Prompt: origami, vibrant nature and technology fusion, in the style of Magritte
CREATIVE GENERATIVE AI
How to write AI art prompts that get results.
The best prompts for AI aren't always obvious. These tips get you to your destination in record time.
What made jaws drop in 1903? The Wright Brothers’ first-ever flight of a powered plane — which lasted a mere 12 seconds. The 2023 equivalent? Creating art with a text prompt for the first time.
Like the Wright brothers, you’ll probably be inspired to keep refining until your AI art really takes off. Think of the following tips as the wind beneath your wings.
Take your AI art prompt from biplane to spaceship.
Creating AI art requires two things: a technology called generative artificial intelligence (AI) and a person who tells the technology what image to make. The most common way for people to instruct AI art generators is with written descriptions known as text prompts.
While prompts can be short — leaving most of the choices to the AI — it’s more fun to fine-tune and describe what you want in detail. Adobe Firefly has a text to image generator that lets you write prompts using nothing more than everyday language.
Let’s say you tell Firefly to make an image of a “full moon” and you get something like this.
Prompt: full moon
Begin to create a scenario with the text prompt “full moon in the sky with a rocket headed toward the moon.” To evoke a particular medium, just tell the AI, for example, “full moon in the sky with a rocket headed toward the moon, cartoon art” or “full moon in the sky with a rocket headed toward the moon, in the style of a Golden Age Dutch Master oil painting.”
Prompts: (left) full moon in the sky with a rocket headed toward the moon; (middle) full moon in the sky with a rocket headed toward the moon, cartoon art; (right) full moon in the sky with a rocket headed toward the moon, in the style of a Golden Age Dutch Master oil painting
Whether you’re going for a straightforward scene, abstract art, character design, or something else, kick your AI art prompts into hyperdrive with descriptors like the following:
- Art movements (Cubism, Pop Art, Impressionism)
- Colors (either specific hues or concepts like vibrant or muted)
- Composition (close up, wide angle, shot from above)
- Emotions (happy, sad, angry)
- Lighting (back lighting, studio lighting, golden hour)
- Materials (photo, yarn, origami)
- Styles (hyper realistic, steampunk, collage)
If you use Firefly to generate AI art, you’ll find many descriptors already listed in the interface, which makes it even easier to try new AI art prompt ideas. You can also specify the image’s aspect ratio in the Firefly interface. Options include square (1:1), landscape (4:3), portrait (3:4), and widescreen (16:9).
Try alternate flight paths
When you’re in an experimental mood, load up your prompts with multiple, even competing, descriptors. Weird combinations could be your ticket to a new way of seeing things.
Nature and technology fusion
Surreal dreamscape
Prompt: flamingo cartoon character in a surreal dreamscape, neon colors, photographed from above
Abstract geometric pattern
Prompt: abstract geometric patterns, vintage surfers, hot colors
Portrait in monochrome
Prompt: happy portraits in monochrome, made out of yarn, graphic, bioluminescent
Pro tips for the best AI art prompts (for now).
AI art generation is still in its infancy, and the ways we interact with the technology will likely evolve far beyond text prompts. In the meantime, here are a few tricks that help in some situations.
Group complex concepts with quotation marks.
When the AI struggles to isolate parts of the prompt, try adding quotes around the elements you want it to treat as one group; for example, “bright green swirling vector shapes in the sky” “red Martian rover on land”, photorealistic.
Prompts: (left) bright green swirling vector shapes in the sky red Martian rover on land, photo realistic; (right) “bright green swirling vector shapes in the sky” “red Martian rover on land”, photorealistic
Indicate elements to avoid.
Add [avoid = purple] to the end of a Firefly text prompt to direct the AI away from the concept in the brackets.
Prompts: (left) 3 pastel-colored spaceships landing on moon; (right) 3 pastel-colored spaceships landing on moon [avoid=purple]
Specify a style’s strength.
Increase or decrease the influence of a Firefly style on an image by adding [stylize = 90,10] at the end of your text prompt. The numbers inside the brackets must equal 100, and the higher the second number, the more closely Firefly will stick to the style in your prompt.
Prompts: (left) red rocket in a blue sky with clouds, layered paper [stylize = 10,90], (right) red rocket in a blue sky with clouds, layered paper [stylize = 90,10]
Weight the prompt’s importance.
Give the AI freedom to pay less attention to a prompt by adding [guidance = 25] to the end of the prompt. The numbers can range from 0 to 25. The lower the number, the more the AI ignores the prompt.
Prompts: (left) steampunk alien dressed like a medieval queen sits on a throne [guidance = 1]; (right) Steampunk alien dressed like a medieval queen sits on a throne [guidance = 25]
Roll the dice.
For results that are surprising but stray from your prompt, add [raw-style = True].
Prompts: (left) double exposure of the Milky Way galaxy inside the helmet of an astronaut; (right) double exposure of the Milky Way galaxy inside the helmet of an astronaut [raw-style = True]
Apply an AI art prompt to only part of an image.
No matter how skilled you are at writing text prompts, you’ll inevitably generate an image that’s perfect — if only you could change just one thing without affecting anything else. That’s when it’s time for Generative Fill, which allows you to add or remove content from your images using simple text prompts.
In Firefly, you use a few simple controls to select part of the image, and then describe with a new text prompt what the AI should do inside the selection. You can also use Generative Fill on images that aren’t AI art, either in Firefly or in the Adobe Photoshop desktop and web apps.
Prompt: (left) International Space Station above the Earth wide-angle composition photo. Generative Fill prompt: (right) astronaut in bright purple space suit waves at camera, strong shadows.
Words can be works of art, too.
Firefly can even apply AI art to letters. Again, you direct the AI with text prompts, and you choose the letters, the font, and how far beyond a letter’s outline the AI art spreads.
The default spreads are Tight, Medium, and Loose. You can add [outline-strength = 10] to the end of your prompt to vary the spread by degrees other than the three defaults. The lower the number in brackets, the greater the spread.
Prompts: (left) double exposure of the Milky Way galaxy inside the helmet of an astronaut; (right) double exposure of the Milky Way galaxy inside the helmet of an astronaut [raw-style = True]
You’re in the pilot’s seat.
By pushing text prompts beyond the basics, you can generate AI art as unique as your imaginings. Start simply and then layer on styles, moods, and other evocative descriptors. Try Generative Fill to finesse areas within your AI art. And don’t be afraid to get weird with your text prompts — unexpected destinations are sometimes the most memorable. When it comes to AI art prompt ideas, the sky’s the limit.
Share this article
Dream Bigger with Adobe Firefly.
Imagine, experiment, and create with generative AI in the Firefly web app. New to Creative Cloud, now available for commercial use.