PreSampling (SDTurbo)(sdTurboSettings)

Description

PreSampling (SDTurbo) configures the sampler settings used by an SDTurbo pipeline before KSampler runs. It is designed for SDTurbo-style models that require very few sampling steps and near-unity CFG, while also offering optional latent upscaling and unsharp masking during the sampling process.

Inputs

• pipe (PIPE_LINE): The pipeline carrying the model, conditioning, latents, and other sampler state. This node modifies the pipeline's sampling settings and passes it forward.

• steps (INT): Number of sampling steps. Default is 1, with a range of 1 to 10. SDTurbo models are intended to run in very few steps.

• cfg (FLOAT): Classifier-free guidance strength. Default is 1, with a range of 0 to 100. A value near 1 is typical for SDTurbo models.

• sampler_name (COMBO): The sampler to use for denoising.

• eta (FLOAT): Stochastic parameter used by certain samplers. Default is 1, range 0 to 10, step 0.01. Higher values add more randomness to the sampling trajectory.

• s_noise (FLOAT): Amount of noise to inject during sampling. Default is 1, range 0 to 10, step 0.01. Only affects samplers that support stochastic noise.

• upscale_ratio (FLOAT): Scale factor applied to the latent during the upscaling phase. Default is 2, range 0 to 16, step 0.01. Set to 1 to disable upscaling.

• start_step (INT): Sampling step at which the upscaling phase begins. Default is 5, range 0 to 1000.

• end_step (INT): Sampling step at which the upscaling phase ends. Default is 15, range 0 to 1000.

• upscale_n_step (INT): Number of steps, or step interval, associated with the upscaling operation. Default is 3, range 0 to 1000.

• unsharp_kernel_size (INT): Kernel size used for the unsharp mask applied after upscaling. Default is 3, range 1 to 21.

• unsharp_sigma (FLOAT): Sigma value for the unsharp mask's Gaussian blur. Default is 0.5, range 0 to 10, step 0.01.

• unsharp_strength (FLOAT): Strength of the unsharp mask. Default is 0, range 0 to 10, step 0.01. A value of 0 disables the unsharp effect.

• seed (INT): Seed used for noise generation in the sampler. Set a fixed value for reproducible results.

Outputs

• pipe (PIPE_LINE): The updated pipeline with the configured sampling settings, ready to be passed to a sampler node or subsequent processing nodes.

Usage Notes

This node is intended for SDTurbo workflows where the model already expects low step counts and low CFG. Keep steps and cfg near their defaults unless you are intentionally experimenting.

The eta and s_noise parameters only influence samplers that support stochastic or ancestral sampling; deterministic samplers may ignore them.

The upscaling parameters take effect during the sampling process. upscale_ratio controls the scale of the latent, while start_step and end_step define the step range in which the upscale is applied. unsharp_strength must be greater than 0 for the unsharp mask to have any effect; unsharp_kernel_size and unsharp_sigma control the sharpening kernel.