SeargeSDXLImage2ImageSampler Node Documentation

Overview

The SeargeSDXLImage2ImageSampler is a custom node for ComfyUI, part of the Searge-SDXL custom nodes extension. It is designed to facilitate image-to-image (I2I) transformations leveraging both a base model and a refiner model. This node is particularly useful in enhancing image resolution, applying High-Resolution (HiRes) fixes, and conditioning image generation with specific inputs such as noise seed and step configurations.

Functionality

The primary purpose of the SeargeSDXLImage2ImageSampler node is to transform input images using a stable diffusion process with customizable sampling techniques. This process is enhanced by utilizing both a base SDXL model and a refiner SDXL model to achieve high-quality image output. The node also supports post-processing features like upscaling, which can be configured for improved image quality.

Inputs

The node accepts several inputs, each serving a distinct role in the image transformation process:

1. Base Model: The primary model used for generating the base latent representation of the image.
2. Base Positive & Negative Conditioning: Conditioning information for the base model positive and negative prompts.
3. Refiner Model: The model used to refine the base output, enhancing details and overall quality.
4. Refiner Positive & Negative Conditioning: Conditioning information for the refiner model's positive and negative prompts.
5. Image: The initial image that will be transformed by the node.
6. VAE (Variational Autoencoder): Used to encode and decode images into a latent space and back.
7. Noise Seed: Determines the randomness in image generation, allowing for controlled variability.
8. Steps: Number of steps in the diffusion process, impacting the level of detail.
9. CFG (Classifier-Free Guidance Scale): Balances the adherence to prompts with creativity, influencing the output image.
10. Sampler Name & Scheduler: Determines the type of sampler and scheduler used in the generative process.
11. Base Ratio: Dictates the proportion of steps allocated to the base model in the total sampling process.
12. Denoise: Controls the level of noise reduction during generation.
13. Optional Parameters:
    • Upscale Model: Model used for upscaling the image.
    • Scaled Width & Height: Target dimensions for upscaling.
    • Softness: Controls blending between the original and upscaled images.
    • Additional Parameters: such as Noise Offset, Refiner Strength (commented out).

Outputs

The node produces a single output:

• Image: The transformed image, processed through the image-to-image sampling pipeline with applied model transformations, upscaling, and any specified conditioning.

Usage in ComfyUI Workflows

In ComfyUI workflows, the SeargeSDXLImage2ImageSampler node can be employed in scenarios requiring complex image enhancements or generation tasks where both base and detailed refinements are needed. It's particularly suitable for workflows that:

• Transform existing images while adding creative alterations based on conditioning inputs.
• Require high-quality image outputs possibly with resolution enhancements.
• Integrate multiple model passes (base and refiner) for creating refined artistic styles.

Special Features and Considerations

• Dual-Model Support: Uses both a base and refiner model for separate stages of the generative process, resulting in detailed and context-aware image outputs.
• High-Resolution Fix: Provides options to upscale the input image, ensuring resolution is adjusted inline with the transformations applied.
• Modulation of Creative Output: Through variable steps, CFG scale, and conditioning inputs, users can balance creativity with precision according to their needs.
• Flexibility with Parameters: While it comes pre-set with defaults, the sampler offers flexibility through its optional parameters to fine-tune the output, significantly impacting the final aesthetic.

By understanding these capabilities and configurations, users can effectively integrate the SeargeSDXLImage2ImageSampler into ComfyUI workflows to achieve advanced image generation tasks tailored to specific needs.