Getting Stable Diffusion to generate photo-realistic images takes some practice and experimentation. The key is crafting a prompt that provides enough detail and context to guide the AI model, while avoiding anything that could lead to strange artifacts or distortions.
In this article, I’ll share my expertise on engineering effective prompts for realism and walk through some examples you can try yourself.
Setting the Scene
When prompting for realism, it’s important to set the scene so Stable Diffusion generates an image grounded in reality. Describe the subject, lighting, environment, camera angle and other photographic elements.
Here is an example prompt to render a realistic portrait:
A high resolution portrait photo of a young woman with long brown hair. She is outdoors on a sunny day, standing in front of a brick wall. Her hazel eyes look directly at the camera and she has a slight smile. The image is perfectly focused and professionally photographed.This provides key details about the woman, scene, lighting and photographic style. The AI model can now render something realistic.
Describing the Subject
Along with the setting, provide descriptive details about the main subject. For portraits, describe facial features, expression, age, gender, ethnicity, hair and any other visual traits.
Here is an example:
A close-up portrait of an elderly Asian man with grey hair and a mustache. He has deep wrinkles on his forehead and crow's feet by his eyes. His skin is weathered from years working outside. He gazes directly into the camera with a somber expression. Describing distinctive facial features and skin textures helps the AI generate realism.
Specifying Image Quality
It’s also important to guide the AI on the image quality, angles and composition. Specify that it should render a high resolution, professionally photographed image.
A perfectly focused, high resolution image with excellent dynamic range and color accuracy, photographed with a full frame DSLR and prime lens.Terms like “full frame DSLR” and “prime lens” signal to the AI that this should have the depth, sharpness and quality of a professional photo.
Putting It All Together
Combining all those elements together, here is an example prompt to generate a realistic portrait photo:
A high resolution, perfectly focused portrait photo of a young hispanic woman, photographed in a studio with soft continuous lighting. She has long black curly hair, dark brown eyes, a round face and light brown skin. She is wearing a white shirt and has a gentle smile, looking directly into the camera. The image has a soft bokeh background captured with a 85mm prime lens on a full frame camera.This provides details on lighting, environment, camera gear, the subject’s appearance, pose and expression. The AI has enough photographic context to render something realistic!
Useful Websites
Here are some great sites to find more examples of Stable Diffusion prompts:
Keep experimenting with prompts and studying examples shared by the creative community. With practice, you’ll be able to guide Stable Diffusion to generate stunning photo-realistic images.