Image Generation Guide

Image generation is a new technology and may produce unpredicatble results. The following guide will help you get the most out of the image generation functionality.

Prompting

Examples

Checkout our example gallery for inspiration. Click "Use Prompt" to open the prompt in the image generation tool.

Include

This section describes what should be in the image. You can use both "tag-based" and "natural" prompts.

Here are some "tag-based" examples:

1girl, extreme beauty, long blonde wavy hair, blue eyes, white dress, in a forest
1boy, handsome, short black hair, brown eyes, black suit, in a city

Here are some "natural" examples:

Adorable cat-like creature with big eyes and perfect fangs / sharp teeth in forest in autumn during full moon.
Giant creature's skull with lush vegetation growing around it with an impressive waterfall cascading down. The skull is seamlessly integrated into the landscape. Ethereal and mystical atmosphere.

Multiple subjects in one image:

1girl, blonde hair and green eyes, 1boy, dark brown hair and blue eyes, having a coffee in a coffee shop in france, sitting outside, smiling

Exclude

This section describes what should not be in the image. Typically used with the "tag-based" approach to avoid underisable elements in the generated pictures.

For example, if the generated image features a human with bad anatomy, you might want to add ugly, deformed, bad hands into the exclude section. Similarly, if you want to avoid NSFW content, you might want to add nsfw, nude.

Prompt Strength

This defines how much the model adheres to the prompt. Also known as "CFG" or "CFG scale".

Lower values: Gives the model more freedom. Tends to also lower color saturation and contrast.
Higher values: Makes the model adhere more closely to the prompt. Tends to also increase color saturation and contrast.
Very high values: May result in generation artifacts.

The best value depends very much on the "include" and "exclude" prompts as well as other parameters such as "fidelity" and selected styles / effects.

Fidelity

This defines how much time the model has to refine the image. Also known as "steps".

Lower values: Faster results, but may be less detailed.
Higher values: Slower results, but should be more detailed.

Styles & Effects

These modules allow you to apply various styles and effects to the generated image. They can be used to enhance the image or to give it a specific look. Some styles / effects may introduce generation biases, for example some of the anime styles tend to bias generation towards adding humans to the image.