Prompts are an essential tool in guiding AI to generate images that match your desired outcome.
Positive prompts tell the AI what to include, while negative prompts specify what to exclude from the generated image.
They are used in both text-to-image and image-to-image processes.
Positive Prompts
Positive prompts are the starting point of your image generation process. They can be a single word or a short phrase used to describe the key elements or features you want in the final image. These prompts help guide the AI toward a specific result and enhance the likelihood of attaining your desired visual outcome.
At the highest level, you are reconstructing an image via a diffusion model. So basic photography terms like composition, lighting, depth, color, and texture can all be positive prompts to effectively communicate your vision to the diffusion model.
In general, you want to keep the following in mind when writing a prompt:
- Specificity: Be as specific as possible. Instead of simply saying “a dog,” describe the dog’s breed, size, color, and any unique features. For example, “a small brown pug with a pink bow.”
- Combinations: You can include multiple elements in your prompt to create diverse and interesting images. For instance, “a medieval castle on a stormy night with lightning striking in the distance.”
- Styles: Artistic styles are a terrific way to get a general look and feel for your generated image. You can prompt the AI with specific styles such as “impressionism,” “cubism,” or “realism” to achieve a desired aesthetic.
- Limitations: Prompt adherence is a big challenge and often depends on the AI’s training data and its understanding of your prompts. Start with a simple prompt, then gradually refine it to get closer to your desired result. Stable Diffusion 1.5 models have poor prompt adherence compared to SDXL or Stable Cascade.
Negative Prompts
Negative prompts are used to refine and enhance your results by excluding certain elements.
Here are some tips for crafting effective negative prompts:
- Exclusion Keywords: Think of negative prompts as a list of things you don’t want in your image. If you generated a picture and there are clouds in the sky and you didn’t want clouds, then adding the term “clouds” as a negative prompt will make them disappear.
- Addressing Issues: Negative prompts are excellent for addressing specific issues in your generated images. If you’re seeing too much noise, fuzziness, or unwanted elements, try using negative prompts like “noise,” “blurry,” “ugly,” or “disfigured limbs.”
Combining Positive and Negative Prompts
Advanced prompting methods including weights, scripts, and more will be covered in more advanced lessons.