Stable Diffusion is a powerful AI image generation model that can create highly realistic images from text prompts. However, it sometimes generates unwanted artifacts or flaws in images. Negative prompts allow you to specify elements you want the AI to avoid, giving you more control over the output.
When generating portraits and images with faces, the eyes are a critical aspect that is easy to get wrong. Using targeted negative prompts for eyes can help Stable Diffusion generate more natural, realistic, and appealing eyes in images.
Here are some useful negative prompt examples specifically focused on improving eyes in Stable Diffusion images:
Remove Unnatural Eye Colors or Patterns
- weird eyes
- unnatural eye color
- eyes with symbols
- patterned eyes
- eyes with shapes
Prevent Crossed or Misaligned Eyes
- crossed eyes
- eyes pointing different directions
- misaligned eyes
- eyes not aligned
- eyes looking away
Fix Glowing or Unnatural Shine in Eyes
- glowing eyes
- eyes emitting light
- shiny eyes
- glossy eyes
- eyes with lens flares
Remove Obstructions Covering Eyes
- obstructed eyes
- covered eyes
- eyes blocked
- eyes covered by hair
- eyes covered by hands
Prevent Dead, Soulless Eyes
- dead eyes
- creepy eyes
- soulless eyes
- lifeless eyes
- haunted eyes
Remove Uncanny/Unrealistic Eye Shapes or Placements
- distorted eyes
- deformed eyes
- wrong shaped eyes
- eyes too far apart
- eyes too close together
Fix Issues With Eyelids or Eyelashes
- extra eyelids
- mutated eyelids
- eyelids covering entire eye
- eyelids missing
- long eyelashes
- dense eyelashes
Best Practices for Using Negative Prompts
When adding negative prompts, there are some best practices to follow:
- Be as specific as possible – “weird eyes” is better than just “eyes”. This gives SD more signal.
- Prioritize the most important fixes – Don’t overwhelm with too many negatives. Focus on 1-3 key issues.
- Check if positives help first – Adding specifics like “gentle eyes” in prompts can help before negatives.
- Try general artifact prompts first – Broad prompts like “artifacts” can cover issues. Then get more specific.
- Use Weighted Negative Prompts if needed – Prefix the negative with a weight (e.g. “-5: shiny eyes”) to strengthen effect.
- Test prompts extensively – Evaluating many outputs is key to fine-tuning negatives.
Example Eye Negative Prompts
Here are some full example eye negative prompts for common use cases:
Realistic Female Portrait
Young woman with long blonde hair, portrait photo
No artifacts, beautiful eyes, very detailed
--
weird eyes, crossed eyes, glowing eyes, obstructed eyes, uncanny eyes
Anime Character Portrait
Anime girl with pink hair, blue eyes, cute smile
No artifacts, vibrant eyes, highly detailed
--
dead eyes, misaligned eyes, extra eyelids, dense eyelashes
Fantasy Portrait
An elf with long pointed ears, golden eyes, in a forest
No artifacts, radiant eyes full of life, highly detailed
--
patterned eyes, deformed eyes, eyes too far apart, eyes emitting light
Useful Resources
Here are some useful sites to find more Stable Diffusion negative prompt examples for eyes:
Conclusion
Using targeted negative prompts for eyes can help you generate more appealing, natural, and realistic eyes in Stable Diffusion portraits. Start by focusing on the 1 or 2 biggest eye issues, be specific with negatives, and test outputs extensively. Refer to the prompt examples and resources outlined here when crafting negatives. With practice, you’ll be able to reliably create stunning AI-generated portraits.