Stable Cascade in ComfyUI with Updated Method and Custom Workflows

How Do?
5 Mar 202423:10

TLDRIn this informative video, the presenter introduces an updated method for utilizing Stable Cascade in ComfyUI, complete with custom workflows. They guide viewers through downloading specific checkpoints from the ComfyUI repository, explain the process of text-to-image and image-to-image workflows, and offer tips for improving the quality of generated images. The video also presents two original techniques to address common issues with Stable Cascade models, such as noise in textures and an oversharpened appearance, providing a comprehensive guide for users to enhance their ComfyUI experience.

Takeaways

  • 🔍 The video introduces an updated method for using Stable Cascade in ComfyUI with custom workflows.
  • 📂 Download the specific checkpoints for ComfyUI from the Stable Cascade examples page in the ComfyUI repo.
  • 📱 For Portable Windows users, place the downloaded checkpoints in the 'models' folder within the ComfyUI directory.
  • 🌐 Ensure ComfyUI is updated to work with the new nodes and features.
  • 🖼️ The text-to-image workflow involves creating a text encoding and feeding it into a sampler, followed by stages C, B, and A for image generation.
  • 🎨 The image-to-image workflow converts an initial image into latents for further processing with Stable Cascade models.
  • 🔧 A highres fix for Stable Cascade involves a two-pass process to improve image quality and reduce noise.
  • 🔄 The first pass of the highres fix is similar to the text-to-image workflow, and the second pass upsizes and denoises the image.
  • 🌟 The highres fix method improves both composition and detail, especially noticeable in lower resolution images.
  • 🛠️ An alternative workflow combines Stable Cascade with an XD XL model to achieve a specific aesthetic while maintaining composition.
  • 📸 A triple pass workflow for Stable Cascade involves upscaling in the latter two passes, resulting in cleaner and more detailed images.

Q & A

  • What is the main topic of the video?

    -The main topic of the video is the updated method for using Stable Cascade in ComfyUI with custom workflows.

  • Where can the checkpoints for ComfyUI be found?

    -The checkpoints for ComfyUI can be found on the Stable Cascade examples page in the ComfyUI repo.

  • What are the two models that need to be downloaded from the Hugging Face repo?

    -The specific models to be downloaded are not named in the script, but they are the ones available on the Hugging Face repo for Stable Cascade in ComfyUI.

  • How does one update ComfyUI to work with the new nodes?

    -To update ComfyUI, one should go to the manager, click on 'update ComfyUI', wait for a few seconds, exit the command prompt, and then restart ComfyUI.

  • What is the purpose of the new node for generating latent images?

    -The new node for generating latent images is used specifically for Stable Cascade models to create the initial stage of the image.

  • What is the significance of the compression factor in the workflow?

    -The compression factor determines the amount of compression for the Stage C in the generation process. Adjusting it can improve the image quality at different resolutions.

  • How does the text-to-image workflow function in the script?

    -The text-to-image workflow functions by creating a text encoding, feeding it into a sampler, generating the initial stage of the image, and then passing it through special Cascade conditioning into Stage B and A for the final image.

  • What is the highres fix method for Stable Cascade?

    -The highres fix method for Stable Cascade involves a two-pass process where the image is first generated, then upscaled and denoised before being passed through Stable Cascade again to improve detail and reduce noise.

  • What are the differences between the highres fix and the Stable Cascade to XD XL workflow?

    -The highres fix is a two-pass method focusing on improving the image quality and reducing noise, while the Stable Cascade to XD XL workflow involves passing the image through a two-stage control net to maintain composition and details before upscaling.

  • How does the triple pass workflow improve the final image?

    -The triple pass workflow improves the final image by running the image through three passes of Stable Cascade with upscaling in the last two, which refines the details and reduces noise, resulting in a cleaner and more detailed image.

  • What is the recommended denoising value for the highres fix method?

    -The recommended denoising value for the highres fix method is 0.5, but it can be adjusted between 0.2 and 0.6 depending on the image and desired outcome.

Outlines

00:00

🖼️ Introduction to Stable Cascade and Comfy UI Workflows

The video begins with an introduction to the updated method for using Stable Cascade in Comfy UI. The focus is on exploring the workflows provided by Comfy Anonymous in the comfi repo and sharing custom workflows. The first step is to ensure readiness for using the updated method by downloading checkpoints specifically made for Comfy UI from the Stable Cascade examples page in the Comfy UI repo. The video provides instructions on where to find and download these models and how to store them for use with Comfy UI. It also mentions the need to update Comfy UI to ensure compatibility with the new workflows.

05:02

📝 Text-to-Image Workflow with Stable Cascade

The second paragraph delves into the text-to-image workflow provided by Comfy Anonymous. It explains the process of using the Stage C model to create a text encoding, which is then fed into a sampler. The video highlights the importance of updating Comfy UI to work with these new nodes and provides a brief overview of the workflow stages. It also discusses the generation of latent images for Stable Cascade models, including the selection of width, height, and compression factors. The video demonstrates the workflow using a default prompt and shows the resulting image, emphasizing the capabilities of Stable Cascade with Comfy UI.

