Image Gen Settings
Last updated
Last updated
Visit Image Generation Page:
At the right side is the Settings! These are very crucial in generating the image that you’re looking for.
Here is each one in further detail:
Sampling Method: It refers to the technique used to generate images by gradually refining noise into clear, detailed visuals based on the model’s understanding. Different sampling methods control the way this refinement process unfolds, affecting the speed and quality of the final image
Steps: Steps affect generation time; recommended range: 25-50.
Scale: How 'literally' the text input should be treated. Higher would be more faithful to the text while lower would have more artistic liberty
Seed: Seed value determines how image generation starts. The same seed will generate the same image.
♻️ — use the same seed number to get the same character.
🎲 — use a random seed for different character.
If no image is selected, ♻️ button will use the seed of the latest image.
Canvas Size: The size of the generated image. Use Enhance button to increase the size of the final image.
CLIP Skip: CLIP skip only works in models that use CLIP text embedding and/or are based on models that use CLIP. Your prompt is digitized and then fed through the layers of the model and the output is used in the image-generation. At the slider value of 1, it goes through all 12 layers of the model. CLIP skip setting allows you to stop the process early and use the output from a different layer. For example, the slider value of 2 will use the output of the next to last layer (layer 11). The higher the value, the less layers of neural network have worked on the prompt
VAE: For txt2img, VAE (variational autoencoder) is used to create a resulting image after the sampling is finished. For img2img, it is used to process user's input image before the sampling, and to create an image after sampling
ENSD: Set seed offset for consistent outputs and image comparisons, while keeping the main seed random
