Image Generation is a neural network module designed to create images using a text description. It can generate either a completely new image or modify an existing one.

 

Select in the menu: AI -> Image Generation. A dialog with settings will be displayed.

At the bottom of the dialog, there is the Results panel, which contains thumbnails of the generated images. Double-clicking a thumbnail pins the result to the panel. It will not disappear the next time you run the generation.

Prompt. Enter a detailed text description of the image to be generated. You can write prompts in any language. However, for complex, detailed descriptions, English provides greater greater accuracy and consistency with the query result.

 

Mask. Image will be generated only in the selected area.

Show Mask. The check-box enables/disables the visibility of the mask, and the drop-down list allows you to change the display type.

Use Mask. The option allows you to generate within the selected area. When it's enabled, generation occurs across the entire image.

Invert Mask. The option allows you to invert the selection.

Blur. This parameter blurs the transition between the selected area (mask) and the surrounding background.

Contrast. The parameter reduces the blurriness of the selection boundaries (mask), making them clearer.

Shift Edges. Changing the parameter leads to a shift in the selection boundaries (mask) within the blurred area.

 

Generation Settings. The Z-Image Turbo model is used for generation.

Size (512-2048). The size of the larger side of the generated image. The aspect ratio matches the document on which the filter was run.

Strength (0.00-1.00). The impact of the source image content on the generated output. If the value is 1, the source image is completely ignored. To create an image from scratch set Strength = 1.

Original Image
(Squirrel)
Strength = 0.6
Prompt = "Knitted Doll"
Strength = 1.0
Prompt = "Knitted Doll"

Results (1-8). The number of results that will be generated in one calculation.

Results = 6

Steps (5-150). The number of generation steps. The more steps, the more detail and the longer the image generation time.

Steps = 5
Steps = 25

Random (0-999999). The starting number for image generation. If the check-box is enabled, you can set a specific number; if it's disabled, a random number will be selected each time.

Precision (0.1-20.0). The default value is 1.0. We highly recommend using this value for the current neural network.

Precision = 1.0
Precision = 2.0

Sampler. In the drop-down list, select the main image generation algorithm.

Different Sampler Options

Scheduler. In the drop-down list, select the auxiliary, refinement generation algorithm.

Different Scheduler Options

 

LoRA Model (LoRA - Low-Rank Adaption). Compact, fast add-on modules trained on top of the main neural network. They allow you to add specific styles, characters, or objects without retraining the entire model.

LoRa Influence (0.1-2.0). The influence of the model on the generation result

 

Server Options (optional). The ability to delegate the image generation process to another (more powerful) computer.

Name. Your name for the server.

Address. IP-address or domain name of the computer.

Port. Access port.

 

Click the Run button to start generating an image from the text description.

Click the Cancel button to stop the generation process.

Click the Save button to open the selected result in the image editor window. If there are multiple pinned results, they will be opened in a single document as separate layers. Select Save as New to open the results in a new document; if the check-boх is disabled, they will open in the original document.

Click the Exit button to close the Image Generation dialog without accepting the changes.