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

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

This video offers a comprehensive guide on implementing and utilizing #SwarmUI on various cloud platforms. It's particularly valuable for those without access to high-powered GPUs or seeking enhanced GPU capabilities. The tutorial covers the installation and operation of SwarmUI, a leading Generative AI interface, on Massed Compute, RunPod, and Kaggle (which provides free dual T4 GPU access for 30 hours per week). You'll learn to deploy SwarmUI on cloud GPU providers with the same ease and efficiency as on a local machine. Additionally, the video demonstrates how to use Stable Diffusion 3 (#SD3) in cloud environments. It's worth noting that SwarmUI employs a #ComfyUI backend.

🔗 Access the Video's Public Post (no login required) with All Relevant Links

➡️ https://www.patreon.com/posts/stableswarmui-3-106135985

🔗 Windows Tutorial for SwarmUI Usage

➡️ https://youtu.be/HKX8_F1Er_w

🔗 Tutorial on Rapid Model Downloads for Massed Compute, RunPod, Kaggle, and Swift Hugging Face Uploads

➡️ https://youtu.be/X5WVZ0NMaTg

🔗 Join the SECourses Discord Community

➡️ https://discord.com/servers/software-engineering-courses-secourses-772774097734074388

🔗 Stable Diffusion GitHub Repository (Please Star, Fork, and Watch)

➡️ https://github.com/FurkanGozukara/Stable-Diffusion

Promotional Code for Massed Compute: SECourses
Valid for 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 on 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 on Massed Compute prior to usage
7:46 Configuring multiple GPUs on SwarmUI backend for simultaneous image generation
7:57 Monitoring GPU status using the nvitop command
8:43 Pre-installed Stable Diffusion models on Massed Compute
9:53 Model download speeds on Massed Compute
10:44 Identifying and resolving GPU backend setup errors in a 4-GPU configuration
11:42 Monitoring the status of all four active GPUs
12:22 Image generation and step speed for SD3 on RTX A6000 (Massed Compute)
12:50 CivitAI API key setup for accessing gated models
13:55 Efficiently downloading generated images from Massed Compute
15:22 Installing the latest SwarmUI on RunPod with proper template selection
16:50 Port configuration for SwarmUI connectivity post-installation
17:50 Downloading and executing the RunPod SwarmUI installer script
19:47 Resolving backend loading issues through pod restart
20:22 Relaunching SwarmUI on RunPod
21:14 Downloading and implementing Stable Diffusion 3 (SD3) on RunPod
22:01 Setting up multiple GPU backends on RunPod
23:22 Generation speed analysis on RTX 4090 (SD3 step speed)
24:04 Quick image download process from RunPod to local device
24:50 SwarmUI and Stable Diffusion 3 installation on a free Kaggle account
28:39 Modifying SwarmUI's model root folder path on Kaggle for temporary storage
29:21 Incorporating a second T4 GPU backend on Kaggle
29:32 Cancelling and restarting SwarmUI runs
31:39 Generating images with Stable Diffusion 3 on Kaggle
33:06 Addressing and resolving out-of-RAM errors on Kaggle
33:45 Disabling one backend to prevent RAM errors with dual T5 XXL text encoder usage
34:04 Analyzing Stable Diffusion 3 image generation speed on Kaggle's T4 GPU
34:35 Batch downloading of generated images from Kaggle to local device

1. Introduction to Cloud-Based AI Image Generation

In this article, we delve into the world of cloud-based AI image generation, focusing on how to utilize SwarmUI, Stable Diffusion 3, and other Stable Diffusion models on various cloud computing platforms. This comprehensive guide is designed to help users who may not have access to powerful GPUs locally but still want to harness the capabilities of advanced AI image generation tools.

1.1 Overview of Platforms Covered

The tutorial covers three main platforms for running SwarmUI and Stable Diffusion models:

1. Massed Compute: Introduced as the cheapest and most powerful cloud server provider, offering pre-installed SwarmUI for immediate use.
2. RunPod: Another cloud service provider that allows users to utilize high-end GPUs for AI image generation.
3. Kaggle: A free platform that offers GPU resources, enabling users to run SwarmUI and Stable Diffusion models without cost.

1.2 Importance of Prior Knowledge

Before diving into the cloud-based solutions, it's strongly recommended to watch a comprehensive 90-minute SwarmUI tutorial for Windows users. This tutorial, described as a "masterpiece," covers all aspects of using SwarmUI and is essential for understanding the full capabilities of the software. The current tutorial focuses primarily on installation and setup on cloud platforms, assuming the viewer has already familiarized themselves with SwarmUI's usage.

