The rise of generative Artificial Intelligence (AI) art applications is transforming the art world. Creative processes that used to make up a sizable portion of the time-on-task—such as ideation, designing mockups, and editing—can now be sped up by software.
But AI art applications that are given basic instructions will produce mediocre results. You need to give good prompts to guide AI tools in the right direction, and then refine those prompts to achieve your desired result.
This guide will go into more detail about how to get that done. We’ll discuss what AI art prompts are and give some tips for effectively prompting AI. You’ll also see some example prompts to illustrate how to guide AI to produce the output you want.
How to write AI art prompts
Crafting effective AI art prompts is both an art and a science. It requires an understanding of how art generators produce results, as well as the strengths and weaknesses of how they process information.
Here’s a brief look at how to create effective prompts for AI art:
- Define your vision. Start by defining what you want in your image. For instance, if you want to create a city sunset, use “sunset over skyline” as the kernel of your prompt.
- Nail your art from style. Guide the AI generator toward the art style you want your image to be in. Going off the initial example, say you want the city skyline sunset to have a watercolor style. In this case, modify your prompt to the following: “sunset over skyline, watercolor style.”
- Add any extra details. Once you define your vision and style, it’s time to add any extra details. In the case of a cityscape, are you interested only in skylines? Or do you want skylines in front of a mountain range? Add any extra details to your image to continue refining your prompt.
- Keep things simple. Although AI art generators have gotten good at translating descriptions into images, they aren’t perfect. Too much text in your prompt can confuse an AI generator and lead to unexpected results. Focus on keeping concise prompts to get more predictable results.
- Experiment and iterate. You may not be satisfied with your first result from an AI prompt. It’s important to keep iterating on your prompt and changing small details to see how they change the AI’s output. Play with your prompt's structure, words, and length to see how each impacts the output.
Understanding these fundamental concepts will leave you better equipped to take advantage of everything generative AI art offers and help you craft prompts and unique images that help you brainstorm and meet your goals.
9 AI prompt tips
While you can learn how to generate AI art by exploring it on your own, the learning curve can be steep. There are some tips and best practices you can use to help you get started. Use these nine AI prompt tips to create better art prompts.
Explore our top tips for AI art prompts
- Understand AI art prompts
- Know the components of effective AI art prompts
- Tailor prompts for specific art styles
- Experiment with text and image prompts
- Use prompt engineering for high-quality results
- Incorporate AI tools in artistic workflows
- Leverage social media and Discord
- Go beyond DALL-E 2
- Build out a lookbook
1. Understand AI art prompts
The first step to getting the most out of your prompts is understanding what the AI apps you use are capable of. AI is trained using large datasets of images, and it learns to mimic the styles in that training data.
For instance, DALL-E and Midjourney get their training data by crawling the internet for social media posts, websites, forums, and image-hosting websites. They also use existing datasets like LAION. As they are trained, images in the training data are labeled with text, and the AI generator is asked to produce similar images from natural language prompts.
A separate program—the discriminator—decides when the AI generator has produced an image that sufficiently matches the prompt.
What this means for you is that you need to prompt AI with similar language to its training and testing data. Often, this means using the language of visual artists.
You can ask AI to use certain color palettes and to mimic artistic styels—such as realism, cubism, or expressionism. You can ask it to produce work similar to specific eras—such as the Renaissance, Romantic, or Victorian periods. And you can even ask it to mimic particular artists—such as Pablo Picasso or Ansel Adams.
If you want a realistic image of a crescent moon over a deserted town in the style of Ansel Adams, describe it using specific terms that are relevant to that style of art. If you want something more imaginative, you might ask for a futuristic cyberpunk city landscape in the style of a manga comic book.
You can also use image generators in combination with other AI tools. Many people use AI chatbots like ChatGPT as a prompt generator for image prompts. Describe what you want for the AI model and have it output a short description for your image generator.
2. Know the components of effective AI art prompts
AI art generator prompts can be written in natural language—so you’ll input plain text to your art generator. While you can use full sentences and even paragraphs, many prompts include a list of phrases and keywords separated by commas to keep things simple.
In this prompt, we will create a small creek in the sunset with fall leaves around: landscape with reflective water, sunrise in background, fall foliage.
Now, say you want a more realistic portrait with a gothic feel. You could use a prompt like this and ensure you include the gothic style keyword: gothic portrait, castle in background, man with a walking stick.
3. Tailor prompts for specific art styles
If you have a specific style you want to incorporate in your AI art, remember to describe it in your prompt. This information will help the AI generator draw from the best inspiration to provide the output you want.
For instance, you can describe specific art styles, including:
Picking the right style will help you draw out the specific emotion you want the viewer to feel when looking at your image.
You can also use artists’ names as keywords to request styles with specific artists—such as the style of Van Gogh, Picasso, and Da Vinci.
Let’s take the case when you want something simple. You can request a minimalist style for the AI: minimalist style, basic shapes, colorful splash.
Now let’s say you want to create something in the style of an artist you like. Let’s look at something similar to Picasso: style of Picasso, cubism, detailed, dog.
4. Experiment with text and image prompts
Experimentation is essential if you want to truly take advantage of everything AI image generators have to offer. You can get great results just by giving a few generations a try—but it’s hard to get exactly what you’re looking for without experimenting with AI-generated images and the different prompts you can give them.
To that end, experiment with different words to see how they impact the image. Try different descriptive text to see how it influences the output.
For example, instead of just saying “sunrise over a lake,” create a text-to-image prompt that says: “sunrise over a lake, birds flying overhead, small ripples in the water, rainy day.”
