sd-webui-semantic-guidance

An unofficial implementation of SEGA: Instructing Text-to-Image Models using Semantic Guidance for Automatic1111 WebUI

Semantic Guidance gives you more control over the fine details of an image by emphasizing or de-emphasizing concepts given in an additional positive / negative text prompt.

Use Cases

• Altering the style of an image without changing the composition

• Adding concepts to an image

Please see the bottom of the page for more examples.

Issues / PR's are welcome!

Settings

• Positive / Negative Prompt: Concepts to emphasize / de-emphasize, separated by commas
• Warmup Steps: How many steps to wait before applying semantic guidance
• Edit Guidance Scale: Globally scale how much influence semantic guidance has on the image
• Tail Percentage Threshold: The percentage of latents to use when calculating the semantic guidance
• Momentum Scale: Scale the influence of the added momentum term
• Momentum Beta: Higher values will make the influence of the momentum term more stable

Feature / To-do List

• SD XL support
• Support A1111 prompt attention syntax and shortcuts for attention strength
• Support restoring parameter values from infotext (Send to Txt2Img, Send to Img2Img, etc.)
• XYZ plot support
• SD 1.5 support (untested)
• Support semantic guidance prompt in regular Positive/Negative Prompt fields

Credits

• The authors of the original paper for their method (https://arxiv.org/abs/2301.12247):

   @misc{brack2023sega,
   	title={SEGA: Instructing Text-to-Image Models using Semantic Guidance}, 
   	author={Manuel Brack and Felix Friedrich and Dominik Hintersdorf and Lukas Struppek and Patrick Schramowski and Kristian Kersting},
   	year={2023},
   	eprint={2301.12247},
   	archivePrefix={arXiv},
   	primaryClass={cs.CV}
   }
  

• Official implementation of Semantic Guidance for Diffusion: (https://github.com/ml-research/semantic-image-editing)

• @udon-universe's extension templates: (https://github.com/udon-universe/stable-diffusion-webui-extension-templates)

More examples

• Enhancing the details of an image