When using AI tools, excellentPrompt wordIt is the basis for high-quality generated results. Whether it's a text generation tool, ChatGPT, or an AI image generation tool, the quality of the prompts is a direct result of the output.
But, while ChatGPT can give good answers even with very simple prompts, AI image generation tools rely more on specific and detailed prompts and often require multiple attempts to generate the results you want.
In this article, we'll detail how to write effective AI art prompts and share some sample prompts that will hopefully help you get started creating quickly.
Important reminder: each AI Image Generation Toolwork differently, and there are differences in the way prompts are handled. Therefore, even if you use the same prompts, the results generated by different tools may not be the same, and the results may vary each time they are generated.
How to Write AI Art Prompts
Among the many great AI image prompts I've observed, I've found that they all have something in common. An effective AI art prompt usually includes the following structure:
1. Describes the subject or content of the image.
2. Describe the art form and style.
3. Add other relevant details such as light, color, and framing.
How long should an AI art prompt be?
There is no hard and fast limit to the length of prompts. For example, Midjourney is good for 60-character prompts, while Stable Diffusion's prompts should ideally be under 380 characters in length.
Regardless of which AI image generator you use, it is important to avoid overloading the prompts with too much content, as the model may confuse the information. It is recommended to use commas to separate different elements, which not only helps the AI to understand, but also makes it easier for you to clearly remember the content of the prompts.
Which language style should be used in AI art prompts?
The choice of language style depends on the effect you wish to generate. For precise and predictable results, use concrete descriptions and vivid details, such as explicitly specifying shapes, materials, or compositions. For more randomness and creativity, use poetic or abstract wording.
You can adjust the structure, length, and language style of your prompts as you gain experience using them. But before we begin, let's dive into the details of each of these key components.
As you gain experience, you can be more flexible in adjusting the structure, length, and style of your prompts to suit different creative needs. Next, let's explore the detailed usage of each component to help you design prompts more efficiently.
1. Describe the subject or content of the image
When writing an AI art prompt, start by describing the subject or content of the image, for example, a cat, a family on the beach, or falling leaves in the fall.
In addition to the topic or content, you may want to describe some other details:
- Describe what action the subject is doing, such as looking up, playing, or running.
- Further instructions on how to do these actions, such as looking up with joy, running with fear, or playing boldly.
- Describe the atmosphere or emotion of the image, such as an early morning drizzle, mountains at sunset, or a coffee table with a cozy feel.
The more detail you provide, the more likely you are to get the results you want. But generating results also depends on the AI model you're using, as each model is trained differently and understands input cues differently.
Here's an example of an AI image generated using fuzzy cues.
"Cat on the Couch". Generated using DALL-E 3 via ChatGPT.
Here's a more detailed tip.
"Cat on couch. The cat is lying on its back on the couch, looking naughty". Generated using DALL-E 3 via ChatGPT.
2. Description of art forms and styles
If you want to achieve a specific visual effect, you also need to describe the art form and style. Here are some common art forms and styles, along with the AI tips used to generate each image.
photography
"A child playing on a sunny Pleasure Beach, laughter as they build a simple sandcastle, mimics Nikon D6 high shutter speed action shot, soft yellow lighting". Generated using Midjourney.
Photography tips are often used to generate photo-realistic effects. In such prompts, you can specify many details, such as the framing method, the lighting conditions, and even the camera brand or lens model, all of which will affect the final result produced.
sculpture
"Impressionist oil painting of a cute robot". Generated using DALL-E.
You can explore all sorts of angles here:
- Techniques such as oil or watercolor painting.
- Art movements such as Impressionism or Fauvism.
- Artist's name, e.g., Cézanne or Kahlo.
For creating unique effects, try mixing different techniques, art movements and artist styles.
artwork
"Illustrations of dinosaurs drawn by a child, with cute and cozy illustrations". Generated using Midjourney.
Illustration styles can also be very diverse:
- pencil drawing
- Charcoal Sketch
- Comic Style
- Poster Design
digital art
"Isolated convenience store in the middle of the desert at sunset, cars parked outside, low fidelity, nostalgia". Generated using Midjourney.
Movie stills
"Building on fire, old movie stills style". Generated using neural.love.
Beyond that, there are many more art forms to try, such as:
- a statue
- collages
- street art
- textile arts
- Ceramic Art
- lithograph
Mixing different styles and forms can inspire more creativity. By experimenting with a variety of combinations, you can explore endless possibilities to create unique works of art.
