Prompt for adjusting skin color in Stable Diffusion images

Generating images of people with accurate and diverse skin tones has been a challenge for AI image generators. However, with careful prompt engineering, Stable Diffusion can produce high-quality images with customized skin colors. This guide will provide tips and examples for adjusting skin tone in Stable Diffusion.

Why Skin Tone Matters

Representing diversity in skin tone is important for generating inclusive and non-biased AI systems. Failure to accurately depict different skin tones can perpetuate harm by excluding people of color and reinforcing stereotypes. By thoughtfully engineering prompts with descriptive skin tone modifiers, we can create AI systems that reflect the beautiful diversity of humankind.

Engineering Prompts for Diverse Skin Tones

Here are some best practices for prompting Stable Diffusion to generate images with accurate skin tones:

Use Specific Skin Tone Descriptors

Rather than vague terms like “light”, “medium”, or “dark”, use specific color descriptors:

  • Alabaster, porcelain, ivory
  • Beige, sand, tan
  • Honey, caramel, golden, olive
  • Almond, pecan, bronze
  • Chestnut, mahogany, espresso

Include Ethnicity for Context

Pair skin tone descriptors with ethnicity for additional context:

  • Light-skinned East Asian woman with alabaster skin
  • Dark-skinned Latino man with bronze skin

Try Different Models

Models like CompVis and Anything V3 may produce better skin tone variation than Stable Diffusion 1.5.

Example Prompts

Light Skin Tones

“Smiling professional headshot of a light-skinned Indian woman with sand-colored skin, long black hair, wearing a red business suit, detailed oil painting”

“Waist-up portrait of a light-skinned Latino man with beige skin and freckles, curly brown hair, denim shirt, soft lighting, depth of field”

Dark Skin Tones

“Black teenage girl with dark chestnut skin, coily hair in braids, gap tooth smile, soft lighting, 35mm photo”

“Highly detailed digital painting of a dark-skinned Pacific Islander man with bronze skin, tribal neck tattoo, looking determinedly into the distance “

Conclusion

With thoughtfully engineered prompts utilizing specific skin tone descriptors and model selection, Stable Diffusion can generate images showcasing the beautiful diversity of skin tones across humanity. This allows for the creation of more inclusive, ethical, and unbiased AI systems.

Useful Websites: