ComfyUI Node
AuK Model Loader X
A ComfyUI node in WorkflowX/Audio/AuK with 3 inputs and 1 output.
AuK Model Loader X
- WORKFLOWX_AUK_MODEL
◄model_name▾►
◄precision▾►
◄attentionauto►
CategoryWorkflowX/Audio/AuK
Inputs (3)
| Name | Type | Default | Description |
|---|---|---|---|
| model_name | COMBO | Converted AuK checkpoint. Base = full quality, Flash = distilled 4-step. The file suffix sets the weight format: bf16 (baseline), int8 or w4a4 (smaller, less VRAM). | |
| precision | COMBO | Compute precision. auto = bf16 on GPU, fp32 on CPU. The weight format (bf16/int8/w4a4) is read from the checkpoint itself, not from this setting. | |
| attentionopt | COMBO | auto | auto uses ComfyUI's attention dispatcher. Explicit flash/sage require CUDA fp16/bf16 and fall back to PyTorch SDPA when unavailable, incompatible, or when an edit mask is present. |
Outputs (1)
| Name | Type | Description |
|---|---|---|
| WORKFLOWX_AUK_MODEL | WORKFLOWX_AUK_MODEL | — |