This guide will cover the basics of working with text prompts and how you can effectively utilize them to render the images you envisioned.
Takeaways
- Prompts for AI image generators can be both text and image-based, and in some cases, a mix of both
- The general structure of text prompts for AI image generators is similar across the board, as is the word count and order of descriptors.
- Different elements of a text prompt have unique effects on the rendering and add complexity to the image.
What Are Prompts?
Prompts for AI image generation are commands, either textual or visual, that dictate what image the AI platform creates. Typically, you will find two different image generation prompt types: text and image.
Text Prompts
To tell the AI what kind of image to render, you should first describe that image. This is done with a text prompt, which is a list of words, sometimes even a grammatically correct sentence, describing what the AI should render.
The highlighted textbox is an example text prompt used to
Image Prompt
In some cases when text is not enough, you can upload images to an AI image generator and use it as a reference to create other images. Depending on the type of AI platform you’re using and the type of results you will need, using image prompts can be even more effective than text.
For example, DALL.E’s Outpainting AI can take an existing image and render its continuation. The image above demonstrates what the AI believes is the continuation of the famous Johannes Vermeer painting of Girl with a Pearl Earring.
The original image of the painting is within the square and everything that surrounds it was rendered by DALL.E.
A Mix of Both
While most of the AI image generators you find on the market today will be text-based with some options for image prompts, there are ways to utilize a mix of both prompts to render desired images.
Writing an Image Generation Prompt
One very important detail you must keep in mind is to understand what a particular AI image generator specializes in and how to use it effectively. Some AI image generators, such as the Dall-E series are very good at rendering images that imitate photorealism.
On the other hand, there’s Midjourney, which is well known for being very effective in digital art or illustration formats.
Our own Let’s Enhance Image Generator is great at simulating the visual aesthetic of product images, on top of also rendering photo-realistic images, illustrations, and 3D models.
Word Count: Use At Least 3-7 Words
While most AI image generators will have a character limit on the text prompts, there is no word limit. This means that you have the freedom to structure your text prompt any way you want.
However, the consensus, for now, seems to be that you need 3-7 words to write a successful text prompt for the AI to render.
Here are the rendered images of the text prompt in the example above. The “wide shot” prompt significantly impacted the images in the left column.
You don’t have to strictly abide by this rule and are allowed to use less than 3 or more than 7 words. But as you’ll see in the examples below, it’s not so much the number of words that matter, but what words you use.
Subject: Who and What
Any AI image render requires some sort of subject. This should be the first thing you write in the text prompt (in most cases). It is possible to just leave a single subject in the text prompt and have the AI generate something a little more general.
However, in most cases, you’re going to need a more descriptive text prompt to generate an image that comes as close to what you were looking to render.
Subjects are typically nouns, so the best place to start with your text prompt regardless of which AI you’re utilizing is to start with one.
Description: What They Are Doing, Where, and How
To add a little more nuance to your rendered image and help the AI narrow down what images to use as references, you need to utilize descriptions. These are your adjectives, and much like how you would use them in any ordinary sentence, they describe the subject.
The trick with using adjectives in AI image generation is that you can combine several of these to add more depth and complexity to the image.
Aside from adjectives that describe the subject, another element of the description is details that answer questions such as:
- What is happening?
- What is the subject doing?
- How is the subject doing this?
- What’s happening around the subject?
That last question is pretty important, as backgrounds play a key role in effective descriptions. Here’s an example of two similar text prompts yielding quite different results. The text prompt for the image on the left was Raccoon reading.
The text prompt for the right image, however, was slightly more detailed; hence why the image came out clearer and more complex: Raccoon reading a book in a library photo close shot.
As you can see by the results, we answered several questions, and the AI was able to narrow down what we were looking for and rendered something as close to what we were looking for as possible. The AI was able to generate an image with greater details and fewer anomalies.
Here’s another great example of how AI image generators work better with more detailed image prompts.
This generated image of a finch demonstrates how AI can render a generic but very accurate image even without additional descriptors.
However, the rendering on the right demonstrates how AI image generators, in this case, Stable Diffusion, can work with more elements to render a significantly more complex visual.
Aesthetic and Style: How It Looks
To complete the text prompt, there are a few more text prompts you must add that will put the finishing touches on your rendered image. Notice how, at the end of the text prompt above, we had words like “photo” and “close up”.
The last words in your prompt will be the ones that dictate the overall aesthetic and style of your image.
Instead of “photo”, you could have had “oil painting”, “fresco”, or “3D rendering”. To dictate the style of your image, you can also use art movements such as impressionist, gothic, steampunk, etc.
Also, if you’re looking to replicate the works of a famous artist, you can even put one or more famous names in the text prompt too. The AI will combine influences from various famous artists to create something that blends elements from all of them.
Finally, you can also dictate the framing of the image as well. Prompts such as “close up” or “medium shot”, much like how you would do in photography, can help the AI understand what angle you want the image to be presented from and from what distance.
Here’s an example of two images generated by Lexica of a handheld computer device. The prompt on the left, though simple, was able to generate a vintage device that even replicates the old-school LCD display.
However, the rendering to the right, thanks to prompts like “vaporwave aesthetic” and “product photography”, has a more defined visual aesthetic and style, as you can tell by the neon colors.
Experience Is the Best Teacher
No amount of tips and tricks are going to teach you how to write effective AI prompts than experience. The best part about AI image generators is that they’re very accessible. All you need is an internet connection, and you can choose from the many free platforms available.
So get out there and experiment with AI image prompts to see what new and fascinating things you can discover in the rising field of image generation.