RF Diffusion Denoiser Config¶
Builds a denoiser configuration for RF Diffusion, controlling how much noise is applied to translations (Cα positions) and rotations during the diffusion process. You can set separate initial (t=T) and final (t=1) noise scales and choose how they interpolate across steps for both translations and frames.
Usage¶
Use this node when you want fine-grained control over the denoising schedule in RF Diffusion. Configure translation and rotation noise scales and schedules here, then connect the output to the 'denoiser_config' input of the RF Diffusion node to influence sampling behavior.
Inputs¶
| Field | Required | Type | Description | Example |
|---|---|---|---|---|
| noise_scale_ca | True | FLOAT | Initial-step (t=T) noise scale for translations (Cα positions). Range 0.0–1.0. | 1.0 |
| final_noise_scale_ca | True | FLOAT | Final-step (t=1) noise scale for translations (Cα positions). Range 0.0–1.0. | 1.0 |
| ca_noise_schedule_type | True | ENUM | Interpolation schedule between initial and final translation noise scales. Allowed values: linear, constant. | constant |
| noise_scale_frame | True | FLOAT | Initial-step (t=T) noise scale for rotations (frame noise). Range 0.0–1.0. | 1.0 |
| final_noise_scale_frame | True | FLOAT | Final-step (t=1) noise scale for rotations (frame noise). Range 0.0–1.0. | 1.0 |
| frame_noise_schedule_type | True | ENUM | Interpolation schedule between initial and final rotation noise scales. Allowed values: linear, constant. | constant |
Outputs¶
| Field | Type | Description | Example |
|---|---|---|---|
| denoiser_config | JSON | A configuration dictionary with translation and rotation noise scales and schedules to feed into RF Diffusion. | {"noise_scale_ca": 1.0, "final_noise_scale_ca": 0.8, "ca_noise_schedule_type": "linear", "noise_scale_frame": 1.0, "final_noise_scale_frame": 0.8, "frame_noise_schedule_type": "linear"} |
Important Notes¶
- Values for noise scales should be between 0.0 and 1.0; higher values inject more noise.
- Schedule type controls how values change from initial to final: constant keeps the same value; linear interpolates across steps.
- This node only constructs configuration; it has effect only when connected to the denoiser_config input of the RF Diffusion node.
- All inputs are required by this node; defaults are provided for convenience.
Troubleshooting¶
- If changes have no visible effect, ensure the output is connected to the RF Diffusion node's denoiser_config input.
- If you see validation errors setting values, keep noise scales within 0.0–1.0 and use allowed schedule values: linear or constant.
- If generation quality degrades, try reducing noise scales or switch schedules from linear to constant to stabilize sampling.
Example Pipelines¶
