ComfyUI Node
WAN Dual KSampler (Boundary)
Boundary-based Dual KSampler for WAN 2.2 MoE: switches from HIGH to LOW model based on timestep boundary.
WAN Dual KSampler (Boundary)
- model_high
- model_low
- positive
- negative
- latent_image
- latent
◄seed0►
◄steps20►
◄boundary0.875►
◄cfg6.0►
◄sampler_name▾►
◄scheduler▾►
◄denoise1.00►
Categorysampling
Inputs (12)
| Name | Type | Default | Description |
|---|---|---|---|
| model_high | MODEL | The HIGH noise model (handles high noise levels, t > boundary). | |
| model_low | MODEL | The LOW noise model (handles low noise levels, t < boundary). | |
| seed | INT | 00–18446744073709550000 | The random seed used for creating the noise. |
| steps | INT | 201–10000 | Total number of denoising steps. |
| boundary | FLOAT | 0.8750–1 | Timestep boundary for model switching. Recommended: 0.875 for T2V, 0.9 for I2V. |
| cfg | FLOAT | 6.00–100 | Classifier-Free Guidance scale. |
| sampler_name | COMBO | The sampling algorithm to use. | |
| scheduler | COMBO | The scheduler for noise removal. | |
| positive | CONDITIONING | Positive conditioning (prompt). | |
| negative | CONDITIONING | Negative conditioning. | |
| latent_image | LATENT | The latent image to denoise. | |
| denoise | FLOAT | 1.000–1 | Denoising strength. |
Outputs (1)
| Name | Type | Description |
|---|---|---|
| latent | LATENT | The denoised latent. |