What are prompts?
Generative AI prompts are written or visual cues, words or sentences, that tell AI Art Generators what kind of image you want them to create.
In order to produce graphics with a specific style, the text prompts must be made in a particular format. This is usually accomplished by adding prompt modifiers, or keywords and key phrases added to prompts. There is a burgeoning online ecosystem of resources, such as lists of awesome AI art prompts and guides to help beginners get started. However, it takes a lot of trial and error and experimentation to master prompt engineering.
There is no one right technique to write text prompts for generative AI platforms. Different AI generators will produce varying results, but the text prompt anatomy is usually the same in general. Textual prompts should be at least 3 to 7 words lengthy, according to most prompt developers, if you want to render something that is more specific and less abstract.
In order for AI to create precise and complex results, you need to use more descriptive prompts. The more descriptive your text prompt, the easier it is for the AI to understand what you’re looking for. This can be anything from how does the subject look like, what is the subject doing, how it is doing this, what is happening around the subject. Focus on multiple adjectives that describe the object or the scene. Specific and clear language that focuses on the subject and the style will give best results. Be creative and experiment with different combinations of prompts to develop and evolve your idea.
Keep in mind that prompt modifiers are weighted–words at the beginning of a sentence carry more weight than words at the end. Here are some basics about creating prompts that will help you on your exciting AI art journey:
The concept of identifying subjects is pretty straightforward. Use your imagination to describe the scene, object, or character you want in your art; the subject is the most basic building block of any prompt and it must be a noun. Although a prompt can be written without a subject, it will be difficult to control the image generation process and you might get undesirable results.
To get the outcomes you want, it's useful to include clear and detailed style modifiers. You can add style modifiers to generate images with specific genres, such as oil on canvas paintings, cartoons, anime, cubism, photography, etc. You can also mention specific artists such as Francisco Goya, van Gogh, Pablo Picasso, or Yayoi Kusama. Style modifiers can include information about the time period, material, medium, or technique.
Adjectives can boost aesthetic qualities and the level of detail of your project. Examples of the type of adjectives you can use to add mood to your art are fantastic, epic, elegant, melancholic, etc.
Action and Scene
Actions indicate what the subject is doing, while scenes describe where it is taking place.
Repetition can reinforce associations formed by generative systems. For instance, using the two phrases in a prompt—a garden mouse and a mouse in a garden—will likely generate better results than if only one of the phrases was used by itself. Different ways of describing objects using synonyms will make the text-to-image process more accurate by activating certain areas in the neural network’s latent space.
Negative prompts tell AI to remove certain subjects and styles from the results. For example, VQGAN-CLIP tends to generate red heart-shaped images when love is included in the prompt. Adding a “heart:-1” will prevent this from happening.