SeargeSDXLSampler2 Node Documentation

Overview

The SeargeSDXLSampler2 node is part of the Searge-SDXL custom node extension for ComfyUI, a user interface for Stable Diffusion models. This node is designed to implement a sampling process for images using both a base model and a refiner model, which can enhance the image generation process by refining the output according to specific conditions and configurations.

Functionality

The SeargeSDXLSampler2 node performs a dual-phase sampling process:

1. Base Sampling Phase: The process begins by generating an image using the base model. This initial phase accounts for a defined portion of the total sampling steps.

2. Refining Phase: After the initial image is generated, the refiner model is applied to further refine the output, focusing on enhancing details and quality. This phase uses the remaining steps of the total sampling process.

Inputs

The node requires several inputs to function effectively:

• Base Model (base_model): The initial model used for generating the base image.
• Base Positive Conditioning (base_positive): Positively conditioned data for the base model.
• Base Negative Conditioning (base_negative): Negatively conditioned data for the base model.
• Refiner Model (refiner_model): The model used to refine the initial output.
• Refiner Positive Conditioning (refiner_positive): Positively conditioned data for the refiner model.
• Refiner Negative Conditioning (refiner_negative): Negatively conditioned data for the refiner model.
• Latent Image (latent_image): The latent space image data to be processed.
• Noise Seed (noise_seed): An integer seed used for generating random noise, influencing the variability of the outputs.
• Steps (steps): Total number of sampling steps to be performed by both models combined.
• Cfg (cfg): The classifier-free guidance scale that adjusts how strongly the models adhere to the conditioning.
• Sampler Name (sampler_name): The algorithm used for iterative image generation.
• Scheduler (scheduler): The scheduling of steps during sampling.
• Base Ratio (base_ratio): Proportion of sampling steps allocated for the base model.
• Denoise (denoise): The intensity of the denoising process applied during sampling.

Optional Inputs

• Refiner Prep Steps (refiner_prep_steps): Preparatory steps for the refiner phase.
• Noise Offset (noise_offset): Adjustment to the noise seed for the refiner phase.
• Refiner Strength (refiner_strength): Intensity of influence of the refiner model.

Outputs

• Latent (LATENT): The output of the sampling process is a latent representation of the generated or refined image, ready for further processing or visualization.

Usage in ComfyUI Workflows

The SeargeSDXLSampler2 node can be seamlessly integrated into ComfyUI workflows where both base and refined image generation are desirable. It allows users to harness the capabilities of dual-model sampling, which enhances image features and quality, making it suitable for diverse applications requiring varying levels of detail and refinement.

Special Features and Considerations

• Dual-Phase Sampling: The distinct use of both a base model and a refiner model offers more control over image quality and features.
• Flexible Configuration: Adjustable inputs such as base_ratio and refiner_strength provide users with extensive control over the sampling process.
• Denoising Control: The denoise parameter lets users fine-tune the degree of noise reduction applied during image generation, impacting the smoothness and detail of the output.

By incorporating the SeargeSDXLSampler2 node into workflows, users can achieve sophisticated image outputs, customizing the sampling approach according to specific project requirements or artistic intentions. For further guidance and updates, users can refer to the SeargeSDXL GitHub repository.