10:02

🖼️ Image-to-Image Workflow and Custom Methods

This paragraph covers the image-to-image workflow, which was not included in previous videos. It explains the process of converting images into latents for Stable Cascade models and the differences between this workflow and the text-to-image workflow. The video also introduces two custom methods devised to address the noise and oversharpened look in Stable Cascade images. The first method is a high-resolution fix, which involves a two-pass Stable Cascade process with upscaling and denoising. The second method involves using an XD XL model to achieve a specific aesthetic, and the video provides a brief overview of this workflow.

15:03

🔄 Stable Cascade Highres Fix and XD XL Workflow

The fourth paragraph elaborates on the Stable Cascade highres fix method, which aims to improve the quality of text and reduce noise in the generated images. It explains the process of upscaling the image and converting it back into latents for further refinement. The video also compares the results of the highres fix method with the initial generation, highlighting the improvements in text clarity and noise reduction. Additionally, the paragraph discusses the second custom method, which involves passing the image through a two-stage control net to maintain composition and adhere to the prompt. The video provides a quick walkthrough of this workflow and shares the results of the process.

20:04

🌐 Triple Pass Workflow for Widescreen Wallpapers

The final paragraph introduces an extension of the highres fix method, involving a triple pass in the Stable Cascade workflow. This method is specifically designed for creating widescreen wallpapers with high resolution. The video explains the process of manually setting the resolution for each pass and emphasizes the importance of maintaining a 16:9 ratio. The paragraph details the improvements in image quality, variation, and cleanliness achieved through the triple pass. The video concludes by demonstrating the final result and suggests upscaling the image for personal use.

Mindmap

Keywords

💡Stable Cascade

Stable Cascade is a method used in the video for generating images through a series of interconnected stages. It is a part of the ComfyUI system and is updated in the video to work with the latest version. The process starts with creating a text encoding or using an existing image, which is then fed into a sampler to generate the initial image. This image goes through stages C, B, and A, with each stage refining the image further. In the context of the video, Stable Cascade is used to create images from text prompts and to improve image quality through multiple passes.

💡ComfyUI

ComfyUI is a user interface or platform used for image generation, as discussed in the video. It provides a repository of workflows and models that users can download and utilize for their image creation tasks. The video specifically mentions an update to ComfyUI that allows for the use of Stable Cascade with custom workflows.

💡Checkpoints

Checkpoints in the context of the video refer to saved states or models that are used in the image generation process. They are crucial for the Stable Cascade method as they represent the different stages of the image creation process. Users are instructed to download specific checkpoints for ComfyUI to ensure compatibility with the updated method.

💡Workflows

Workflows are the step-by-step processes or sequences used to generate images with specific characteristics or meet certain objectives. In the video, workflows provided by ComfyUI are discussed, and the creator also shares custom workflows they have developed. These workflows guide users through the image generation process using Stable Cascade and ComfyUI.

💡Text-to-Image

Text-to-Image is a workflow where textual descriptions are used as inputs to generate corresponding images. In the video, this workflow is demonstrated using the Stable Cascade method within ComfyUI. The process involves creating a text encoding and feeding it into the sampler to generate an image based on the text prompt.

💡Image-to-Image

Image-to-Image is a workflow where an existing image is used as a base to generate a new or modified image. The video introduces an image-to-image workflow using Stable Cascade, where a base image is converted into latents and then processed through the Stable Cascade stages to produce a new image with different characteristics or improved quality.

💡Latent Images

Latent images refer to the intermediate representations of images in the image generation process. In the context of the video, these are created specifically for the Stable Cascade models. They are generated by compressing the input data and are used as a basis for further refinement through the Stable Cascade stages.

💡Compression Factor

The compression factor is a value used in the image generation process to control the level of compression applied to the latent images. Adjusting the compression factor can affect the quality and detail of the final image, with lower values resulting in less compression and potentially more detail, while higher values increase compression and can affect the image quality.

💡Denoising

Denoising is a process used to reduce or eliminate noise in images generated through the Stable Cascade method. In the video, denoising is applied as part of the highres fix workflow to improve the quality of the images by reducing noise patterns and artifacts, particularly in textures like grass and gravel.

💡Upscaling

Upscaling refers to increasing the resolution of an image. In the video, upscaling is used in the highres fix workflow to improve the quality of the images generated by the Stable Cascade method. The process involves enlarging the image and then converting it back into latents for further processing.

💡Custom Nodes

Custom nodes are additional tools or functionalities that can be installed in ComfyUI to extend its capabilities. In the video, the creator mentions using custom nodes for pre-processing in the image-to-image workflow, which involves using a two-stage control net to maintain the composition of the image.

Highlights

Updated method for using Stable Cascade in Comfort UI

Custom workflows for Stable Cascade in Comfort UI

Downloading checkpoints specific to Comfort UI from the examples page

Explanation of the text-to-image workflow provided by Comfy Anonymous

The process of creating a text encoding and feeding it into the sampler

The unique node for generating latent images for Stable Cascade models

Adjusting compression factor for different resolutions

Running the image through special Cascade conditioning into Stage B and Stage A

The highres fix method for Stable Cascade to improve image quality

Using a two-pass Stable Cascade workflow for better text and composition

The Stable Cascade to XD XL workflow for achieving a particular look

Utilizing custom nodes for the two-stage control net

The triple pass Stable Cascade method for generating high-quality wallpapers

Adjusting the denoising value for different outcomes

The importance of maintaining the same aspect ratio throughout the passes

The recommendation to experiment with compression and denoising values

The final result of a high-quality, detailed image suitable for wallpapers