2. Massed Compute: The Powerhouse of Cloud AI

Massed Compute is highlighted as an exceptional cloud service provider for AI image generation tasks. This section of the article details the process of setting up and using SwarmUI on Massed Compute.

2.1 Registration and Deployment

To get started with Massed Compute:

1. Use the specially provided registration link, which offers benefits to both the user and the tutorial creator.
2. After registration, enter billing information and load some balance into your account.
3. Navigate to the "deploy" section of the Massed Compute dashboard.

2.2 Selecting the Right Configuration

Massed Compute offers various configurations for AI tasks. The tutorial recommends:

1. Selecting the "RTX A6000" or "RTX A6000 Alt config" option.
2. The main difference between these configurations is the amount of RAM, with the Alt config being an alternative when the standard A6000 is unavailable.
3. For the tutorial, four GPUs are used to generate four images in parallel, though it's noted that only one GPU is necessary to run SwarmUI.

2.3 Deployment Process

The step-by-step deployment process on Massed Compute is as follows:

1. From the category options, select "creator."
2. For the image, choose "SE courses."
3. Apply the special coupon code "SECourses verify" to reduce the hourly rate from $2.5 to $1.25.
4. Click "deploy" to initiate the instance creation.

2.4 Connecting to the Virtual Machine

To access the Massed Compute virtual machine:

1. Download and install the ThinLinc client appropriate for your operating system.
2. Configure the ThinLinc client:
   • In the options, uncheck all local devices except "drives."
   • Add a folder for synchronization to facilitate file uploads and downloads.
3. Use the provided login IP address and credentials to connect to the virtual machine.

2.5 Setting Up SwarmUI

Once connected to the Massed Compute virtual machine:

1. The SwarmUI application is pre-installed and ready to use.
2. Use the provided updater button to ensure SwarmUI is on the latest version.
3. The update process is automatic and will start SwarmUI upon completion.

2.6 Utilizing Multiple GPUs

To make full use of multiple GPUs on Massed Compute:

1. Navigate to the "Server" and "Backends" sections in SwarmUI.
2. Add additional ComfyUI self-starting backends, one for each GPU.
3. Ensure each backend is assigned to a different GPU by setting the appropriate GPU ID.

2.7 Downloading and Using Models

Massed Compute comes with several pre-installed models, including:

1. StableDiffusionXL base version
2. RealVisXL version 4
3. Stable Diffusion HyperRealism version 3
4. StableDiffusion3 medium model

To use these models:

1. Select the desired model from the dropdown menu in SwarmUI.
2. Adjust settings such as sampler, scheduler, and text encoders as needed.
3. Enter your prompt and generate images.

2.8 Performance and Cost-Effectiveness

The tutorial demonstrates the impressive performance of Massed Compute:

1. Generating multiple images simultaneously across four GPUs.
2. Achieving speeds of over 4.4 IT (Inference Time) per second.
3. Offering this performance at a cost-effective rate of $1.25 per hour.

2.9 New Features and CivitAI Integration

The tutorial introduces a new feature in SwarmUI:

1. The ability to use a CivitAI API key for downloading gated models.
2. Instructions on how to obtain and input your CivitAI API key are provided.

2.10 Downloading Generated Images

To retrieve the images generated on Massed Compute:

1. Navigate to the output folder within the SwarmUI directory.
2. Copy the output folder to your synchronization folder.
3. Access the synchronized files on your local machine.
4. RunPod: High-Performance Cloud GPUs

The second part of the tutorial focuses on using SwarmUI on RunPod, another cloud service provider known for its high-performance GPU offerings.

3.1 Registration and Setup

To begin using RunPod:

1. Use the provided registration link to create an account.
2. Set up billing and load credits into your account.
3. Navigate to the "Pods" section to deploy a new instance.

3.2 Selecting the Right Configuration

For optimal performance on RunPod:

1. Choose the "Community Cloud" option for temporary storage.
2. Select "extreme speed" from the filters.
3. Opt for NVME storage for faster data access.
4. Choose a configuration with 48 GB RAM for better performance.
5. Select the desired number of GPUs (the tutorial uses 3x 4090 GPUs).

3.3 Template Selection

Choosing the right template is crucial for compatibility:

1. Select "RunPod PyTorch 2.1 with CUDA 11.8" as the template.
2. This template ensures support for all necessary applications.

3.4 Deployment and Connection

After selecting the configuration:

1. Set the disk volume according to your needs (100 GB is recommended in the tutorial).
2. Set the proxy port to 7801 for accessing the SwarmUI interface.
3. Deploy the pod and wait for initialization.

