Learn how to generate images with AI by writing quality text prompts. Get tips and tricks on how to adjust your text, so you can create imagery without limits.
Have an idea? Describe it with text and AI will generate an image. Shutterstock.AI’s newest innovations create any image you can dream up. Simply describe your image using text, click a button, and in just a few seconds, a variety of image interpretations will appear before you.
Using AI to generate an image is an easy thing to do. That’s why we’ve partnered with OpenAI, the creators of DALL-E 2, to bring this technology to the masses. It allows users to create without limits in a safe, ethical, and convenient way. All you need is imagination and text.
When it comes to writing that text though, you may be presented with a few challenges. Before clicking the Generate button for the first time, you may ask yourself:
- What kind of language should I use?
- How should I adjust language to get my images just right?
- I wasn’t an English major . . . So will my writing even make sense to the AI?
The truth is, it takes some practice and refinement to write a good text prompt. All Shutterstock.AI users tinker around with text prompts to get their images just right. Here we give you tips on how to write better text when generating images with AI. Let’s dig into the writing process, tricks on how to adjust your prompts, and an example of refining an image through editing text.
Side note: If you’re looking for something super-specific, we still recommend exploring Shutterstock’s massive image library!
Now . . . let’s start writing and creating!
Tackle Your Writing Troubles
Enter text into a field. Click a button. Receive a handful of images. Seems simple enough, right? Well, it is, but getting that text just right takes a bit of effort.
Generative AI is very literal. It takes every word into account and needs detailed language to produce exactly what you’re looking for. The AI will always generate a few different takes on an image concept. Still, you’ll probably need to make adjustments on your text prompt, in order to get image results that are closer to your desired result.
See how “oil painting of tulips in a field under a blue sky with a hot air balloon flying overhead” delivers more descriptive image results than “balloon over flowers” does?
Here are a few common problems you may encounter with your text:
- Language is too vague. We’ll teach you how to be more descriptive.
- Even if you are very descriptive, details may be placed in the wrong spot in text syntax.
- There are limitless ways to customize your text. Are you accounting for all of these possibilities? Probably not on your first try, but that’s ok!
Even though there are a handful of issues to get caught up on, there are so many more ways to solve them. Let’s dig into how to write a more perfect prompt.
Tips and Tricks to Writing Generative AI Prompts
Getting the hang of writing for generative AI takes practice, sure, but here are a few tips and tricks on how to master your text prompts more quickly.
When writing your very first text prompt:
- Thoroughly think through what your desired imagery is. Are there a few different ways to describe it? If so, nail down what’s the most literal interpretation of your idea. Start there.
- Enter your idea into the AI image generator. Then click the Generate button. In about 15 seconds, you’ll see four image interpretations of that idea.
- Examine each of your four image results. They may not look exactly like the image you had in your mind . . . or they may not look like what you imagined at all! That’s ok. This just means you should adjust your text.
- Ask yourself what image elements are missing. What did you forget to type? Are major elements–like people, settings, accessories, or animals–missing? Where should those elements be within your image? Assessing an image’s missing elements will help you understand what to include in your adjusted text prompt.
How to Edit Generative AI Prompts
- Add in any missing elements. Be sure to include the parts of your image that failed to make it into your first round of images.
- Be thoughtful of text placement. Take care when adding more details to your prompt. Be sure to add descriptive text to areas where it’ll make sense. Avoid adding descriptions to areas within your prompt that may confuse the AI. For example, writing a prompt that says, “a blue man performing on a stage” is a better prompt than “a man on a stage blue.”
- Examine all angles. Did you know that you can add in descriptive language to tell the AI how to display the image? You can add in phrases like “¾ view” or “from overhead” to change up the angles of your imagery.
- Use colorful language. Let’s say you want your image to look warmer. You can include language that describes the color palette of your desire. Try adding phrases like “warm tones” or “in a warm color palette” to your next text prompt.
- Apply artistic styles. Want your images to look like a Renaissance painting? Or do you prefer paintings with a more modern art aesthetic? You can apply any artistic style to your generative AI queries. Toy around with these different looks as you continue to generate images. A new masterpiece awaits.
Tutorial: Editing Text Prompts for AI
The tips and tricks above are helpful, but what do they really look like in action? Here’s an example of how to update and adjust a text prompt, in order to get an AI-generated image that’s just right.
For this example, here’s a look at our first image and final image. They’re pretty different. So how did we get here?
The very first text prompt I wrote was, “sparkly soft serve ice cream on top of a chocolate waffle cone.” I received a handful of images of delicious ice cream cones. Still, none of them looked exactly how I wanted.
First, I asked myself, “What are these images missing?” Upon reflection, I discovered that my images were missing:
- A waffle cone that’s chocolate flavored or dipped in chocolate.
- Ice cream that’s purple. In my mind, I imagined it to be purple.
- A hand holding the ice cream cone. I don’t want it floating in space.
With these things in mind, I updated my text prompt to read, “a hand holding a chocolate-dipped waffle cone that’s topped with purple soft serve ice cream with sparkly sprinkles.” From this updated search, the results were getting closer to my desired result.
While these images were closer to what I had in mind, they weren’t perfect. I had some text refining to tackle. When examining these new images, I noticed:
- Only one cone included chocolate. I needed to be more specific with my language around the chocolate waffle cone.
- Not all cones included sprinkles. I needed to tell the AI that the dessert had to be covered in sparkly sprinkles.
- I decided that I didn’t just want any hand holding the cone. I wanted a child to be holding it, while enjoying the ice cream cone at a playground.
I then readjusted my prompt to read, “a kid holding a chocolatey waffle cone that’s topped with violet soft serve ice cream and covered in glitter with a playground in the background.” Of course, this drastically changed my image results. I was delivered imagery that included the person and setting of my choice.
From there, I continued to edit my language to refine my image results. Following the same adjustment process, I nailed down an AI-generated image that was closer to what I had initially envisioned. Thoughtfully choosing my words, and adjusting them as needed, delivered the image I had in my mind’s eye all along.
Would you like to generate your own imagery using AI? Get started with Shutterstock.AI’s image generator right now!
License this cover image via Ikon Images.