Generation Settings

When generating images, you can find multiple settings that provide more control over the output, such as Prompt Guidance, Quality and Details, and Strength.

Prompt Guidance

The Prompt Guidance parameter can be found on the right panel for supported models. There are three options available:

• Low: Corresponds to a Guidance scale of around 5.

• Medium: Corresponds to a Guidance scale of 7.5.

• High: Corresponds to a Guidance scale of 12.

The higher the Prompt Guidance value, the more closely the model will follow the provided prompt. However, this may reduce the model's creative aspect. Conversely, a lower Prompt Guidance value allows the model to be more creative, but the generated image may not align as closely with the original prompt.

Quality and Details

The Quality and Details parameter is similar to the number of inference steps for the diffusion model. It is typically set to a value of 30, as this generates good-quality images within a reasonable time frame. You can decrease or increase this value.

• Higher values will take longer to generate the image but will produce more detailed results.

• After a certain range (around 30-50 steps), there may not be much noticeable difference in the generated images.

• Lower values will generate the image faster, but it may have some defects.

Images for different inference steps (article)

Strength

The Strength parameter can be considered a dial that controls how much you want to change the generated image. If you're using an image-to-image model:

• Higher Strength values will result in more significant changes, like turning a cat into a tiger.

• Lower Strength values will make smaller, more subtle changes, like adding stripes to a cat without drastically altering its appearance.

The Strength parameter is like telling the model how bold or subtle you want the changes to be.

Inference Type

Next to the generation button, you'll find three options: Normal, Fast, and Hyper. These options represent the GPU type used for the image generation:

• Normal: Runs the container on an A10G GPU, which is a cost-effective option but slower.

• Fast: Runs the container on an A100 GPU, which is faster but more expensive.

• Hyper: Runs the container on an H100 GPU, which is the fastest but also the most expensive option.

Choose the option that best fits your requirements, balancing speed and cost.