KSampler (Efficient) Node Documentation

Overview

The KSampler (Efficient) node is a modified version of the KSampler designed to streamline and optimize the sampling process in ComfyUI workflows. This node provides enhanced functionality, including the ability to preview image generations and decode images using a VAE (Variational Autoencoder). Its design aims to reduce the need for additional nodes by combining multiple functions, making it a robust choice for efficient processing.

Features

1. Live Preview: Users can view image generations in real-time, which is useful for immediate feedback during the creative process.
2. Enhanced Seed Management: The node includes a seed management box that allows clear control over seed values. Setting the seed to -1 applies the selected seed behavior.
3. Script Execution: It supports executing various scripts, such as the XY Plot script, by connecting an input connection.

Inputs

Required Inputs

• Model: The model to be used for sampling.
• Seed: An integer value to initiate the random number generator. It defines the starting point of the sampling process.
• Steps: Defines the number of diffusion steps, determining the iterations for refining the image.
• CFG (Classifier-Free Guidance): A float value guiding the strength of adherence to the conditioning prompts.
• Sampler Name: Specifies the sampling method used for the operations.
• Scheduler: Describes the noise schedule method to dictate the noise reduction process over time.
• Positive Conditioning: Defines the conditions or prompts to positively influence the sampling output.
• Negative Conditioning: Refers to prompt conditions that negatively influence the sampling process, discouraging unwanted features.
• Latent Image: The starting latent image that the model builds upon through sampling.
• Preview Method: The strategy for generating previews (auto, latent2rgb, taesd, vae_decoded_only, none).
• VAE Decode: Determines whether to decode the VAE output (true, true (tiled), false).

Optional Inputs

• Optional VAE: Connects a different VAE for decoding purposes if the preview or output requires it.
• Script: Allows for custom script execution.
• Denoise: Controls the level of denoising during sampling, with a range from 0.0 to 1.0.

Hidden Inputs

• Prompt: Metadata information about the applied prompt.
• Extra PNG Info: Additional PNG metadata.
• Unique ID: A unique identifier for the node in a workflow, assisting in tracking and caching.

Outputs

• Model: Outputs the updated model post-sampling.
• Positive Conditioning: The altered positive conditioning based on sampling results.
• Negative Conditioning: The modified negative conditioning after processing.
• Latent: The final latent images obtained from the sampling process.
• VAE: The Variational Autoencoder used within the process.
• Image: The resulting image generated from the latent space, typically formatted as a decoded image.

Usage in ComfyUI Workflows

The KSampler (Efficient) node is pivotal in ComfyUI workflows, particularly when efficiency and streamlined operations are necessary. By integrating multiple functionalities into one node, it reduces the overall node count, thus simplifying and speeding up processing pipelines. It is used in workflows where live previews and quick iterations are desirable. Users can leverage its advanced seed management and script execution capabilities for custom, automated tasks in image generation pipelines.

Special Features and Considerations

• Robust Control: Provides extensive control over the sampling process, allowing users to fine-tune generation outputs closely.
• Efficiency: By combining multiple functions, it reduces node clutter and enhances efficiency.
• Compatibility: Integrates well with other nodes, especially script and plot nodes, to expand functional use cases like generating multiple image variations or experimental plots.

These attributes make the KSampler (Efficient) an invaluable asset to those working within the ComfyUI framework, pushing the boundaries of creative expression and technical efficiency.