Dall-e Prompting Tips & Tricks

January 17, 2023

Tory Barber

illustration of a skateboarding cyberpunk cat character

There is of course no perfect way to engineer a prompt with DALL-E 2. The image generator largely resembles the creative process itself. While design has rules and general guidelines we are recommended to follow, these rules can (and should) be broken for the most wonderfully creative results.

Not to mention, it can be easy to forget, DALL-E 2 is just in its infancy. It only recently finished its beta testing.

The more you accept the random nature of the technology, the happier you will be with the results.

With all that aside, I hope you enjoy these DALL-E tips & tricks! I’ve certainly found them useful for my own creations!

1. Give Your Prompt Structure

There’s no pefect structure to give your prompt, but I find I have the most success if I follow something like this:

{adjective}{emotion} [Subject], [Style], {color}

Dall-e prompt structure infographic

At the minimum, there are two key components in the above structure that your final prompt should certainly have:

  • A subject;
  • and a style

The subject is the core idea of your image; it is the focus of your artwork. Your subject could be more than just a person/thing. In my example above, my subject was “cat character in city streets”. Including a location as part of my subject helps fill in the background.

The style of your image is the second more important part of your prompt. This is where you can get REALLY creative. You can combine subjects with styles never seen before like a portrait of a guinea pig painted in the style of a Renaissance oil painting. Time periods (i.e. the 80’s, or medieval) and in the style of [artist name], tend to work well with DALL-E.

You can add adjective, emotion, or color prompts around your subject and style to help narrow down the variations between your results, but without a subject or style, DALL-E becomes the world’s cheesiest Stock Photography tool.


See how when I remove the style from my prompt, it defaults to a realistic photograph? And without a subject, DALL-E doesn’t know what to create! But when I have both subject and style prominent, DALL-E can create something conclusive.

No Style

realistic photo of a cat generated by dalle

No Subject

abstract digital art generated by dalle

Subject & Style

Cat character sitting in the city streets, digital art

The last thing to keep in mind when you are prompt crafting is to try and front-load the most important information.

DALL-E tends to add importance to the words at the beginning of the prompt. So if the style of the image is most important to you, add it before the subject. If you’re finding DALLL-E’s struggling to understand what the subject of your prompt is, try reducing adjectives/emotions so the subject is closer to the front.

2. Keep It Simple Stupid

As an artist, I know that inspiration can strike at any moment, leading us to create elaborate and detailed images. However, when it comes to using DALL-E 2 it’s important to remember that simplicity is key.

Unlike Midjourney and other popular AI Art tools that encourage their users to enter sprawling paragraphs in their image prompts, DALL-E users have found more success in being direct, and brief. This reduces the chances of the AI getting confused.

This means avoiding overly complicated descriptions or concepts, and instead focusing on a single, clear idea. By doing so, you’ll be more likely able to generate high-quality images that are true to your vision, without the AI getting bogged down in unnecessary details.


Complicated Prompt

illustration of a skateboarding cyberpunk cat character

Prompt: graphic novel illustration of a cool skateboarding cat character skating in a cyberpunk halfpipe, in a cyberpunk city, cosmic, anime illustration, blue and purple, cosmic, synthwave, animal characters

Simple Prompt

illustration of a skateboarding cyberpunk cat character

Prompt: An illustration of a skateboarding cat character in a cyberpunk city, blue and purple

See how the AI got confused with the more complicated prompt in the example above?

You can see DALL-E tried to create each of the items in the more complicated prompt, but it had too many different requests fighting for its focus that it didn’t really do a good job of creating any of them.

As long as DALL-E can generate the subject of your photo, you can always add the other details through DALL-E’s Editing/Outpainting feature afterward.

One way to achieve this simplicity is by breaking down your idea into its most basic components. For example, instead of describing a detailed landscape with multiple elements, focus on a single object or feature, such as a tree or a river. This will allow Dalle 2 to generate a clear, focused image that captures the essence of your idea.

Avoid using overly technical terms or jargon, and instead, use language that is easy to understand and visualize. This will help to ensure that Dalle 2 accurately interprets your prompts and creates the images you want.

By breaking down your ideas into their most basic components, and using clear, descriptive language, you’ll be able to create powerful and effective image prompts that allow Dalle 2 to do its best work.

