Text-to-Motion in Blender: Uthana vs Kimodo (Cloud vs Local)

You can now generate character animation from a text prompt inside Blender without touching a mocap suit. There are two very different ways to do it — a hosted service or a model running on your own machine — and the right choice depends on cost, privacy and how good the result needs to be.

Option 1: Uthana (Hosted)

The Uthana Blender add-on sends your prompt to Uthana’s servers and imports the result into your scene. It also does video-to-motion, so one add-on covers both text and footage.

  • Cost: pay per second — $0.10/s on Text-to-Motion 3.0, $0.075/s on 3.0 Mini
  • Needs: Blender 4.0+, account, API key, internet
  • Strength: quality tiers, video mocap with hand refinement and foot IK, FBX/GLB/BVH export
  • Weakness: ongoing cost and your prompts and footage leave your machine

Option 2: Kimodo (Local)

Kimodo Blender Bridge runs NVIDIA’s Kimodo text-to-motion model locally and bakes the result onto your rig. kimodo.cpp is the C++ port that runs on CPU or Vulkan, without needing CUDA.

  • Cost: free
  • Needs: a capable machine and some setup
  • Strength: no per-second cost, fully offline, nothing leaves your computer
  • Weakness: setup effort, and text-only — no video mocap

How to Choose

Choose local if you generate a lot, work under NDA, have unreliable internet, or simply do not want a running bill. Free and private beats convenient when volume is high.

Choose hosted if you want video mocap as well as text, need the quality ceiling of a large commercial model, or do not want to maintain a local install.

Plenty of people use both: local for exploration and blocking, hosted for the shots that matter.

Either Way: The Same Cleanup Rules

  • Apply transforms on your character before retargeting — unapplied scale breaks everything downstream
  • Check foot contacts frame by frame on any walk
  • Use the NLA editor to layer corrections on top rather than editing generated keys directly
  • Keep prompts short-clip scoped and describe transitions

Where This Leaves Hand Animation

Generated motion is excellent for blocking, background characters and previs. Hero performances still benefit from an animator — weight, anticipation and accents are judgement calls, not pose estimation. The tools remove the blank-timeline problem; they do not remove the craft.

For video-based capture specifically, see our AI mocap comparison and why AI mocap gets hands and feet wrong.

Leave a Comment

Your email address will not be published. Required fields are marked *

Scroll to Top