Guide to Stable Diffusion Steps Parameter

What do sampling steps mean?

First, let's briefly introduce the steps parameter in Stable Diffusion and diffusion models in general. Generally speaking, using more steps can improve the image quality, but there's no need to always max them out. The optimal number depends on the specific results you're aiming for.

Diffusion models are iterative processes—a repeated cycle that starts with random noise generated from text input. Some noise is removed with each step, resulting in a higher-quality image over time. The repetition stops when the desired number of steps is completed.

Around 25 sampling steps are usually enough to achieve high-quality images. Using more may produce a slightly different picture, but not necessarily better quality. In addition, the iterative nature of the process makes generation slow; the more steps you'll use, the more time it will take to generate an image. In most cases, it's not worth the additional wait time.

Sampling steps: 1 to 100 comparison

Below, you can see single steps (1-100) used to generate a picture of the 'puppy in space'. Drag the slider to move between steps, starting with noise at step 1.

Step: 1

Note how the magic happens around steps 4-7, and the dog emerges from the blob. Then, it reaches high quality around 20-25 steps into the generation. Steps above 25 do not create a significant difference in quality; the dog's form repeatedly changes without producing more details.

Here's another example showcasing 'a bear astronauts' at different steps.

Steps: 15

Steps: 25

Steps: 45

Steps: 75

Notice that with higher steps, the main form of a 'bear astronaut' stays the same. However, in this case, some minor details are improving with more steps, such as the hair quality, colors, and space suit details.

Of course, it's all subjective, and we leave the final decision on how many steps to use to you—the creator. We hope our guide helped you understand the steps parameter in the Stable Diffusion generation process, and you'll use what you've learned to create something unique.

Frequently Asked Questions

What are the sampling steps in Stable Diffusion?

Sampling steps control how many times the model refines an image. It starts with rough noise and gradually turns it into something clear.

How many sampling steps should I use?

25 steps are usually just right for a good balance between quality and speed.

Will more steps make my image better?

Not necessarily. After 25 steps, any extra improvement is usually very small and might not be worth the extra time.