Sampling steps refer to the number of steps the AI will take to create an image. The number you need will depend on the size and complexity of the image.
In diffusion models, a series of repeated cycles are used to generate an image from text input. During each cycle, some noise is removed, resulting in a higher-quality image over time. The number of cycles, or "steps," used to generate the image can be adjusted to control the quality and speed of the process. Generally, using more steps can lead to slightly higher-quality images, but the difference may not be significant beyond a certain point. Using a large number of steps can also slow down the generation process.
As a default, start with a range of 30-60 steps, increasing the number if the quality is low. Consider the trade-off between quality and speed when deciding how many steps to use. The number of steps can affect the appearance of the generated image. We encourage users to experiment and decide for themselves how many steps to use.
A common misconception is that more sampling steps will always produce better results. However, users have found that using too many sampling steps for simple sketch images or small images can actually add unnecessary detail that leads to poor results.
For more complex images, such as 3D and photorealistic ones, higher sampling steps may be necessary to maintain quality.
As a general rule, it is ideal to have fewer steps for 2D illustrative styles and more steps for complicated compositions and photorealistic images.