This node is designed to enhance a model’s sampling capabilities by integrating continuous EDM (Energy-based Diffusion Models) sampling techniques.
This node is designed to enhance a model’s sampling capabilities by integrating continuous EDM (Energy-based Diffusion Models) sampling techniques. It allows for the dynamic adjustment of the noise levels within the model’s sampling process, offering a more refined control over the generation quality and diversity.
Inputs
Parameter
Description
Data Type
Python dtype
model
The model to be enhanced with continuous EDM sampling capabilities. It serves as the foundation for applying the advanced sampling techniques.
MODEL
torch.nn.Module
sampling
Specifies the type of sampling to be applied, either ‘eps’ for epsilon sampling or ‘v_prediction’ for velocity prediction, influencing the model’s behavior during the sampling process.
COMBO[STRING]
str
sigma_max
The maximum sigma value for noise level, allowing for upper bound control in the noise injection process during sampling.
FLOAT
float
sigma_min
The minimum sigma value for noise level, setting the lower limit for noise injection, thus affecting the model’s sampling precision.
FLOAT
float
Outputs
Parameter
Description
Data Type
Python dtype
model
The enhanced model with integrated continuous EDM sampling capabilities, ready for further use in generation tasks.
MODEL
torch.nn.Module
This documentation was AI-generated. If you find any errors or have suggestions for improvement, please feel free to contribute! Edit on GitHub