Voxta HY-Motion
Tencent's HY-Motion 1.0 text-to-motion model — generates 3D character animations from natural-language prompts.
HY-Motion uses Tencent's HY-Motion 1.0 model to turn a sentence into full-body motion, up to 10 seconds long. The motion plays on the character's 3D avatar in Voxta Talk.
Marked experimental. Behavior and config may change.
Setup
Add the service
Manage Modules → Add Modules → HY-Motion → Install. Only an admin can add it. Voxta installs the Python runtime on first use; watch the Terminal for progress.
Pick a preset
| Preset | |
|---|---|
| HY-Motion 1.0 (Default) | The standard model, about 7 GB of VRAM plus the text encoder. |
| Lite (Low VRAM) | A smaller model for smaller GPUs. |
The prompt is read by a Qwen3-8B text encoder. The recommended IQ4_XS build is about 4 GB; the full FP16 one is about 16 GB. Pick one and press Download.
Try it
Open Playground → Animations, write a prompt and press Generate. See the Animations playground.
What it's for
- Letting a character act things out in chat with
/performand/animate. - Letting a scenario say "have the character walk to the window and look out" and getting actual motion instead of canned animations, through
chat.perform(). - Making idle, talking and reaction clips for a character in the playground, then dropping them into its assets. See animations by name.
For scripted Timeline animations on the VAM side, see Routimator.