Are you curious about how to make AI-generated art with just text prompts? Do you want to know expert AI artists’ best tips and tricks?
This process is known as prompt engineering, which involves finding the right balance of specificity and ambiguity in the text prompts you to use. Start with a subject matter, and then add specific keyphrase modifiers to create an image more aligned with your vision.
I was curious about how to find the perfect balance when creating text prompts for an AI art generator. I’ve collected advice (5 tips) from various AI art communities that I found helpful, and I’m excited to help you become a better AI artist.
Prompt Engineering and Prompt Modifiers
Most AI-art generator tools are text-to-image generators. Deep generative models can synthesize artworks from textual inputs (“prompts”)—a nerdy way of saying that you feed in a short descriptive text, and the AI algorithm generates an image based on that text prompt.
Prompt engineering is the practice and skill of how to phrase effective input prompts for your AI-art generator. It’s about carefully selecting and composing sentences to achieve a particular visual style in the synthesized image.
An effective textual input prompt includes a subject matter, which is what the image will be about, as well as keywords and key phrases to the prompts (referred to as “prompt modifiers,” or “style phrases”), which communicate to the AI algorithm the specifics of how you want the generated image to look.
As I mentioned, I’ll share a “recipe” for effectively phrasing input prompts. However, some artists do not share all of the prompts they used to create their AI-made artworks. So, there will still be a lot of trial-and-error and experimenting with different phrases as input prompts to determine which works best for your use case. That’s probably why “engineering” is included in the prompt engineering name.
Let’s move on to the tips and tricks from the experts.
Great Tips on How To Make Good AI-Generated Art
Tip 1: Start With A General Idea Of What You Want To Create
Before you start prompt engineering, you must have a general idea of the kind of image you would like to generate. This is known as your subject matter.
Your subject can be anything from a specific object or scene to a more general concept or feeling. Once you have decided on your subject matter, you can start brainstorming different ways to describe it. What kind of text prompt would generate an image that looks like your idea?
- “an old car in a meadow“
- “a landscape at dawn,“
- “orchestra conductor leading a chorus of sound wave audio waveforms swirling around him on the orchestral stage,
- the sun is setting over the city, casting a warm glow on the buildings and streets below.“
You can also use Google Images or Pinterest to do a reverse image search of what you have in mind and look at the tags associated with the images. This can give you a better idea of what keywords to use in your prompt.
Tip 2: Select a Styling Keyword
Whose style do you want to imitate? Do you want the output image to look like it was painted by Vincent van Gogh or generated by a contemporary artist?
This is known as a “styling keyword or style modifier.” It is the first prompt modifier added to your subject matter. It is added to produce images in a particular style.
For instance, the modifier “by Francisco Goya” will generate digital images that look like paintings by the late Spanish painter.
Other examples of this type of modifier include:
- “surrealistic painting,”
- “oil painting,”
- “oil on canvas,”
- “Cubism” or “cubist”,
- “mixed media,”
- “cabinet card,” “in the style of a cartoon,” “by Claude Lorrain,” or “in the style of Hudson River School,” to name but a few.
As shown in the preceding list, style modifiers can include information about art periods, schools, and styles, as well as art materials and media, techniques, and artists.
There is no right or wrong answer here.
It depends on what you are going for and what you think would look best. If you’re unsure, it might be helpful to look at examples of art in the style you are interested in to get a better idea of what kind of images are typically associated with that style.
Tip 3: Add A Quality Booster
Quality boosters are adjectives that increase the aesthetic qualities and the level of detail in the generated image.
For example, adding the adjective “sparkling” to a glass of water might make the water look more refreshing while adding “clear” might make it look purer.
Similarly, adding “fertile” to a “field of grass” might make the grass look greener and healthier while adding “sparse” might make it look dry and unhealthy.
Other examples of quality boosters include:
- “trending on artstation” or
- “rendered in UnrealEngine,” or
- “highly detailed,” or
- “AWESOME,” “eclectic,” “fantastic,” “beautiful,” and “epic,” to name but a few.
Example of text prompt provided in the DISCO Diffusion notebook: “A beautiful painting of a singular lighthouse, shining its light across a tumultuous sea of blood by greg rutkowski and thomas kinkade, Trending on artstation.”
By carefully selecting quality boosters, it is possible to improve the appearance of an image significantly.
Tip 4: Sometimes Repetition Helps
It is often better for AI algorithms when there are prompts with repetitions. This is because when the AI algorithm is presented with the exact prompt multiple times, it can better learn the associations being made.
This was demonstrated by Twitter user @nshepperd1, who found that the prompt “space whale. a whale in space” resulted in better results than either of the two subject terms alone.
This is likely because the repeated prompt helped the AI algorithm better to understand the concept of a “whale in space.”
Similarly, the prompt “a very very very very very beautiful landscape” will, for instance, likely produce a better image than a prompt without repetitions.
Tip 5: Use Adjectives To Describe The Mood or Emotion You Want To Convey
Mood and emotion are some of the most essential aspects of art. They are what can make an image truly striking and memorable.
Think about what kind of mood or emotion you want to convey in your output image. Do you want it to be peaceful and calming? Or playful and energetic?
Some examples of mood and emotion adjectives include:
- “serene and calming, with a feeling of peace and tranquility.”
- “full of life and energy, with a sense of joy and happiness.”
- “The picture is dark and ominous, conveying a sense of fear and dread.”
Bonus Tip: AI Is Starting to Write Prompts for AI Art Generators
It’s a bit meta, but you can use new AI writing tools to help write text prompts for AI art generators.
I love Chibi AI, and Chad published this video about Chibi Studio, which will help you in the text prompt engineering. Watch this quick intro video (7m6s@2x)