How to Use SwarmUI & Stable Diffusion 3 on Cloud Services Kaggle (free), Massed Compute & RunPod

Tutorial Video Link : https://youtu.be/XFUZof6Skkw

This video provides a comprehensive guide on installing and utilizing #SwarmUI on various cloud services. It's particularly valuable for those without access to a powerful GPU or seeking additional GPU power. The tutorial covers the implementation of SwarmUI, a leading Generative AI interface, on platforms such as Massed Compute, RunPod, and Kaggle (which offers complimentary dual T4 GPU access for 30 hours weekly). This instructional content aims to make using SwarmUI on cloud GPU providers as straightforward and efficient as on a local PC. Additionally, it demonstrates how to employ Stable Diffusion 3 (#SD3) in cloud environments. It's worth noting that SwarmUI operates on a #ComfyUI backend.

🔗 Access the Video-Referenced Public Post (no login or account required) with Relevant Links ➡️ https://www.patreon.com/posts/stableswarmui-3-106135985

🔗 Windows Tutorial for SwarmUI Usage ➡️

🔗 Tutorial on Rapid Model Downloads for Massed Compute, RunPod, and Kaggle, plus Swift Model/File Uploads to Hugging Face ➡️

🔗 Join the SECourses Discord Community ➡️ https://discord.com/servers/software-engineering-courses-secourses-7727740977340...

🔗 Stable Diffusion GitHub Repository (Please Star, Fork, and Watch) ➡️ https://github.com/FurkanGozukara/Stable-Diffusion

Exclusive Discount Code for Massed Compute: SECourses

This coupon is applicable to both Alt Config RTX A6000 and RTX A6000 GPUs

0:00 Introduction to the SwarmUI cloud services tutorial (Massed Compute, RunPod & Kaggle)

3:18 SwarmUI installation and usage guide for Massed Compute virtual Ubuntu machines

4:52 ThinLinc client synchronization folder setup for Massed Compute virtual machine access

6:34 Connecting to and initiating use of a running Massed Compute virtual machine

7:05 One-click SwarmUI update process on Massed Compute

7:46 Configuring multiple GPUs on SwarmUI backend for simultaneous image generation

7:57 Monitoring GPU status using the nvitop command

8:43 Overview of pre-installed Stable Diffusion models on Massed Compute

9:53 Massed Compute's new model download speed demonstration

10:44 Identifying and addressing GPU backend setup errors in a 4-GPU configuration

11:42 Monitoring the status of all four active GPUs

12:22 SD3 image generation and step speed on Massed Compute's RTX A6000

12:50 CivitAI API key setup for accessing gated models

13:55 Quick method for downloading generated images from Massed Compute

15:22 Installing the latest SwarmUI on RunPod with proper template selection

16:50 Port configuration for SwarmUI connection post-installation

17:50 Downloading and executing the RunPod SwarmUI installer script

19:47 Resolving the "backends loading forever" error through Pod restart

20:22 Reinitiating SwarmUI on RunPod

21:14 Downloading and implementing Stable Diffusion 3 (SD3) on RunPod

22:01 Setting up a multi-GPU backend system on RunPod

23:22 SD3 generation speed assessment on RTX 4090

24:04 Efficient bulk download of generated images from RunPod

24:50 SwarmUI and Stable Diffusion 3 installation and usage on a free Kaggle account

28:39 Modifying SwarmUI's model root folder path on Kaggle for temporary disk space utilization

29:21 Adding a secondary backend to leverage Kaggle's second T4 GPU

29:32 Canceling and restarting SwarmUI runs on Kaggle

31:39 Implementing and generating images with Stable Diffusion 3 on Kaggle

33:06 Troubleshooting and resolving out-of-RAM errors on Kaggle

33:45 Disabling one backend to prevent RAM errors when using T5 XXL text encoder twice

34:04 Evaluating Stable Diffusion 3 image generation speed on Kaggle's T4 GPU

34:35 Efficiently downloading all Kaggle-generated images to your local device

1. Introduction

In this article, a comprehensive guide is provided on how to use SwarmUI, Stable Diffusion 3, and other Stable Diffusion models on various cloud computing platforms. The tutorial covers three main options for running these powerful AI image generation tools without requiring a high-end GPU:

1.1 Massed Compute

Massed Compute is introduced as the cheapest and most powerful cloud server provider. The process of setting up and using SwarmUI on Massed Compute is explained in detail, including how to deploy virtual machines with pre-installed software.

