🤖 AI Video Summary: To run a stable diffusion model, a GPU is recommended for faster performance. If you don’t have one, consider purchasing or using cloud-based GPU services like Run Diffusion, Think Diffusion, Runpod, or Vast. These services offer various plans and pricing, with options for affordable alternatives, and some require scripting knowledge.
A GPU can significantly reduce image generation time, making it ideal for creating stable diffusion art.
When working with a diffusion model, it is always best to have a dedicated graphics card (GPU).
However, most modern computers should be able to generate images at varying speeds.
To give you an idea of the expected generation times for your machine, this spreadsheet of user-reported benchmarks is an excellent source (here’s a local copy if the link ever becomes unavailable). You can see that a new RTX 4090 performs 40 iterations per second whereas a M1 Mac performs .57 for a 512×512 image.
What does this mean?
In about 20 steps with the Stable Diffusion 1.5 model, the RTX 4090 can make an image in about .5 seconds. Whereas a Mac will take somewhere in the neighborhood of 10 – 30 seconds.
Mid-range graphics cards, like the RTX 3060 with 12GB of RAM work well for most use cases.
Additionally, cloud-based providers like RunDiffusion.com are also a good way to still generate images without having to own a high-end GPU.
Requirements will vary depending on the models that you are using. Stable Diffusion 1.5 models are much smaller and can run on most machines whereas Stable Diffusion XL or Stable Cascade requires a higher-end graphics card with suitable VRAM (greater than 8GB) to produce an image.
One response to “Realistic System Requirements for Using Diffusion Models”
Here’s some other services that can be used to run Stable Diffusion:
Leave a Comment