SDXS: 90 millisecond renders with comfyUI

Stephan Tual
3 Apr 202407:21

TLDRThe video discusses the challenges of slow渲染 speeds with KY due to numerous custom nodes and offers a solution by reinstalling KY and properly configuring the model path. The use of SDXS technology is highlighted for its rapid image generation, and the video concludes by demonstrating the improved performance with the new setup, achieving faster渲染 times.


  • 🚀 The video discusses a solution for slow loading times in KY's interface, particularly due to a high number of custom nodes and models.
  • 🛠️ To address performance issues, the speaker suggests using developer tools to identify problematic functions, but acknowledges this as a temporary fix.
  • 🎯 The speaker introduces SD XS, a technology that rapidly generates images, albeit with moderate quality, and demonstrates its speed in rendering.
  • 🔧 A more permanent solution is proposed, which involves reinstalling KY and managing the custom nodes and models more efficiently.
  • 📂 The video provides a step-by-step guide on reinstalling KY, including downloading the latest version and setting up the correct model paths.
  • 📈 The importance of correctly configuring the base path in the yaml file is emphasized, to ensure the new KY installation can access the existing models.
  • 🚀 After the reconfiguration, the new KY installation loads significantly faster, with a noticeable improvement in rendering times.
  • 📊 The comparison between the old and new installations shows a substantial decrease in image generation time, from 0.19-0.20 seconds to 0.09-0.10 seconds.
  • 🔍 The use of verbose command line parameters is recommended to understand and diagnose what KY is doing during its operation.
  • 💡 The video concludes by highlighting the potential for even faster rendering times by batching operations and using a clean KY installation.

Q & A

  • What is the main issue discussed in the transcript?

    -The main issue discussed is the slow loading time and performance problems of an existing Confy UI installation, specifically in the browser, which takes up to 5 seconds to load the GUI.

  • What is the problem with the custom nodes and models in the user's situation?

    -The high number of custom nodes and models is causing performance issues, making the interface and rendering slow due to the excessive load and time consumption.

  • How can one identify the function causing performance issues?

    -By using developer tools to select the function that's currently causing trouble, which can help identify the source of performance problems.

  • What is a quick fix for slow interface performance?

    -Disabling the previews in the manager can be a quick fix, but it is more of a temporary hack and the previews are still useful for the user.

  • What does the user plan to do to solve the performance issue?

    -The user plans to kill the current server instance, download a fresh copy of Ky, and set it up properly to avoid the performance issues associated with the high number of custom nodes.

  • What is the purpose of using the 'verbose' command line parameter?

    -Using the 'verbose' command line parameter allows Ky to provide detailed information about its operations, such as where it's getting its models from, which can help diagnose and fix path-related issues.

  • How does the user ensure the new Ky installation reads from the old folder containing the custom models?

    -The user edits the 'extra_model_path.yml' file in the new Ky installation's 'config UI' folder, setting the base path to the location of the old 'confy UI' folder containing the custom models.

  • What is the significant improvement in the new Ky installation's performance?

    -The new Ky installation loads in microseconds, significantly faster than the old version, and it can generate images at a rate of 10 frames per second, which is a substantial improvement over the previous 0.19 to 0.20 seconds per image.

  • How can the user further increase the speed of image generation?

    -The user can increase the speed by using an empty latent image batch of 100, processing them all at once, which can result in faster generation times.

  • What is the main advantage of using SDXS technology?

    -The main advantage of using SDXS technology is its ability to quickly generate images using a one-step diffusion process, although the quality may be less than ideal.

  • What is the final recommendation for users with performance issues?

    -The final recommendation is to use a clean Ky installation, avoid unnecessary custom nodes and models, and follow the steps outlined in the transcript to set up the new installation properly for improved performance.



🚀 Optimizing Interface Performance

The paragraph discusses the challenges of managing a large number of custom nodes and IP adapters, which can slow down the interface and rendering performance. The speaker shares their personal experience with a slow-loading GUI and proposes a solution to improve efficiency. They demonstrate using SD XS, a technology that rapidly generates images, albeit with lower quality. The main focus is on streamlining the setup by disabling previews and reinstalling the software to a clean state, which significantly improves load times and overall performance. The speaker also emphasizes the importance of correctly configuring the file paths to ensure the new installation can access the existing models.


🎨 Dynamic Image Generation with SD XS

This paragraph showcases the capabilities of SD XS for dynamic image generation, highlighting its speed and adaptability. The speaker changes the prompt to generate images of dogs and then to create a more complex scene involving a cat with wings flying in the sky. The emphasis is on the real-time image generation, where the visuals change as the prompt is typed. While the quality may not be the highest, the speed of generation is impressive, with the potential to reach up to 100 frames per second. The paragraph also compares the performance of a fresh installation with that of a cluttered one, demonstrating the significant improvement in image generation speed when using a clean setup.




SDXS refers to a technology mentioned in the video that uses a one-step diffusion process to rapidly generate images. Although the quality of the images produced by SDXS may not be the highest, its main advantage lies in the speed at which it can produce these images, measured in 'bears per second' in the context of the video. This technology is particularly useful for those who require fast rendering capabilities, as demonstrated by the user's experience of generating 100 images at a time in about 10 seconds.


