finds another way to arrive at the curves.

Studios should test the release on representative characters, production animation and target hardware before deploying it into an active pipeline.