Stable diffusion negative prompt

Using Negative Prompts in Stable Diffusion

Reading Time: 2 minutes

Stable Diffusion, a cutting-edge deep learning text-to-image model released in 2022, has revolutionized the generation of detailed images based on text descriptions. While its primary application is generating art conditioned on text prompts, it can also be utilized for various tasks, including inpainting, outpainting, and image-to-image translations guided by text prompts. One of the key parameters to master when using Stable Diffusion is the “Negative Prompt.”

Using Negative Prompts in Stable Diffusion

Understanding Negative Prompts in Stable Diffusion

A negative prompt is a powerful tool that allows users to specify what they don’t want to see in the generated image without requiring additional input. This parameter serves as a high-dimensional anchor, guiding the generation process away from undesired elements. Negative prompts can enhance image output by defining abstract concepts like “blurry” or “pixelated,” enabling users to create unique, detailed, and precise images.

How Negative Prompts Influence Generation

By providing negative prompts, users can exert greater control over the generation process. This means that even if Stable Diffusion produces duplicates, users can intervene by specifying negative prompts like “duplicate” or “poorly rendered face,” leading to more accurate and customized results. Remarkably, Stable Diffusion often comprehends negative prompts better than human language, resulting in improved output.

Using Negative Prompts in Stable Diffusion

To employ negative prompts in Stable Diffusion or Dream Studio, you simply add “| negative prompt>: -1.0” to the prompt. For example, “deformed, ugly: The error ‘-1.0, too many fingers: -1.0′” serves as a workaround to prevent an excessive number of fingers in the generated image.

Universal Negative Prompt List for Stable Diffusion

To assist you in leveraging negative prompts effectively, we present a comprehensive list of options to remove specific elements from the generated image:

  1. poorly drawn hands
  2. underexposed
  3. overexposed
  4. blurry
  5. malformed limbs
  6. jpeg artifacts
  7. low contrast
  8. missing legs
  9. morbid
  10. out of frame
  11. duplicate
  12. extra arms
  13. deformed
  14. worst quality
  15. extra fingers
  16. disfigured
  17. low quality
  18. extra limbs
  19. watermark
  20. error
  21. lowres
  22. ugly
  23. missing arms
  24. cropped
  25. username
  26. signature
  27. poorly drawn face
  28. long neck
  29. bad anatomy
  30. body out of frame
  31. text
  32. extra legs
  33. bad proportions
  34. dehydrated
  35. gross proportions
  36. mutation
  37. fused fingers
  38. cloned face
  39. mutated hands
  40. too many fingers
Share this article