KSamplerAdvanced (Inspire) Node Documentation

Overview

The KSamplerAdvanced (Inspire) node is part of the ComfyUI-Inspire-Pack, designed to enhance the functionality of the standard KSampler node in ComfyUI. This advanced version aims to provide more control and flexibility, specifically tailored for compatibility with the A1111 interface, while introducing additional features for noise and seed management. It maintains key functionalities required to manipulate latent images during sampling, thus allowing for more varying and detailed image generation processes.

Functionality

The primary function of the KSamplerAdvanced (Inspire) node is to perform image generation by adding controlled noise to latent images and subsequently applying a series of transformations. It uses user-defined parameters, such as the model, seed, CFG scale, and a variety of noise options, to output refined latent images. This node is particularly advantageous for users requiring intricate control over noise application, batch processing, and using specialized seeds during the sampling process.

Inputs

The KSamplerAdvanced (Inspire) node accepts the following required and optional inputs:

• Required Inputs:

  • model: The model used for sampling.
  • add_noise: A boolean that indicates whether noise should be added.
  • noise_seed: The seed value for generating the initial noise applied to the latent.
  • steps: Number of steps to run the sampler.
  • cfg: The CFG (Class-Free Guidance) scale value.
  • sampler_name: The name of the sampler to be used.
  • scheduler: The scheduler that determines the sampling schedule.
  • positive & negative: Conditioning elements for guiding the image generation.
  • latent_image: The latent image that serves as the base for applying noise and transformations.
  • start_at_step: The step where sampling should begin.
  • end_at_step: The step where sampling should end.
  • noise_mode: The mode of noise application (options include CPU and GPU modes).
  • return_with_leftover_noise: A boolean indicating whether to return the latent with leftover noise.
  • batch_seed_mode: The mode for handling seed applications across batches.
  • variation_seed: The seed for introducing variations in noise generation.
  • variation_strength: The strength to determine how much variation seed influences the generated noise.

• Optional Inputs:

  • variation_method: The method (e.g., linear, slerp) used for applying variations.
  • noise_opt: Optional pre-generated noise image.
  • scheduler_func_opt: An optional scheduler function.
  • internal_seed: An optional seed used for intermediate noise generation, particularly in ancestral and SDE-based samplers.

Outputs

• LATENT: The refined latent image after applying all the specified transformations and noise.

Usage in ComfyUI Workflows

The KSamplerAdvanced (Inspire) node can be used in ComfyUI workflows where precise control over noise and seed manipulation is essential. This is crucial for complex image generation tasks that require consistency across batches or additional variation within the same base image.

Common use cases include:

• Creating complex images with layer-specific noise variations.
• Consistent batch processing where each output needs distinct variation seeds and strengths.
• Advanced artistic endeavors where the subtle and precise manipulation of latent spaces is required.
• Ensuring reproducibility for iterative tasks by setting specific seeds.

Special Features and Considerations

• A1111 Compatibility: This node is designed to mimic the noise generation process of A1111, using GPU for noise if specified, thus enabling better reproduction of results across different environments.
• Batch Handling: Through batch seed modes, users can avoid generating duplicate images due to seed duplication.
• Variation Control: The node allows fine-tuning of how much influence a variation seed should have on the image, using parameters like variation_strength and variation_method.
• Intermediate Seed Use: internal_seed provides users the capability to generate noise during intermediate steps, crucial for specific sampler types.
• Return with Noise: The node offers the choice to output the latent images along with residual noise, offering more flexibility post-sampling.

This node caters to advanced users who require detailed control over image sampling processes, ensuring versatility and control in complex ComfyUI workflows.