Musubi Tuner
A local LoRA training toolkit for image and video models that runs on Windows and Linux with NVIDIA GPUs and supports multi-GPU training.
Musubi Tuner provides training scripts for people who want to adapt image and video generation models with their own datasets. It's aimed at users comfortable with a script-based workflow.
- Image and video training: Train LoRA adapters for HunyuanVideo, Wan2.1/2.2, FramePack, FLUX.1 Kontext, FLUX.2 dev/klein, Qwen-Image, Z-Image and MiniMax-H3.
- Single-frame training: Wan and FramePack also support training on individual frames.
- GPU memory savings: Memory-saving options reduce GPU memory use when training supported models.
- Multi-GPU training: Accelerate lets training use multiple GPUs.
- Generation: Inference scripts let you generate images or videos with supported architectures.
The Python toolkit runs locally on Windows, including Windows on ARM64, and Linux with NVIDIA CUDA GPUs. Recommended hardware is at least 12GB of VRAM for image training or 24GB for video training, plus 64GB of system RAM. Actual needs depend on resolution and training settings; 32GB of RAM with swap may work.
The project is unofficial and unaffiliated with the supported model teams. It's experimental, isn't intended for production use, and may introduce breaking changes.
Logotyper
