AI Animation Running Now Locally for FREE (6gb vram)
How to Animate 3D Characters with AI Offline
Introduction to Kimodo and Animation Process
- The video introduces a method for animating 3D characters using Nvidia's Kimodo, which operates offline on personal hardware.
- Kimodo is described as Nvidia's text-to-motion model that generates realistic human motion from text prompts, trained on extensive optical motion capture data.
- A custom quantized text encoder allows users to run the setup offline, requiring only 6 GB of VRAM. Installation is simplified through a one-click script provided in the video description.
Installation Steps
- The installation requires three files:
install kimodo.bat,start kimodo.bat, and a Python wrapper file. Links for downloading these files are provided.
- Users must pre-install Python and Git; any version of Python (3.11 or higher) will suffice for the installation process.
- Running
install kimiko.batcreates a virtual environment and downloads necessary models, taking approximately 50 minutes depending on internet speed.
Running Kimodo
- For systems with less than 8 GB of VRAM, an offload tag can reduce VRAM usage during operation.
- After installation, users launch Kimodo by double-clicking
kimiko.bat, which initializes the program while monitoring system resources.
Using the Interface
- To access Kimodo’s interface, users copy a specific path into their web browser where they can find a quick start guide.
- The interface features options like dark mode and different skeleton models; the Soma human body rig is recommended for most users.
Generating Motion from Text Prompts
- Users can input single text prompts to generate animations; starting prompts with "a person" yields better results due to training data specifics.
- After generating motions based on prompts (e.g., "a person walking forward confidently"), users can play back animations directly within the interface.
Utilizing Constraints for Enhanced Control
- Constraints allow users to define keyframe poses that enhance control over character movements during animation generation.
- Adjustments made at specific frames (like arm rotation or foot positioning), followed by regenerating motion, improve animation fluidity between keyframes.
Advanced Animation Techniques
- Users can combine multiple prompts for complex animations (e.g., walking then performing flips), showcasing flexibility in character actions through constraints.
Preparing Character Models for Export
- Before exporting animations, characters should be set in T-pose at frame zero to ensure proper retargeting later in Blender or other software.
Exporting Animations and Importing into Blender
- Animations are exported as BVH files; users need additional tools like Rokoko add-on in Blender for retargeting purposes.
Rigging Characters with Mixamo
- Characters are uploaded to Mixamo for auto-rigging before being re-imported into Blender along with textures applied correctly.
Finalizing Animation Workflow
- Once imported back into Blender, animated characters can be manipulated further. However, hand animations may not be included by default when using Mixamo's rigging system.
Conclusion
- The entire workflow utilizes free tools allowing unlimited animation generation locally on personal hardware while providing detailed steps throughout the process.