1.2 RunPod

The second part of the tutorial covers how to set up and use SwarmUI on RunPod, another cloud service provider offering high-performance GPUs for AI tasks.

1.3 Kaggle

If you want to use SwarmUI on a free Kaggle account, the last part of the tutorial fully covers this option, allowing users to leverage Kaggle's free GPU resources.

Before diving into the specifics of each platform, the tutorial emphasizes the importance of first watching a comprehensive 90-minute SwarmUI tutorial for Windows users. This foundational knowledge is crucial for understanding how to use SwarmUI effectively, regardless of the platform.

1. Massed Compute Setup and Usage

Massed Compute is highlighted as an excellent option for running SwarmUI and Stable Diffusion models due to its competitive pricing and powerful hardware offerings. The process of setting up and using SwarmUI on Massed Compute is explained in detail:

2.1 Registration and Deployment

Use the specially given link for registration to register on Massed Compute. After registration, users need to enter their billing information and load some balance to their account. To deploy a virtual machine, navigate to the "deploy" section on the Massed Compute dashboard. There are two configuration options available for the RTX A6000 GPU: the standard config and an alternative config with different RAM amounts. Users can choose based on their specific needs and availability.

The tutorial provides a special coupon code that reduces the hourly rate from $2.5 to $1.25 for the RTX A6000 GPU, making it a very cost-effective option compared to other cloud providers. To deploy the virtual machine, users should select the "creator" category and choose the "SE courses" image. After applying the coupon code, click "deploy" to start the instance.

2.2 Connecting to the Virtual Machine

To connect to the deployed virtual machine, users need to download and install the ThinLinc client. The tutorial provides step-by-step instructions for downloading, installing, and configuring the ThinLinc client for various operating systems.

An important step in the configuration process is setting up a synchronization folder for file transfers between the local machine and the virtual machine. This is done through the ThinLinc client's options menu, where users can specify a local folder for synchronization and set appropriate permissions.

Once the ThinLinc client is set up, users can connect to their Massed Compute virtual machine using the provided IP address, username, and password. Upon successful connection, users will see a Linux desktop environment with pre-installed applications, including SwarmUI.

2.3 Updating and Starting SwarmUI

Before using SwarmUI, it's recommended to update it to the latest version. This can be done by double-clicking the updater button on the desktop. The update process is automatic and will start SwarmUI once completed.

2.4 Using Multiple GPUs

To fully utilize the power of multiple GPUs on Massed Compute, users need to configure additional backends in SwarmUI. This process involves adding ComfyUI self-starting backends and assigning them to different GPUs. The tutorial provides detailed instructions on how to add and configure these additional backends, ensuring that each GPU is utilized effectively.

2.5 Generating Images with SwarmUI

The tutorial demonstrates how to generate images using various Stable Diffusion models, including SDXL and Stable Diffusion 3. Users can select different models, adjust parameters like sampling methods and step counts, and generate multiple images simultaneously across multiple GPUs.

The article showcases the impressive speed of image generation on Massed Compute, with examples of generating 100 images using Stable Diffusion 3 across multiple GPUs. The tutorial also explains how to monitor GPU usage and generation speeds using tools like nvitop and SwarmUI's built-in logging features.

2.6 Downloading Generated Images

To download generated images from the Massed Compute virtual machine to a local computer, users can utilize the synchronization folder set up earlier with the ThinLinc client. The tutorial provides step-by-step instructions on how to copy the output folder to the synchronization directory, making it easy to access generated images on the local machine.

2.7 Using CivitAI API for Model Downloads

A new feature introduced in SwarmUI is the ability to use a CivitAI API key for downloading gated models. The tutorial explains how to obtain an API key from CivitAI and add it to the SwarmUI user settings, enabling users to download a wider range of models directly through the SwarmUI interface.

1. RunPod Setup and Usage

The second part of the tutorial focuses on setting up and using SwarmUI on RunPod, another cloud service provider that offers high-performance GPUs for AI tasks. The process is similar to Massed Compute but with some key differences:

3.1 Registration and Pod Deployment

Use the provided link to register for a RunPod account. After registration and setting up billing, users can deploy a pod (virtual machine) with their desired GPU configuration. The tutorial recommends using the "Community Cloud" option and selecting the "extreme speed" filter for optimal performance.

For the tutorial, a configuration with 3x RTX 4090 GPUs and 48GB of RAM is selected. The important step here is choosing the correct template: "RunPod PyTorch 2.1 with CUDA 11.8". This template ensures compatibility with all required applications.

