How to Run Stable Diffusion (without owning a GPU)

Want to use Stable Diffusion, but feel sidelined by your hardware? You’re not alone.

The entry barrier for the world of AI art is real: online services can be restrictive, and high-end hardware is incredibly pricey. This guide will teach you how to run Stable Diffusion for under $1/hour, regardless of your current setup.

Stable Diffusion Minimum Requirements

To run Stable Diffusion locally, the hardware demands vary significantly based on your expected output. If you try to run it on a standard setup, you’ll likely face "Out of Memory" (OOM) errors.

There are plenty of ifs regarding the bare minimum specs for image generation with Stable Diffusion. There is broad agreement around these minimum system requirements:

<ul><li><strong>OS:</strong> Windows 10/11, macOS, or Linux.</li><li><strong>GPU:</strong> NVIDIA (Recommended). Some AMD and Intel GPUs may work with extra configuration.</li><li><strong>VRAM:</strong> 4GB (Minimum for basic generation).</li><li><strong>Storage:</strong> 20GB+ (Significant space needed for models and LoRAs).</li><li><strong>RAM:</strong> 16GB.</li></ul>

You could build a computer to match these requirements for under $1000. However, as we’ll see in the benchmarks, entry-level hardware won't get you very far.

Stable Diffusion Minimum VRAM

The Stable Diffusion minimum VRAM requirements of 4GB are strictly for basic, low-resolution generation. If you want to use high-resolution models or advanced tools like ControlNet, your system needs much more overhead.

A high-end card can churn out images in seconds, while integrated graphics might take ten minutes if the system doesn't crash first.

Usually, the main limitation for image generation is VRAM. That's because models have to be loaded on to memory, and as they grow more complex they also become larger.

Recommended VRAM for Popular Models

Here is how the most popular models stack up:

Strictly speaking Flux doesn't use latent diffusion and is not part of the Stable Diffusion brand. It was included in this table as one of the most popular models today.

Picking the Right Stable Diffusion Model

Picking the right model has a massive impact on the end result. But, how do you choose the right one? AI evolves constantly, so any manual or guide you find is probably outdated.

Luckily, there are common-sense strategies to pick the right model:

<ul><li>Find AI art references, check out what models were used.</li><li>Ask a community. Reddit is great for that.</li><li>Try it out for yourself.</li></ul>

Spin up an A100 with 80GB of VRAM for $0.78/hour and start testing models in minutes.

How to Run Stable Diffusion in the Cloud

If your current PC doesn't meet the minimum requirements, you don't need to splurge on a new build. You can run Stable Diffusion on a cloud GPU.

Cloud computing allows you to rent GPUs that would require a massive investment to own. Thunder Compute makes it even simpler by providing pre-configured templates.

Don’t wrestle with local Python environments or driver updates, and focus on creating.

Why Use Thunder Compute?

Thunder Compute provides high-performance GPU instances optimized for AI workloads.

By using our ComfyUI template, you bypass the complex installation process entirely. ComfyUI is a node-based interface that allows you to run AI generation models with granular control.

A few scenarios when a cloud GPU can become invaluable:

<ul><li><strong>You have the hardware</strong>, but need it for another project.</li><li><strong>You don’t have the hardware</strong>, but want to learn.</li><li><strong>You are on the move</strong>, but can’t take your system with you.</li></ul>

The Quick Runthrough:

Getting started is simple. Here is a brief overview:

<ul><li><strong>Select Your Instance:</strong> Log into Thunder Compute and choose a GPU.</li><li><strong>Deploy the Template:</strong> Choose the ComfyUI Template. This pre-installs necessary drivers.</li><li><strong>Launch &amp; Create:</strong> Open the provided link to your workspace and start generating immediately.</li></ul>

Our official guide on using instance templates offers a detailed walkthrough on setting up your environment.

Hardware for Every Creator

We offer a range of GPUs to fit your project. Whether you’re running a quick test or a massive batch of high-res upscaling, we have the right chip for the job:

Thunder Compute lets you modify instance specs at any time. If your project changes you can scale up or down easily, without having to build your workflows all over again.

Stop Waiting, Start Creating

You shouldn't have to be a hardware expert to be an artist. By running Stable Diffusion on a cloud GPU, you get professional-grade speed on any device.

Ready to see what the A100 can do for your art? Try Thunder Compute GPUs today and run Stable Diffusion without investing in hardware.

FAQ