In brief
- ComfyUI is a node-based, open-source interface for Stable Diffusion-based image and video generation.
- SwarmUI is an interface that uses ComfyUI as its backend, offering easier setup through ready-made presets.
- GPUs like the RTX Pro 6000, H100 and A100 can be rented hourly on RunPod and Massed Compute; parallel generation with multiple GPUs is possible.
- Wan 2.2's TI2V-5B model runs on a single RTX 4090 (24GB VRAM recommended), while the 14-billion-parameter A14B models require multiple GPUs or the cloud.
For those who want to try open-source AI models like FLUX Krea and Wan 2.2 for high-resolution image or video generation but don't have a graphics card with enough VRAM, cloud GPU services offer a practical solution. This guide explains step by step how to set up and run the ComfyUI and SwarmUI interfaces on the RunPod and Massed Compute platforms, based on openly licensed (Apache-2.0) tutorials published by SECourses instructor Furkan Gözükara.
What you'll learn
By the end of this guide, you'll know how to connect to a cloud GPU machine, install ComfyUI and SwarmUI, download the required AI models, run parallel generation with multiple GPUs, and stop billing once you're done.
Requirements
A RunPod or Massed Compute account with balance (hourly billing; for example, prices of around $1.47/hour for an RTX Pro 6000 96GB and around $1.80/hour for a 2x A100 configuration have been observed; current prices may vary on the platforms).
A ComfyUI installation package (a frequently updated version; version 57 is cited as an example in the sources) and a SwarmUI installation/model downloader package (version 89 in the example).
If using Massed Compute for remote desktop access, the ThinLinc client; on RunPod, the browser-based Jupyter Lab interface is sufficient.
Enough VRAM depending on the model you want to generate with: Wan 2.2's TI2V-5B model can run on a single RTX 4090 (24GB VRAM recommended), while the 14-billion-parameter T2V-A14B and I2V-A14B models require multiple GPUs or cloud resources.
Step-by-step setup
Create an account and add balance. After registering on RunPod or Massed Compute, set a balance under the billing section.
Choose a GPU. Massed Compute offers options like the RTX Pro 6000 Blackwell or L40S; if you want to use multiple GPUs at once, you can select a machine with dual GPUs.
Start the machine and connect. On Massed Compute, access is via the ThinLinc client; on RunPod, you connect directly through the browser-based interface. On Massed Compute, enabling local folder sharing (clipboard and drive sync) makes file transfer easier.
Upload the installation files. Transfer the ComfyUI and SwarmUI installation packages to the remote machine and extract them from the archive.
Install ComfyUI. Run the command in the installation instructions file in the terminal to install ComfyUI.
Update or install SwarmUI. SwarmUI usually comes pre-installed on Massed Compute images; running the update script is enough. On RunPod, it's installed from scratch.
Download the models. Use SwarmUI's model downloader tool to download models like FLUX, FLUX Krea, or Wan 2.2, along with the required files (such as VAE and text encoder weights).
Import presets. Ready-made preset packages let you get started quickly without manually adjusting parameters.
Configure multi-GPU backends. In SwarmUI, you can define a separate backend for each GPU to run parallel generation; acceleration options like Sage Attention can also be enabled at this stage.
Start generation and download the files. After starting image or video generation, you can transfer the outputs to your computer via the shared folder (Massed Compute) or workspace (RunPod).
Shut down the machine. On Massed Compute, delete the machine; on RunPod, terminate the pod instead of just stopping it, to avoid unnecessary billing.
Which model suits which hardware
Wan 2.2 is offered in several model variants. The TI2V-5B model can generate text-to-video and image-to-video, and runs on a single consumer GPU (RTX 4090, 24GB VRAM recommended). The 14-billion active-parameter T2V-A14B and I2V-A14B models use a Mixture-of-Experts architecture and require multiple GPUs (up to 8) or a cloud service for best performance. FLUX Krea Dev has a 12-billion-parameter rectified flow transformer architecture and aims to produce more photorealistic, less "AI-looking" results compared to the standard FLUX Dev.
Common mistakes
Forgetting to terminate the machine instead of just stopping it after you're done, which can lead to unexpected charges.
Trying to run model files directly from the shared folder instead of first copying them to the local disk (downloads), which can cause synchronization issues.
Trying to run a large model on a single GPU without checking its VRAM requirements, leading to memory errors.
Next steps
Once setup is complete, you can experiment with different FLUX Krea and Wan 2.2 presets to test which parameter combination gives the best results for your own projects.
Source and license
This guide is adapted into Turkish—translated here into English—from the community articles "How to Install and Use ComfyUI and SwarmUI on Massed Compute and RunPod," "Ultimate ComfyUI & SwarmUI on RunPod Tutorial," and "Wan 2.2 & FLUX Krea Full Tutorial," published by Furkan Gözükara on Hugging Face under the Apache-2.0 license.
Sources
3 sourcesSource texts are not republished; short quotes are marked, everything else is our own summary and commentary.
For users without a powerful local GPU who are interested in visualization and video production, renting cloud GPUs can be a practical way to try models like FLUX Krea or Wan 2.2 without investing in hardware; this approach can also be an option for small offices and students working on visual presentations and concept studies. The hourly billing model can be budget-friendly for short-term trials or one-off projects, but costs can add up quickly with regular use.
Since the setup process involves technical steps such as terminal commands, file synchronization, and GPU/backend configuration, there's somewhat of a learning curve for complete beginners. Still, combined with free, open-source software (ComfyUI, SwarmUI), it offers a low-cost testing environment for users in Turkey.
Frequently asked questions
What are ComfyUI and SwarmUI?
ComfyUI is a node-based, open-source interface for Stable Diffusion-based image and video generation; SwarmUI is a layer that uses ComfyUI as its backend and simplifies setup with ready-made presets.
How much VRAM is needed to run Wan 2.2?
The TI2V-5B model can run on a single RTX 4090 (24GB VRAM recommended), while the 14-billion-parameter A14B models require multiple GPUs or cloud resources.
What's the difference between RunPod and Massed Compute?
Both platforms offer hourly-billed GPU rentals; Massed Compute uses the ThinLinc client for remote desktop access, while RunPod provides access through a browser-based Jupyter Lab interface.