3.2 Installing SwarmUI

Unlike Massed Compute, RunPod requires manual installation of SwarmUI. The tutorial provides a modified installation script that works specifically for RunPod. Users need to upload this script to their RunPod instance and execute it through the JupyterLab interface.

The installation process involves several steps, including downloading base models and setting up the SwarmUI environment. The tutorial guides users through each step, explaining how to monitor the installation progress and troubleshoot any issues that may arise.

3.3 Configuring SwarmUI for Multiple GPUs

Similar to the Massed Compute setup, users can configure SwarmUI to utilize multiple GPUs on RunPod. The process involves adding additional backends and assigning them to different GPUs through the SwarmUI interface.

3.4 Generating Images and Performance

The tutorial demonstrates generating images with various Stable Diffusion models on RunPod, showcasing the impressive speed and performance of the RTX 4090 GPUs. Users can monitor the generation process and performance metrics through the SwarmUI interface and RunPod's built-in monitoring tools.

3.5 Downloading Generated Images

For downloading generated images from RunPod, the tutorial suggests several methods, including using RunPod's built-in file browser, uploading to Hugging Face, or using the RunPodCTL tool. Each method is briefly explained, with references to more detailed instructions in separate tutorials.

1. Kaggle Free Account Setup and Usage

The final part of the tutorial covers how to set up and use SwarmUI on a free Kaggle account, allowing users to leverage Kaggle's free GPU resources for AI image generation:

4.1 Setting Up Kaggle Notebook

To use SwarmUI on Kaggle, users need to create a new notebook and import a specially prepared Kaggle notebook file provided in the tutorial. This notebook contains all the necessary setup instructions and code cells for installing and running SwarmUI on Kaggle's infrastructure.

4.2 Configuring GPU Resources

Users need to select the GPU accelerator option (T4 x2) to utilize Kaggle's free GPU resources. The tutorial explains how to properly configure the notebook settings to ensure both GPUs are available for use.

4.3 Downloading Models

The first step in the Kaggle setup involves downloading the desired Stable Diffusion models. The tutorial provides code cells for downloading multiple models, including SDXL and Stable Diffusion 3, to Kaggle's temporary storage.

4.4 Installing and Configuring SwarmUI

The installation process on Kaggle involves running several code cells that download and set up SwarmUI. The tutorial guides users through each step, including how to access the SwarmUI interface through Kaggle's proxy system.

An important configuration step specific to Kaggle is changing the model root directory to point to Kaggle's temporary storage where the models were downloaded. This ensures SwarmUI can access the models properly within Kaggle's environment.

4.5 Using SwarmUI on Kaggle

Once set up, users can generate images using SwarmUI on Kaggle similarly to other platforms. The tutorial demonstrates generating images with both SDXL and Stable Diffusion 3 models, showcasing the capabilities of Kaggle's free GPU resources.

However, it's noted that when using more complex models like Stable Diffusion 3 with additional text encoders, users may need to disable one of the GPUs due to memory constraints on Kaggle's free tier.

4.6 Downloading Generated Images

To download generated images from Kaggle, the tutorial provides a code cell that zips all generated images. Users can then download this zip file directly from the Kaggle notebook interface.

1. Additional Resources and Community

The tutorial concludes by encouraging users to engage with the community and access additional resources:

5.1 Discord Community

Users are invited to join a Discord server with over 7,000 members, where they can chat, ask questions, and interact with other SwarmUI users and developers.

5.2 GitHub Repository

The tutorial highlights the SwarmUI GitHub repository, encouraging users to star, fork, and watch the project for updates and contributions.

5.3 Patreon Exclusive Content

For those interested in more in-depth content, the tutorial mentions a Patreon exclusive post index available on GitHub, where users can preview Patreon-only tutorials and guides.

1. Conclusion

This comprehensive tutorial provides users with multiple options for running SwarmUI and Stable Diffusion models on cloud platforms, catering to different needs and budgets. Whether using the cost-effective Massed Compute, the high-performance RunPod, or the free resources on Kaggle, users can now access powerful AI image generation tools without the need for expensive local hardware.

The step-by-step instructions, coupled with explanations of potential issues and their solutions, make this tutorial an invaluable resource for both beginners and experienced users looking to leverage cloud resources for AI image generation. By following this guide, users can set up their preferred environment and start generating high-quality images using state-of-the-art Stable Diffusion models through the user-friendly SwarmUI interface.