daVinci-MagiHuman
Single-file bf16 conversion of the base DiT checkpoint from the official daVinci-MagiHuman release.
File
base_bf16.safetensors
Upstream source
- Official base checkpoint: https://huggingface.co/GAIR/daVinci-MagiHuman/tree/main/base
Intended use
This file is mainly for Musubi Tuner MagiHuman LoRA training.
Use it as:
--dit path/to/daVinci-MagiHuman/base_bf16.safetensors
Other required weights
You still need:
- T5-Gemma encoder-only: https://huggingface.co/PhatcatDK/t5gemma-9b-2b-ul2-encoder-only/tree/main
- Wan 2.2 VAE: https://huggingface.co/Wan-AI/Wan2.2-TI2V-5B/blob/main/Wan2.2_VAE.pth
- Stable Audio Open 1.0: https://huggingface.co/audo/stable-audio-open-1.0/tree/main
Notes
- base model only
- not the SR stage
- converted from the original shard-based release for easier loading
Inference Providers NEW
This model isn't deployed by any Inference Provider. 🙋 Ask for provider support