3. Add other relevant details to the prompt
After describing the theme and art form in detail, you can further optimize the generated results by adding more relevant details.
(art) composition
Composition refers to the position of the subject or content in the image, which directly affects the visual effect of the picture.
Depending on how the first part of the prompt is written, controlling the framing can be challenging. For example, if you're having trouble with close-ups, try stating the subject first and then describing the background. Example:
Initial cue: man standing in the middle of a forest fire, determined, close-up, realistic. midjourney doesn't handle "close-up" well.
Adjusted cue: "Man with determined attitude, forest fire background, close-up shot, realistic". Highlighted the subject and clarified the background, and the effect is closer to the desired one.
You can borrow expressions from movie directors and cinematographers, such as "wide angle lens", "point of view" or "Dutch angle", to further enhance the compositional effect.
If you need more precise control over image composition, try the AI Outpainting tool, which expands the image and then crops it to fit the needs.
illumination
"Lonely evil banana on a table, strong light and dark contrast, realistic". Generated with Midjourney.
Lighting is the soul of an image, especially in photography, and it determines the overall atmosphere of the picture.
Especially in photography, lighting is crucial and can change the look and feel of an image.
Consider adjusting your light cues using "soft light," "strong light," or even "dramatic light." You can also mention the time of day, such as "morning," "sunset," or "golden hour."
color matching
"A bustling city full of futuristic hope with a purple and green color scheme". Generated by Midjourney.
While the lighting settings already have an effect on color, by specifying a color scheme you can further control the overall tone of the picture.
For example, "blue and gold" conveys a sense of nobility and harmony, while "red and black" is more dramatic.
Detail and realism
"Distant galaxy filled with small colorful stars with a blue earth-like planet in the foreground, realistic, colorful, 8K, popular on artstation". Generated using Midjourney.
To increase the detail of an image, you can add a "4K" or "8K" cue to allow the AI to prioritize the rendering of finer image details. While these hints don't really increase the resolution, they do add realism and complexity to the image.
You can also use hints such as "Unreal Engine" to generate images similar to those rendered by the game engine. If you need more realism, make sure to include the description "realistic" or "ultra-realistic" in the hint.
Advanced Tools and Tips for AI Art Tips
After mastering the basics, the following tools and methods can help you move up to the level of a professional creator.
Reverse engineer real images to find new hints.CLIP Interrogator is an AI model powered by Hugging Face that analyzes images and generates textual hints describing their content. You can further optimize and feed into the text to generate image models based on these hints.
Website: https://huggingface.co/spaces/pharmapsychotic/CLIP-Interrogator
The image generates an image.Text-generated images are just one option. You can also provide an image as a source of inspiration and tweak it in conjunction with the text or other controls of the model. For example, you can use a tool such as Runway to upload an image and then modify the color, composition, or add new elements.
Website: https://runwayml.com/ai-magic-tools/image-to-image
AI Painting.You can upload an image to an AI paint model and it will expand the image beyond the original frame to create a larger image. This method is perfect for adjusting the composition of a picture, such as precisely controlling framing, or expanding a complete scene from a small detail. It's not limited to AI-generated images, but regular images can use this feature as well. You can try using neural.love.
Website: https://neural.love/uncrop
Adjusts the advanced settings for generating models.Some models, such as Stable Diffusion, provide a rich set of generation parameters that you can fine-tune to control image effects. For example, by increasing or decreasing the number of generation steps, you can adjust the level of detail in the image. In general, more steps will result in higher detail, but may also increase generation time.
Train your own AI image models.If you want to generate images that are more in line with your personal style or specific needs, you can use tools such as Leap AI to train your own models. This approach reduces the bias that can occur when using a generic model, and allows you to generate results that are more in line with your desired results.
Edit images with AI.You can not only generate images, but also edit existing ones. For example, select certain parts of the image and modify them with text prompts. AI features such as Canva or Photoshop have been integrated to help you do these edits easily.
Use ChatGPT to generate better tips.Yes, you can use AI to help you write prompts. When you let ChatGPT write your prompts for you, it may offer ideas you didn't expect, such as new aesthetic styles, unique ways of composing images, or novel subject pairings that can help you achieve more unexpected artistic effects.
at last
Using AI to generate images is simple, and as technology advances, these types of tools will get better and easier to use. The key is to be willing to experiment and keep exploring the possibilities.
It is possible that the images generated at the beginning are not the effect you want, but by staying patient and iterating, you will find that the quality of the images will continue to improve and eventually approach or even exceed your expectations.