TemporalScoreRescaling - ComfyUI Built-in Node Documentation

This node applies Temporal Score Rescaling (TSR) to a diffusion model.

This node applies Temporal Score Rescaling (TSR) to a diffusion model. It patches the model so that, during sampling, the predicted score or noise is rescaled to steer the diversity of generated results. This is implemented as a post-CFG (Classifier-Free Guidance) function.

Inputs

Parameter	Description	Data Type	Required	Range
model	The diffusion model to be patched with the TSR function.	MODEL	Yes	-
tsr_k	Controls the rescaling strength. Lower k produces more detailed results; higher k produces smoother results in image generation. Setting k = 1 disables rescaling. (default: 0.95)	FLOAT	Yes	0.01 - 100.0
tsr_sigma	Controls how early rescaling takes effect. Larger values take effect earlier. (default: 1.0)	FLOAT	Yes	0.01 - 100.0

Note: Rescaling is skipped when tsr_k is set to 1, when the sigma value is 0, or when the signal-to-noise ratio is 0.

Outputs

Output Name	Description	Data Type
patched_model	The input model, now patched with the Temporal Score Rescaling function applied to its sampling process.	MODEL

This documentation was AI-generated. If you find any errors or have suggestions for improvement, please feel free to contribute! Edit on GitHub

---

Source fingerprint (SHA-256): 4d4e3c64fb6e3a3fe4725ea944a361b46d871943a10e65d72d70e0e6d757dfca