ComfyUI Node
RH MagiHuman Model Loader
A ComfyUI node in MagiHuman with 5 inputs and 1 output.
RH MagiHuman Model Loader
- pipeline
◄model_type▾►
◄use_turbo_vaetrue►
◄vram_modemid_vram►
◄sr_modelnone►
◄torch_compilefalse►
CategoryMagiHuman
Inputs (5)
| Name | Type | Default | Description |
|---|---|---|---|
| model_type | COMBO | 2 options: base, distill | |
| use_turbo_vae | BOOLEAN | true | — |
| vram_mode | COMBO | mid_vram | mid_vram: INT8 DiT + GPU/CPU ping-pong (~16 GB). low_vram: INT8 DiT + layer-by-layer CPU offload (~6.5 GB base). |
| sr_model | COMBO | none | Super-resolution model. Upscales base output to 540p or 1080p in a second denoising pass. |
| torch_compile | BOOLEAN | false | Enable torch.compile (dynamo). First run compiles ~5 min, subsequent runs faster. Disable for eager mode. |
Outputs (1)
| Name | Type | Description |
|---|---|---|
| pipeline | MAGIHUMAN_PIPELINE | — |