KY, as used in the video, appears to be a software or platform that the speaker is using for some form of graphical or interface-based work. The speaker mentions issues with KY's loading times, particularly in the browser, which can take up to 5 seconds. The speaker's goal is to improve the performance of KY, not just for the interface but also for the rendering process. This keyword is central to the video's theme of optimizing performance and resolving issues related to slow loading times.


confyUI is referenced as a user interface that the speaker is working with, which seems to have issues with performance due to a high number of custom nodes and other elements running within it. The speaker describes the process of reinstalling confyUI to improve its loading times and performance. This keyword is integral to the video's narrative as it represents the software environment that the speaker is seeking to optimize.


In the context of this video, performance refers to the speed and efficiency with which the software or platform (KY and confyUI) operates. The speaker is concerned with the slow loading times and rendering issues that affect the overall performance. The video discusses various methods to enhance performance, such as reinstalling software, using the developer tools to identify problematic functions, and disabling previews. The keyword 'performance' is a core concept around which the video's theme of troubleshooting and optimizing software efficiency revolves.

💡custom nodes

Custom nodes are user-created components or extensions within the software that the speaker is using. The video mentions that the high number of custom nodes is contributing to the performance issues experienced with confyUI. The speaker needs to keep these custom nodes running for their work but also seeks ways to manage their impact on the software's performance. This keyword is significant as it represents the additional features or functionalities that users may incorporate into their software environment, which can sometimes lead to unintended performance consequences.


Previews in the context of the video refer to the visual representations or snapshots of the models or nodes within the software interface. The speaker mentions disabling previews as a temporary solution to improve performance, indicating that these visual elements consume resources and can slow down the software. However, they are also considered useful, so the speaker seeks a more permanent solution that doesn't involve completely removing them. The keyword 'previews' is relevant as it highlights the trade-off between performance optimization and maintaining the utility of visual aids in the software interface.

💡Nvidia GPU

Nvidia GPU refers to the graphics processing unit (GPU) produced by Nvidia, a company known for its high-performance GPUs used in computing and gaming. In the video, the speaker provides a tip about accessing the Nvidia GPU settings and using command line parameters to enhance the software's performance. The keyword 'Nvidia GPU' is significant as it represents the hardware component that can be optimized or configured to improve the rendering and overall performance of the software being discussed.

💡yaml file

A yaml file, short for 'YAML Ain't Markup Language' file, is a human-readable data serialization format often used for configuration files. In the video, the speaker instructs the viewer on how to modify a yaml file to specify the search path for models, which is crucial for the software to function correctly. The keyword 'yaml file' is important as it represents a key part of the software's configuration process that the user can manipulate to optimize performance and manage the software's behavior.

💡config UI

Config UI likely stands for 'configuration user interface' and refers to the interface within the software where users can adjust settings and preferences. The speaker mentions using the config UI to observe what the software is doing, particularly in terms of finding and using model paths. This keyword is significant as it relates to the process of customizing and fine-tuning the software to improve its performance and functionality according to the user's needs.


Autoq is a term used in the video to denote an auto-queue or automated queuing feature that allows for the batch processing of tasks. The speaker uses this feature to enhance the speed of image generation by processing multiple images at once. The keyword 'autoq' is relevant as it exemplifies one of the methods the speaker employs to optimize the rendering speed and overall performance of the software.

💡latent image batch

A latent image batch refers to a group of images that are processed together in a non-visible or 'latent' state until they are ready to be displayed or saved. In the video, the speaker mentions using a batch of 100 latent images to process them all at once, which significantly speeds up the rendering process. This keyword is important as it illustrates the speaker's strategy for achieving faster image generation by utilizing the software's batch processing capabilities.


The video discusses a solution for slow loading times in KY's interface due to a high number of custom nodes.

The issue is not entirely server-related but also due to the browser and specific elements like light graph.

Using developer tools can help identify problematic functions, such as those linked to custom nodes.

Disabling previews in the manager can be a temporary workaround for performance issues.

SDXS technology is introduced as a method to quickly generate images, albeit with varying quality.

The video demonstrates running a queue with SDXS, achieving fast rendering times.

A new approach is proposed, which involves killing the current server instance and downloading a fresh copy of KY.

The video provides a detailed guide on installing a new version of KY, including using the verbose command line parameter.

A method to specify the model path in the config UI folder is explained to avoid path-related issues.

The new KY installation shows a significant reduction in folder count, indicating fewer custom nodes.

The server starts instantly compared to the old version, and the verbose parameter reveals the search path being used.

The workflow is tested, and the desired models are found in the correct locations.

SDXS generates 100 images at a time, showing its speed in rendering.

The video shows real-time image generation with SDXS, changing prompts dynamically.

The speed of image generation is compared between the clean install and the usual packed version.

The clean install achieves 10 frames per second, a significant improvement over the old version.

Batching images into latent can further increase the speed of rendering.

The video concludes by highlighting the benefits of the new workflow and provides a link for further exploration.