How to Use DALL·E 3 to create AI images with ChatGPT

How to Use DALL·E 3 to create AI images with ChatGPT

DALL·E 2 will likely be considered one of the most important tech products of the 2020s, and it undeniably kickstarted the current generative AI craze. But since it launched in 2022, other image generators, like Stable Diffusion and Midjourney, have been creating increasingly impressive AI art. For the past six months or so, it's really felt like DALL·E was falling behind.

But that's now changed with the release of DALL·E 3.

What is DALL·E 3?

DALL·E 3 is the latest version of OpenAI's AI art generator. It's a massive upgrade over DALL·E 2, both in how you use it and the quality of what it can generate. Once again, it can go toe-to-toe with all the other AI image generators available.

Create AI images from any app you use

Automate AI image generation

Learn how

The biggest change is that DALL·E 3 is no longer a standalone app, at least not for the time being. Instead, it's integrated with ChatGPT. This makes it easier to use, but there are still a few quirks—as we've come to expect with any AI-powered tool.

So let's dig in.

First, sign up for ChatGPT Plus

At the moment, DALL·E 3 is only available to ChatGPT Plus subscribers. To get access to it, sign up for a ChatGPT account, and then click Upgrade to Plus at the bottom of the left sidebar. Select the $20/month ChatGPT Plus subscription plan, enter your payment details, and you should be good to go.

So far, DALL·E 3 seems to have the same cap as GPT-4: you can send 50 requests every three hours. In theory, this means you could create well over a thousand images a day. That's a lot more than you typically get with an AI image generator like Dream Studio (which uses Stable Diffusion) or Midjourney, although pricing across the board isn't consistent. And I suspect if you routinely start cranking through that many prompts, OpenAI might have something to say.

If you want to try DALL·E 3 out without signing up for ChatGPT Plus, you can also test it out in Bing Image Creator. You don't have the same level of control, but it uses the same underlying model. Technically, it's free to use, though the images are watermarked, and it uses a "boosts" credit system. Once you're through your allotment, it may take longer for Image Creator to render your prompts.

Getting started with DALL·E 3

To access DALL·E 3, head to ChatGPT, and hover over GPT-4 at the top of the screen. From the dropdown menu, select DALL·E 3 Beta.

After that, using DALL·E 3 is just like using ChatGPT. You can either enter a normal text prompt or interact with it using natural language. Each prompt will get you four different images to choose from.

Some things to try are:

• "A cubist painting of a large cow in a small field"

• "An oil painting of a monkey in a spacesuit on the moon"

• "A Canadian man riding a moose through a maple forest in the style of an impressionist painting"

• "A stock photo of an Irish man working remotely"

One thing that DALL·E 3 does differently is automatically improve and iterate on your prompts. It also uses a slightly different prompt for each of the four images it generates.

For example, when I entered "a cubist painting of a very large purple cow in a small field," DALL·E 3 actually used the following prompts:

• "Cubist painting showcasing a massive purple cow dominating a compact field, with fragmented geometric shapes highlighting its form."

• "Oil painting in a cubist style depicting a large purple cow, its body composed of angular shapes, set against the backdrop of a confined meadow."

• "Art piece in cubist fashion illustrating an oversized purple cow with distorted dimensions, standing in the center of a limited grassy area."

• "Cubist artwork where a voluminous purple cow, made up of disjointed planes and angles, stands out prominently in a petite field."

To see what DALL·E 3 used as a prompt for any image, just click on it. What it attempted to generate will appear in the Prompt box.

To download an image, hover over it, and click the download icon in the top-left.

How to direct DALL·E 3 with ChatGPT

The best thing about using DALL·E 3 is that you can follow up with natural language requests through ChatGPT. If there's something you don't like or something you want emphasized with the different results, you can just request it.

Some of the things I've tried are:

• Asking for more variations of one particular image

• Changing the point of view of each image

• Changing where the subject appears in each image

• Changing the aspect ratio of each image

• Changing the number of subjects in each image

• Adding, removing, and modifying subject details, like color and size

• Adding and removing background details

• Hanging the generated works on gallery walls

Unfortunately, at the moment, DALL·E 3 creates a new prompt based on your requests and then generates a new set of images, rather than making edits directly. Sometimes the changes between the two images will be small and exactly what you want—when that happens, it feels like magic. But other times, DALL·E 3 will also throw out what you liked about any given image.

Still, working with DALL·E 3 to fine-tune the prompt that gets you exactly what you want is much easier and more efficient than relying on trial-and-error, like you had to with DALL·E 2. It helps that DALL·E 3 also continues to jazz things up and make more interesting and evocative prompts for you.

How to get the best results with DALL·E 3

While DALL·E 3 is still in beta, it's possible to get some great results with it. In particular, I found that DALL·E 3 was excellent at creating drawings, paintings, and other kinds of artwork, rather than photorealistic images. Here are some things to keep in mind to help you get the best results.

Give detailed prompts

Even though DALL·E 3 makes it easier to use simpler prompts by extrapolating out a lot of things itself, if you want a particular image, add lots of details to your prompt. For example, the screenshot below started with my prompt:

Pretty good, right?

DALL·E 3 understands numbers and position

Although it's still possible to overload DALL·E 3 with a ridiculous number of details in your prompt, it's much much harder than it was with DALL·E 2. And while it's still not perfect, DALL·E 3 has a significantly better understanding of things like numbers and the position of different elements within your image.

For example, you can ask it to generate something in the foreground or on the left side of the image, and it will most likely do it. Similarly, if you ask it for a specific number of something, it will get it right more often than not.

Ask for subtle variations

If you ask DALL·E 3 to make variations based on one of its results, it can sometimes make pretty big changes to the initial prompt. If you would prefer it kept things a bit more similar, ask it to make "subtle variations." While this doesn't stop it from generating entirely new images, I found that it would change the initial prompts less.

50 requests every three hours is a lot

I tested DALL·E 3 a lot over two days to write this article, and I never hit the limit. Take the time to tell it what to do and work through each image. You're unlikely to hit the cap without really trying.

Have fun and play around

Seriously, the only way to really grasp what DALL·E 3 is—and isn't—capable of is to play around with it for yourself. ChatGPT was able to pull off some requests that I really thought it would struggle with, but it also completely messed up what I thought were simple changes.

Getting more from DALL·E

DALL·E 3 is one of the best image generators out there, but using it through ChatGPT won't fit with every workflow. DALL·E 2 is still available if you want to send requests through an API or play around with more advanced features like in-painting and out-painting.

And if you connect DALL·E to Zapier, you can automate your AI image generation to create images straight from the other apps you use most. Learn more about how to automate AI image generation, or get started with one of these workflows.