3. Build Your Subject FIRST

In keeping with the theme of simplicity, I’d highly recommend you build your subject FIRST. Meaning, you should start your image generation by striping your prompt of styling, color, and adjective keywords so that only your core idea is left.

Once you know DALL-E 2 can create the subject you have in mind, you can start layering in additional details to your prompt.

You can also use DALL-E’s outpainting feature to adjust the background around your now-perfect subject.

Don’t Assume the AI Knows Anything

DALL-E’s knowledge is only as expansive as its training data. While DALL-E’s training data covers a staggering amount of topics, its knowledge isn’t endless.

Building your subject first, allows you to ensure DALL-E know’s what your subject even is!

There is nothing more frustrating than wasting a bunch of credits on an idea that isn’t even possible.


While creating the featured image for this post, I wasn’t impressed with the skateboards DALL-E was creating. True to my own tips, I simplified my prompt until I was happy with the subject (the cat), and then tried to refine the skateboard with DALL-E’s Outpainting.

However, when I simplified my outpainting prompt to “Skateboard” DALL-E didn’t generated an image with a skateboard in it.

This likely means DALL-E’s training data on skateboards is limited. If the skateboard was the subject of my photo, I would have wanted to know early on in the process.


dalle outpainting skateboard selection


Results from DALL-E skateboard outpainting fail

Test Out the Descriptor before adding the style

Similar to testing whether DALL-E’s been trained on your subject, sometimes it’s helpful to test one of your descriptor to see if DALL-E understands it, or if they AI has some built in biases associated with the word.


See how when we give Dall-E the prompt “Thinking” all of the resulting subjects have their hands on their face?

stock photos of people thinking

This is of course, is not the only way a person looks when they think. DALL-E has clearly learned a bias from its training data.

Therefore it’s reasonable to assume that if you were to use this descriptor in your prompt, your subject will likely use a similar pose.

Dalle results showing descriptor bias

If you would like your subject to look different than the bias, you now know you’ll have to use a different descriptive word.

4. Reinforce Your Goal

If you are finding DALL-E is struggling to develop your subject (and you’ve already tested that the AI’s been trained on it), you can trying using a technique called “reinforcing your goal”.

A slightly less generous name for this technique is keyword stuffing.

It is possible that DALL-E is struggling to identify the subject of your prompt, and that repeating the subject in different ways throughout your prompt will help indicate to DALL-E where you’d like it to focus.



With Keyword Stuffing

futuristic cat character sitting on a skateboard in city streets

skater cat character sitting in cyberpunk city streets, digital art, cat character holding a skateboard, illustrated cat character, skater cat, blue and purple


childish drawing of a cat on a skateboard

a graphic novel illustration of a skateboarding cat character in a cyberpunk city, blue and purple

5. Try Different Combinations

If you ever find yourself close to your goal, but DALL-E just isn’t quite getting it, try experimenting with different combinations of words and phrases. Dall-e is able to understand and interpret complex combinations of words and concepts, so I always try out different combinations and see what kind of results I get.

Don’t be afraid to get creative with your prompts! DALL-E is capable of generating a wide range of images, from realistic to surreal. So I like to let my imagination run wild and see what kind of unique and interesting images DALL-E can create.

The Final Tip

Most importantly, keep an open mind and have fun!

Cheesy way to end this post, I know, but some of the greatest designs come from a place of play. Why wouldn’t that be the same case with DALL-E?

I hope these tips and tricks help you get the most out of using DALL-E. It’s such a fun and powerful tool, and I can’t wait to see what kind of amazing creations you come up with!

Post your creations with the #digitalunicorn or tag me using any of the social networks outlined below for a chance to get featured!

Tory Barber Headshot Thumbnail

About Me

Hi I’m Tory!I’m a Graphic Designer, Web Developer, and AI OBSESSED artist from good ol’ Canada, eh!I record all the things I learn as I dive deeper into the Generative AI rabbit hole so I can teach people like you!

Follow Me

Prompt Inspo In Your Inbox!

Latest Articles

Latest Posts

Dall-e 3 Announcement Breakdown

Dall-e 3 Announcement Breakdown

OpenAI just announced that they will be releasing Dall-e 3 in the fall. Learn when you can get access and what Dall-e 3 improves on the latest model.