ComfyUI Node
ROCm KSampler
KSampler with auto-detected architecture and model optimizations
ROCm KSampler
- model
- positive
- negative
- latent_image
- LATENT
◄seed0►
◄steps20►
◄cfg8.0►
◄sampler_name▾►
◄scheduler▾►
◄denoise1.00►
◄optimize_for_videofalse►
◄precision_modeauto►
◄compatibility_modefalse►
CategoryROCm Ninodes/Sampling
Inputs (13)
| Name | Type | Default | Description |
|---|---|---|---|
| model | MODEL | The model to sample from | |
| seed | INT | 00–18446744073709550000 | Random seed for generation |
| steps | INT | 201–10000 | Number of sampling steps |
| cfg | FLOAT | 8.00–100 | Classifier-free guidance scale |
| sampler_name | COMBO | Sampling algorithm to use | |
| scheduler | COMBO | Scheduler for noise timesteps | |
| positive | CONDITIONING | Positive conditioning | |
| negative | CONDITIONING | Negative conditioning | |
| latent_image | LATENT | Latent image to sample from | |
| denoise | FLOAT | 1.000–1 | Denoising strength |
| optimize_for_videoopt | BOOLEAN | false | Disable previews/progress for multi-frame latents |
| precision_modeopt | COMBO | auto | ROCm precision hint (no-op on CUDA/CPU) |
| compatibility_modeopt | BOOLEAN | false | Force pure stock behavior |
Outputs (1)
| Name | Type | Description |
|---|---|---|
| LATENT | LATENT | — |