AI character animation had an unusually busy week. NVIDIA fully open-sourced MotionBricks, Uthana shipped new text and video motion models, a free local mocap app went viral, and an agent workflow started rigging entire machines in Blender.
Here is what each one actually is, what it costs, and who it suits.
NVIDIA MotionBricks — Real-Time Motion Between Poses
Give it a character’s recent poses and the poses to reach next, and it generates the motion in between, in real time. Trained on 350,000+ motion clips, now released with a pretrained human model, a Python runtime and authoring tools.
Cost: free. Licence: Apache 2.0 code; weights under the NVIDIA Open Model Agreement, which allows commercial use with attribution — but the bundled sample motions are evaluation-only. Needs: an NVIDIA GPU, or the community motion-bricks.cpp port. Full details.
Uthana Text-to-Motion 3.0 — Animation From a Prompt
Describe an action and get editable skeletal animation back, exportable as FBX, GLB or BVH. Now in three tiers: 3.0 for hero shots, Lite for iterating, Mini for volume.
Cost: $0.10 per second for 3.0, $0.075 for Mini, pay as you go. Full details and which tier to use.
Uthana Video-to-Motion 3.0 — Hosted Mocap
Single-person video to 3D motion, now with a dedicated hand refinement step and foot IK — targeting the two failures animators most often fix by hand.
Cost: $0.06 per second. Works from Blender and Maya via free plugins. Full details.
RGB MoCap — Free Local Mocap
A Windows app wrapping several research pose models, with contact correction, finger tracking, multi-person and multi-camera capture, retargeting and Blender project export. Everything runs on your machine.
Cost: free — but 41 to 63 GiB of disk, Windows plus an NVIDIA GPU, and model licences to check before commercial use. Full details and our hands-on look.
AnyMachine — Agents That Rig Machines
Not a motion model but an agent workflow: references, parts breakdown, one Python build script for geometry, rig and animation, then self-checking renders. Built a 30-metre walker with a 104-bone rig.
Cost: MIT and free, plus your agent and image-API usage. Needs: Blender 5.2. Full details.
Which One Is For You?
- You have footage and an NVIDIA PC — RGB MoCap
- You have footage and a Mac, or want it fast — Uthana Video-to-Motion
- You have no footage — Uthana Text-to-Motion, or free local kimodo.cpp
- You are building a game with NPCs or crowds — MotionBricks
- You need machines, not people — AnyMachine
The Pattern Across All Five
Every one of these produces editable data — skeletal animation, rigs, build scripts — rather than finished video. That is the important shift. AI animation is becoming a pipeline input you keep refining, not a black-box output you accept or reject.
Which also means the craft has not gone anywhere. Generated motion still needs timing, weight and contact work on anything hero-facing. These tools remove the blank timeline; they do not remove the animator.
Comparing approaches? See Uthana vs Kimodo for text-to-motion in Blender and why AI mocap gets hands and feet wrong.

