ComfyUI-RH-daVinci-MagiHuman INT8

INT8-quantised daVinci-MagiHuman DiT and super-resolution weights for ComfyUI-RH-daVinci-MagiHuman. This repository contains only the four converted .pt files. TurboVAE, the T5 text encoder, and the video/audio VAEs still come from the official releases.

The ComfyUI nodes load these INT8 files directly. Official BF16 DiT shards and official 540p_sr/ / 1080p_sr/ config directories are not required.

Full snapshot is about 57 GiB. Download only the files you need.

Mirror: Hugging Face / ModelScope

File	Size	Role
`base_int8.pt`	14.25 GiB	INT8 Base DiT (32 steps, higher quality)
`distill_int8.pt`	14.25 GiB	INT8 Distill DiT (8 steps, faster)
`sr_540p_sr_int8.pt`	14.25 GiB	Optional INT8 540p super-resolution
`sr_1080p_sr_int8.pt`	14.25 GiB	Optional INT8 1080p super-resolution

You need one of base_int8.pt / distill_int8.pt. Add an SR file only when the loader's sr_model is 540p_sr or 1080p_sr.

Install into ComfyUI

ComfyUI/models/MagiHuman/
├── base_int8.pt
├── distill_int8.pt
├── sr_540p_sr_int8.pt          # optional
└── sr_1080p_sr_int8.pt         # optional

Hugging Face

cd /path/to/ComfyUI
python3 -m pip install -U huggingface_hub

hf download Gluttony10/ComfyUI-RH-daVinci-MagiHuman base_int8.pt \
  --local-dir ./models/MagiHuman
hf download Gluttony10/ComfyUI-RH-daVinci-MagiHuman distill_int8.pt \
  --local-dir ./models/MagiHuman
hf download Gluttony10/ComfyUI-RH-daVinci-MagiHuman sr_540p_sr_int8.pt \
  --local-dir ./models/MagiHuman
hf download Gluttony10/ComfyUI-RH-daVinci-MagiHuman sr_1080p_sr_int8.pt \
  --local-dir ./models/MagiHuman

ModelScope

pip install modelscope
cd /path/to/ComfyUI
modelscope download --model Gluttony10/ComfyUI-RH-daVinci-MagiHuman base_int8.pt \
  --local_dir ./models/MagiHuman
modelscope download --model Gluttony10/ComfyUI-RH-daVinci-MagiHuman distill_int8.pt \
  --local_dir ./models/MagiHuman
modelscope download --model Gluttony10/ComfyUI-RH-daVinci-MagiHuman sr_540p_sr_int8.pt \
  --local_dir ./models/MagiHuman
modelscope download --model Gluttony10/ComfyUI-RH-daVinci-MagiHuman sr_1080p_sr_int8.pt \
  --local_dir ./models/MagiHuman

Other required assets

These files are not in this repository:

# Official MagiHuman assets (TurboVAE, T5 text encoder)
hf download GAIR/daVinci-MagiHuman --local-dir ./models/MagiHuman
pip install modelscope
modelscope download --model GAIR/daVinci-MagiHuman --local_dir ./models/MagiHuman

# External VAEs
hf download stabilityai/stable-audio-open-1.0 \
  --local-dir ./models/audio_checkpoints/stable-audio-open-1.0
hf download Wan-AI/Wan2.2-TI2V-5B \
  --local-dir ./models/Ovi/Wan2.2-TI2V-5B

Expected layout after everything is in place:

ComfyUI/models/
├── MagiHuman/
│   ├── base_int8.pt
│   ├── distill_int8.pt
│   ├── sr_540p_sr_int8.pt
│   ├── sr_1080p_sr_int8.pt
│   ├── t5gemma-9b-9b-ul2/
│   └── turbo_vae/
├── audio_checkpoints/stable-audio-open-1.0/
└── Ovi/Wan2.2-TI2V-5B/

Plugin

Current plugin: RH-RunningHub/ComfyUI-RH-daVinci-MagiHuman

Node	Purpose
`RH MagiHuman Model Loader` (`RH_MagiHumanModelLoader`)	Load INT8 DiT, VAEs, and optional SR
`RH MagiHuman Generate` (`RH_MagiHumanGenerate`)	Talking-head video from a reference image and text prompt

Loader options:

model_type: base (32 steps) or distill (8 steps)
vram_mode: mid_vram (~~16 GB) or low_vram (~~6.5 GB)
sr_model: none / 540p_sr / 1080p_sr

mid_vram + 540p SR needs about 24 GB. mid_vram + 1080p SR needs about 48 GB.

License

Converted weights follow the upstream Apache-2.0 license of daVinci-MagiHuman.

Model tree for Gluttony10/ComfyUI-RH-daVinci-MagiHuman

Base model

GAIR/daVinci-MagiHuman

Finetuned

(5)

this model

Gluttony10
/

ComfyUI-RH-daVinci-MagiHuman