Be as expressive as you need to describe your vision, but keep the prompt concise enough for the AI to understand your request.
Let’s start by requesting the image of the sunset and lake: sunset by the lake, cloudy sky, mountains in background.
But let’s say you wanted a shoreline that wasn’t rocky. Modify the prompt with more detail: sunset by the lake, cloudy sky, mountains in background, expansive grass shoreline.
5. Use prompt engineering for high-quality results
Prompt engineering is the process of refining your text input for AI to generate new results more in line with your vision. You experiment with different descriptors and AI generator options to tailor the output to your needs.
We’ve used examples of this strategy already, such as altering the image style and features. But you can also add different AI application features to modify the output.
Take Midjourney prompts, for instance. They offer several prompt parameters to change the output:
- Aspect ratio. Modify the image’s aspect ratio.
- Chaos. Change how varied the results are.
- Quality. Change the rendering time used.
- Style. Switch between Midjourney versions.
Different AI generators use different algorithms, and were trained on different data. So if you have tried several iterations of text prompts and aren’t getting the results you hoped for, you should explore other platforms.
Let’s look at a basic Midjourney AI prompt: city block, pack of dogs, daylight.
While an image such as this might pass a cursory glance, a closer look shows that many of these dogs have only two or three strangely placed legs. Some of them lack hindquarters, or their snouts and eyes are strangely shaped. Light posts are facing all directions, and there appears to be a set of stairs in the middle of the road.
Depending on your needs, an image like this might be okay for a quick mockup for a client, or to help an artist get a sense of what you’d like the final product to be. But it shows the ways that AI art has to be looked at closely and edited, and shouldn’t be used as a final product.
So say you’ve seen the flaws in that image, and want another iteration of it, but this time without trees. You can use Midjourney’s no command as a negative prompt: city block, pack of dogs, daylight no trees.
6. Incorporate AI tools in artistic workflows
AI can play a significant role in improving the workflow of digital art. Artists may be stuck trying to develop new ideas for projects, and this creative block wastes time that could be better spent making art.
Using AI-generated art can jumpstart that process. Use AI generators to describe the artistic style you want—whether it’s oil paintings, watercolor, or another style—to have it generate a sample for you.
Artists can then streamline their workflows by using AI images as a jumping-off point for their own creative process.
Many AI generators also have more advanced features outside of text-to-image generation. Adobe Firefly, for instance, can extend images, recolor images, and automatically fill sections.
Let’s imagine an artist who wants to create a digital oil painting. They want to create a forest scene, so they can create a prompt like this to get ideas: oil painting style, lush trees, small amount of wildlife.
Likewise, you can use the watercolor keyword to get another output in a watercolor style: watercolor style, lush trees, small amount of wildlife.
7. Leverage social media and Discord
One of the great things about the AI art space is how social it is. AI enthusiasts regularly share their work and experiments online using social media.
For instance, Discord has many communities dedicated to digital art. People share their AI prompts and the art they produce. You can browse the prompts of other users to get ideas to use yourself.
The Midjourney app exclusively uses Discord to generate art. You can browse their official Discord to see users generate art in real-time.
You can also ask questions in Discord communities and other social media websites. If you’re having trouble getting results, ask experienced users what they would do and use that advice to change your AI prompts.
8. Go beyond DALL-E 2
DALL-E was one of the first art generators to start making waves on the internet. OpenAI released it in January, 2021, and subsequently released its successor, DALL-E 2, with more advanced features.
Compared to DALL-E, DALL-E 2 had a more extensive training dataset, could create more complex objects, output higher-quality images, and had improved composition. Since its release, several other AI generators have come out—of which Midjourney and Stable Diffusion are two of the most popular.
Each of these generators has strengths and weaknesses, so explore each one to its fullest and see how you can adapt them to your art style.
You can also push the limits of AI art algorithms by further training AI apps on your own work. Stable Diffusion is an open-source generator that offers this ability through a process called fine-tuning. During this process, you provide Stable Diffusion with a set of new training data in your own style—teaching it to output images more in line with your desired output.
This is useful for artists who want output more in their own style or companies that want output matching their brand image.
9. Build out a lookbook
As artists now use AI algorithms more often for creating new and innovative artwork, you can examine what’s available to create and use it as inspiration for your own work.
Traditionally, artists did this by using a lookbook. A lookbook is a collection of images showcasing an artist’s work. You can use the concept of a lookbook by creating a collection of digital images you enjoy and created yourself.
Keep track of your favorite art and the prompt techniques you used to create it. A log will help you reflect on your efforts and continue producing amazing AI art.
Show off your AI art skills on Upwork
AI has transformed digital art. It offers a way for creatives to quickly generate and test new ideas in a short time. And when you understand how to create AI art prompts effectively, you can get even better results from AI art applications.
As time goes on, AI art will continue to get better as algorithms improve. Continue testing new features the popular tools to stay ahead of the curve as an AI artist.
Have you spent time perfecting your craft and are ready to use your skills to produce AI art for clients? Head to Upwork’s AI artist freelance marketplace to find companies looking for independent AI artists.
Disclosure: Upwork is an OpenAI partner, giving OpenAI customers and other businesses direct access to trusted expert independent professionals experienced in working with OpenAI technologies.
Upwork does not control, operate, or sponsor the other tools or services discussed in this article, which are only provided as potential options. Each reader and company should take the time to adequately analyze and determine the tools or services that would best fit their specific needs and situation.
Get This Article as a PDF
For easy printing, reading, and sharing.