motionsmith-animate-install
Install and
set up.
From the extension ZIP to your first generated motion: the MotionSmith Runtime, the Llama 3 text encoder and the worker, step by step.
Before you start
MotionSmith Animate has three parts, installed separately: the extension (this ZIP), the MotionSmith Runtime with the motion model, and the Llama 3 8B Instruct text encoder from Hugging Face. Everything runs locally on your computer.
- Blender 5.2 LTS on Windows x64.
- An NVIDIA GPU with CUDA for motion generation (the default). A CPU option exists, without a performance guarantee.
- Disk space for the runtime (several GB) and the Llama 3 8B model.
- A free Hugging Face account with access to Meta’s Llama 3 8B Instruct.
- Online Access turned on in Blender (Edit → Preferences → System → Network) to download the runtime from the preferences.
Install
Every step below happens in Edit → Preferences. The preferences of MotionSmith Animate are numbered in the same order.
Install the extension
- Open Edit → Preferences → Get Extensions.
- Open the menu at the top right and choose Install from Disk, then pick the MotionSmith Animate ZIP.
- In Add-ons, make sure MotionSmith Animate is enabled, then expand it.
MotionSmith Runtime
The top of the preferences shows three checks: Runtime, Llama 3 and Worker. Each turns into a tick when its part is ready. How to Install opens the documentation.
- Keep Runtime Folder at its default,
%LOCALAPPDATA%\MSR, unless you want the runtime on another drive. In use shows the folder that is used. - Press Install Runtime. It downloads the MotionSmith Runtime 1.0.0 (several GB) and unpacks it into the Runtime Folder, with a progress bar you can cancel.
- When it finishes, the panel reads Runtime 1.0.0 is already installed and the button becomes Reinstall Runtime. An older runtime shows Update Runtime.
Llama 3 8B Instruct (text encoder)
The motion model reads your prompt through Meta’s Llama 3 8B Instruct. Meta’s licence must be accepted on Hugging Face, so you download it yourself.
- Sign in to Hugging Face and open the model page with Llama Download Page.
- Accept Meta’s licence on that page. The approval can take a while.
- Create an access token (Settings → Access Tokens). Then, in a terminal with the Hugging Face command-line tool, run:
hf download meta-llama/Meta-Llama-3-8B-Instruct
Copy Download Command copies the second line. The model lands in the Hugging Face cache, where it is found automatically: the panel shows Found: and its path. If you keep the model in another folder, set Llama Model Folder.
Worker
The worker is the local process that generates motion. It starts only when you press a button and stops with Blender.
- Leave Worker Script empty to use the worker installed with the MotionSmith Runtime. Find Worker fills in its path.
- Press Run to try it. It loads the model, which takes about a minute, and the status becomes Worker: Ready. Stop shuts it down.
- Keep Diffusion Device on CUDA and Text Encoder Device on CPU, the defaults.
kimodo_worker.py
file you received.Library and Output
| Setting | Use |
|---|---|
| Library Folder | Your saved motions and poses. Empty uses MotionSmith Library next to Blender’s version folders, shared by every Blender version. In use shows the folder. |
| Auto-save to Library | Saves every generated motion to the library. On by default. |
| Output Folder | Raw generated jobs. Empty uses the extension’s own user folder. |
| Animation Folder | Folders of your FBX animations for the External Animations panel. + adds another folder; × removes one. |
| Download Free Starter Animations | Downloads a free pack of ready clips with previews. Unpack it and pick its folder as an Animation Folder. |
Updates
Check for Update asks the MotionSmith update server whether a newer version exists; it needs Online Access and runs only when you press it. When there is one, Get Update opens the MotionSmith site. Nothing is downloaded or installed automatically.
First generation
- In the 3D Viewport, press N, open the MotionSmith tab and choose Motion.
- Press Start in the status row and wait for Ready. The model loads once per worker session.
- In Character, pick your character’s armature as the Target Rig. For an unrigged mesh, build one first with Rigging → Quick Rig.
- In Motion Generation, write a prompt such as “A person walks forward at a relaxed, natural pace.”, set Seconds and press Generate. Generation needs Object Mode.
The documentation explains every panel from here on.
If something is missing
A check at the top of the preferences is not ticked
Runtime, Llama 3 and Worker each match one section below them. Open that section: it says what it could not find and in which folder it looked.
Install Runtime does nothing
Turn on Online Access in Edit → Preferences → System → Network. Behind a firewall, download and unpack the runtime yourself and set Runtime Folder to it.
Llama 3 is not found
Check that the download finished and that Meta’s licence was approved for your account. If the model is
outside the Hugging Face cache, set Llama Model Folder to the folder that holds its
config.json.
The worker does not reach Ready
Check Worker Script and that no other MotionSmith worker (for example in Unreal Engine) is running. Open Window → Toggle System Console to read the worker’s messages.
For support, include the Blender and add-on versions, the exact message in the panel and screenshots.
Contact support ↗
Comments
Post a Comment