3.5 Installing SwarmUI

To install SwarmUI on RunPod:

1. Connect to the JupyterLab interface provided by RunPod.
2. Upload the modified "install_linux.sh" file provided in the tutorial.
3. Execute the installation commands in the terminal.

3.6 Configuring SwarmUI

After installation:

1. Access the SwarmUI interface through the provided HTTP service port.
2. Follow the setup wizard to customize settings and select models for download.

3.7 Using Multiple GPUs

To utilize multiple GPUs on RunPod:

1. Add additional backends in the SwarmUI server configuration.
2. Assign each backend to a different GPU.

3.8 Performance and Model Usage

The tutorial demonstrates:

1. Downloading and using various Stable Diffusion models, including SDXL and Stable Diffusion 3.
2. Generating multiple images simultaneously across different GPUs.
3. Achieving high-speed image generation with RunPod's powerful GPUs.

3.9 Downloading Generated Images

To retrieve images generated on RunPod:

1. Use the JupyterLab interface to access the output folder.
2. Download the images as a zip file.
3. Kaggle: Free GPU Resources for AI Image Generation

The final section of the tutorial covers how to use SwarmUI on a free Kaggle account, providing access to GPU resources without any cost.

4.1 Setting Up Kaggle

To get started with Kaggle:

1. Create a free Kaggle account and verify your phone number.
2. Download the provided Kaggle notebook file.
3. Create a new notebook on Kaggle and import the downloaded file.

4.2 Configuring the Kaggle Environment

To set up the environment:

1. Select "GPU T4 x2" as the accelerator to use both available GPUs.
2. Ensure internet access is enabled for the notebook.

4.3 Downloading Models

The tutorial provides instructions for downloading models to Kaggle's temporary disk space:

1. Execute the provided cell to download selected models.
2. Monitor disk space usage to ensure you stay within Kaggle's limits.

4.4 Installing SwarmUI

To install SwarmUI on Kaggle:

1. Follow the step-by-step instructions provided in the notebook.
2. Execute the cells in order to set up and start SwarmUI.

4.5 Configuring SwarmUI for Kaggle

Some specific configurations are necessary for Kaggle:

1. Change the model root directory to "/kaggle/temp" in the server configuration.
2. Add an additional backend to utilize both T4 GPUs.

4.6 Using SwarmUI on Kaggle

The tutorial demonstrates:

1. Generating images with SDXL and Stable Diffusion 3 models.
2. Using both T4 GPUs for parallel image generation.
3. Handling potential RAM limitations when using complex models.

4.7 Performance on Free Resources

The article highlights the impressive capabilities of free Kaggle resources:

1. Generating SDXL images in about 20 seconds per image.
2. Running Stable Diffusion 3 with reasonable performance.
3. Utilizing 30 hours of GPU time provided by Kaggle each week.

4.8 Downloading Generated Images

To retrieve images generated on Kaggle:

1. Use the provided cell to zip all generated images.
2. Download the zip file through the Kaggle interface.
3. Additional Resources and Community

The tutorial concludes by providing additional resources and encouraging community engagement:

5.1 Discord Community

Users are invited to join a Discord server with over 7,000 members, where they can chat and ask questions about SwarmUI and AI image generation.

5.2 GitHub Repository

The SwarmUI GitHub repository is highlighted:

1. Users are encouraged to star, fork, and watch the repository.
2. The repository has gained significant popularity, with 1.9K stars at the time of the tutorial.

5.3 Patreon Exclusive Content

The tutorial mentions a Patreon exclusive post index available on GitHub, allowing users to browse and potentially subscribe to additional content.

6. Conclusion

This comprehensive guide provides users with multiple options for utilizing SwarmUI and Stable Diffusion models in cloud environments. Whether opting for the powerful and cost-effective Massed Compute, the high-performance RunPod, or the free resources available on Kaggle, users can now access advanced AI image generation capabilities without the need for local high-end hardware.

The tutorial emphasizes the importance of understanding SwarmUI's functionality through the recommended Windows tutorial before diving into cloud-based solutions. It also highlights the rapid development in this field, with new features like CivitAI integration being added to enhance the user experience.

By following this guide, users can harness the power of cloud computing to generate high-quality AI images, experiment with various Stable Diffusion models, and participate in a growing community of AI enthusiasts and creators. The combination of detailed instructions, performance insights, and resource management tips makes this tutorial an invaluable resource for anyone looking to explore the cutting edge of AI image generation technology.