Guide

Beta

Studio

Guide

API

Guide

Quality

Quality steps are an essential part of the AI image generation process. This guide will help you understand how to utilize the Quality steps setting to achieve the best image quality.

The Quality slider under Parameters

Quality Slider

AI image generation models work by progressively removing noise from an input image. The quality slider allows you to adjust how many "steps" of noise removal you'd like the model to perform. Generally, higher step counts result in cleaner, more detailed images. However, this also increases the generation time.

Real-World Example

Prompt: "A panda bear holding a lightsaber on Tatooine"

Quality

Play around with the example above to see how the model progressively 'denoises' and transforms a blurry image into a clear one. This can help you understand how many steps are ideal for your specific needs. Generally, you will find that the image starts to stabilize around 20 steps. Generating images with steps counts higher than 30 will start to yield diminishing returns.

Trade-offs

While higher quality steps may yield better results, they also consume more computational resources and take longer to generate. Therefore, it's essential to find a balance that suits your requirements.

Best Practices

For quick drafts or previews, consider using lower step counts. Once you are satisfied with the image's overall composition and elements, you can then generate a final version with higher quality steps. Generally leaving the Quality setting at 20 is good for most cases.