Guide

Beta

Studio

Guide

API

Guide

Parameters

In order to exercise even more control over how images are generated, you can adjust one of several parameters located under the "parameters" tab on the Studio page. This section will give you a brief overview of how each of these parameters can change the resulting image.

Image Dimensions

You can change the aspect ratio and size of an image using the "Image Dimensions" dropdown selectors. You can click the <> button in the center to swap width for height. Keep in mind that an image 1024 x 1024 pixels in size is four times larger than a 512 x 512 image, and thus will take longer to generate.

Quality Slider

AI image generation models work by progressively removing noise from an input image. You can adjust how much noise is removed by changing the number of quality "steps" using the quality slider. Play around with the example below to see how this model progressively 'denoises' and transforms a blurry image into the panda that we are expecting. Notice how the image seems to stabilize around 20 steps. Generating images with step counts higher than 30 will start to yield diminishing returns, along with taking longer to generate.

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

Quality

Prompt Guidance Slider

Using prompt guidance allows you to influence the output by specifying how closely the model should follow the input prompt. A higher prompt guidance value will ensure images stay closer to the provided prompt, but may result in images taking on a "deep fried" look if set too high. Conversely, a low prompt guidance will give the model more creative freedom at the expense of not following the prompt as accurately. We generally recommend leaving prompt guidance between 5 and 12. Some models prefer a lower guidance, while others do better with a higher guidance.

Prompt: "Powerful panda bear mage casting a spell"

Prompt Guidance

Seed

AI art models are deterministic, if you specify the same input you should expect to see the same output. One of the major inputs for AI art models besides the prompt is the seed. If you set the seed to a fixed number, you should be able to generate the same image multiple times given the same prompt and parameters. Setting the seed allows you to experiment with other parameters to see how they affect the final result. You can go back to using a random seed at any point by simply deleting the number in the input field. Generating more than one image at a time will always set the seed to random.

Advanced Parameters

Scheduler

Changing the Scheduler can have an effect on the final quality of the generated image, as well as the number of quality steps required before the image stabilizes. We recommend leaving the default dpm++ Scheduler for most cases as it tends to achieve good results in as little as 10-15 quality steps.