ComfyUI Extension: ComfyUI-ppm

ComfyUI-ppm Introduction

ComfyUI-ppm is an extension designed to enhance the functionality of ComfyUI by providing a collection of modified, fixed, and newly created nodes. These nodes are tailored to improve compatibility, add new features, and solve common issues faced by AI artists when working with prompts and models. By integrating ComfyUI-ppm into your workflow, you can achieve more precise control over your prompts, reduce unwanted effects, and streamline your creative process.

How ComfyUI-ppm Works

ComfyUI-ppm works by introducing new nodes and modifying existing ones to enhance their functionality and compatibility. Nodes are the building blocks in ComfyUI that perform specific tasks, such as processing prompts or generating images. By using these nodes, you can create complex workflows that allow for greater customization and control over your AI-generated art.

For example, the CLIPNegPip node allows you to use negative weights in your prompts, which can help in reducing or eliminating unwanted elements in your generated images. Similarly, the AttentionCouplePPM node lets you define specific regions in your image to focus or defocus attention, providing more control over the final output.

ComfyUI-ppm Features

CLIPNegPip

The CLIPNegPip node is a modified implementation that allows you to use negative weights in your prompts. This can be particularly useful for removing unwanted elements from your images. To use it, connect the CLIPNegPip node before other model or clip patches. You can then enter negative weights in your prompts, such as (worst quality:-1.3). This node increases compatibility with other nodes by using ModelPatcher instead of monkey-patching.

AttentionCouplePPM

The AttentionCouplePPM node is a modified version of the AttentionCouple node, compatible with CLIPNegPip and PatchModelAddDownscale (Kohya Deep Shrink) nodes. It allows you to add or remove regions by right-clicking the node and selecting Add Region or Remove Region. You can use multiple LatentToMaskBB nodes to set bounding box masks, providing precise control over which parts of the image to focus on.

CFG++SamplerSelect

This node adapts samplers to the CFG++ method, which helps reduce overexposure effects in your images. Use it together with the SamplerCustom node and set the CFG scale to 1.0-2.0 and PAG scale to 0.5-1.0 for optimal results.

Guidance Limiter

The Guidance Limiter node implements a method to apply guidance in a limited interval, improving the quality of samples and distribution in diffusion models. It also includes RescaleCFG functionality.

Empty Latent Image (Aspect Ratio)

This node generates an empty latent image with a specified aspect ratio, respecting the target resolution. It is useful for creating base images with specific dimensions.

CLIPTextEncodeBREAK

A lightweight wrapper over the ConditioningConcat node, CLIPTextEncodeBREAK can split prompts by the BREAK keyword into chunks and produce a single concatenated conditioning. This is useful for managing complex prompts.

CLIPTokenCounter

The CLIPTokenCounter node counts tokens in your prompt and returns them as a string. It can also print the token count and individual tokens if debug_print is enabled.

ComfyUI-ppm Models

ComfyUI-ppm does not introduce new models but enhances the functionality of existing ones through its nodes. Each node is designed to work seamlessly with the models you are already using, providing additional features and customization options.

What's New with ComfyUI-ppm

Updates and Changes

• CLIPNegPip: Now uses ModelPatcher for increased compatibility.
• AttentionCouplePPM: Added compatibility with CLIPNegPip and PatchModelAddDownscale (Kohya Deep Shrink).
• CFG++SamplerSelect: Adapted samplers to reduce overexposure effects.
• Guidance Limiter: Implemented limited interval guidance and RescaleCFG functionality.
• Empty Latent Image (Aspect Ratio): Generates empty latent images with specified aspect ratios.
• CLIPTextEncodeBREAK: Splits prompts by BREAK keyword for better prompt management.
• CLIPTokenCounter: Counts and prints tokens in prompts. These updates provide AI artists with more tools and flexibility, improving the overall experience and quality of generated images.

Troubleshooting ComfyUI-ppm

Common Issues and Solutions

1. Node Errors After Update:

• Solution: If a node starts throwing errors after an update, try re-adding the node to your workflow.

2. Negative Weights Not Working:

• Solution: Ensure that the CLIPNegPip node is connected before other model or clip patches. Check that the weights are entered correctly, e.g., (worst quality:-1.3).

3. AttentionCouplePPM Region Issues:

• Solution: Right-click the node and select Add Region or Remove Region to adjust the regions. Use LatentToMaskBB nodes to set bounding box masks accurately.

Frequently Asked Questions

• Q: How do I use negative weights in my prompts?
• A: Connect the CLIPNegPip node before other model or clip patches and enter negative weights in your prompts.
• Q: How can I focus on specific regions in my image?
• A: Use the AttentionCouplePPM node and LatentToMaskBB nodes to define and adjust regions.

Learn More about ComfyUI-ppm

For additional resources, tutorials, and community support, consider exploring the following:

• ComfyUI Documentation
• NegPiP Original Repository
• CFG++: Manifold-constrained Classifier Free Guidance for Diffusion Models
• Applying Guidance in a Limited Interval Improves Sample and Distribution Quality in Diffusion Models These resources provide valuable insights and support to help you make the most of ComfyUI-ppm in your AI art projects.