Diffusion offers a few different ways you can generate or refine an image.

Text Prompts

Text prompts are the most common way to generate images with Diffusion. With a text prompt, you can describe the type of generated image you want. To get started, click the Prompt tool () to enter a text prompt.

You can start typing in the text field or click the menu to use the following options:

  • Presets – Style presets use both regular and negative prompts to generate a certain look and feel. For more information on how they work, see Preset Prompts.
  • Prompt – Enter a text prompt that generates an image using the model viewport or generated image displayed in the Diffusion window. For more information on creating text prompts, see Prompt Engineering.
  • Additional Settings
    • Negative Prompt – Enter text that performs the opposite of what a prompt does, describing what you DON’T want Diffusion to include when generating an image.
    • Maintain Seed – Activate to force Diffusion to reduce randomness in image generation. When selected, the same prompt will always generate the same output.
    • Respect Model Geometry – A slider to control how much of the original geometry Diffusion is allowed to alter. At 0, Diffusion will only adhere to the text prompt and any geometry can be altered. At the maximum of 1, diffusion will only generate images that respect the original geometry and not alter it.
    • Prompt Influence – Control the influence of the entered text prompt on Diffusion’s output. Lower values will disregard the prompt, while higher values will put more emphasis on the prompt than the source input.
    • Add a Reference Image – This option is available only in the “No Style” Preset and leverages Stable Diffusion’s IP Adapter to generate a custom style based on an image of your choice. Clicking this option opens your device's file browser where you can select a .jpg or .png.
Note: In the latest version, Diffusion enables the Text Prompt Tool and selects the Auto-Prompt Style by default

Erase

You can use the Eraser tool () to paint over an area in your viewport. Activate the tool and paint over an area to erase it, allowing Diffusion to use contextual information to fill in that area.

You can select Brush mode, designate a diameter for your brush, and color in an area you want to erase. With Lasso mode, you can circle the area you want to erase. After designating an area with either mode, click Generate. Diffusion will generate results in the Gallery. If you leave the tool and come back your selection will persist. You can clear your mask by clicking Clear and Reset at the bottom of the dialog.

Paint

With the Paint tool () you can paint over an area and enter text prompts that will only affect that selected area.

In the options for the Paint tool you can also add a Negative Prompt or select Brush or Lasso for your selection method. Once you have selected an area and entered a prompt, click Generate. Diffusion will generate results in your Gallery. If you leave the tool and come back your selection will persist. You can clear your mask by clicking Clear and Reset at the bottom of the dialog.

Sketch

With the Sketch tool () you can sketch over an image and combine that sketch with an additional prompt, generating added elements in your output.

You can add both a regular or negative prompt as well as adjust the diameter of the lines you draw. Clicking Generate creates output using the drawn image and your prompts. If you leave the tool and come back your selection will persist. You can clear your mask by clicking Clear and Reset at the bottom of the dialog.

Was this article helpful?