ChatGPT DALL-E 3: Complete Guide (Generate Images with Text)

In October, 2023, OpenAI launched the open beta of ChatGPT DALL-E 3 for all ChatGPT Plus and Enterprise users. The model is also available through the API since November 6th, 2023.

DALL-E 3 significantly improves upon DALL-E 2, which couldn't keep pace with alternatives like Stable Diffusion and Midjourney. This new model raises the bar in image generation capabilities.

Explore its key features and best use-cases, including its unique ability to produce readable text. After getting to know DALL-E 3, you'll find it hard to return to older, less efficient image generators.

What Is DALL-E 3?

DALL-E 3 is an advanced text-to-image generation system developed by OpenAI. It is designed to understand more nuance and detail compared to its predecessors, enabling users to translate their textual ideas into highly accurate images. 

Unlike other text-to-image systems that often ignore specific words or descriptions, DALL-E 3 adheres closely to the text prompts provided by the user.

Key differentiators include: 

1. DALL-E 3 can understand prompts starting from a barebone input. Other AI generators require intense prompt engineering to produce quality outputs, but that’s not the case with this one. DALL-E 3 automatically enhances your prompts to ensure the best output;
2. ChatGPT includes DALL-E 3 in its ecosystem. You don’t need a separate subscription, or a separate app, to access it; 
3. Contrary to the previous DALL-E 2 model, DALL-E 3 can generate images of all proportions, including 1:1 and 16:9;
4. DALL-E 3 is now capable of generating text.

How To Access DALL-E 3

DALL-E 3 likely started the rollout to ChatGPT plus users on Friday, 13th of October. It's now available to all ChatGPT plus subscribers in multi-modal mode, meaning it's included in the GPT-4 model.

On Monday, 6th of November, OpenAI released DALL-E 3 in the API as well, with an HD and non-HD version.

ChatGPT DALL-E 3

In order to access ChatGPT DALL-E 3, you need to purchase a Plus subscription. This will give you access to all add-ons: Code Interpreter, Web Search, and Image Generation. You also get access to the new custom GPTs.

Using DALL-E 3 in ChatGPT doesn’t incur in any extra cost. You can generate a large number of images for the cost of your ChatGPT plus subscription. The limit is 50 messages every 3 hours but can vary depending on the account.

API And Labs

OpenAI now offers DALL-E 3 in their API as well. For now, the generations are pricier than DALL-E 2: 

- DALL-E 3 Standard, resolution 1024×1024, costs $0.040 / image

- DALL-E 3 Standard, resolution 1024×1792 and 1792×1024, costs $0.080 / image

- DALL-E 3 HD, resolution 1024×1024, costs $0.080 / image

- DALL-E 3 HD, resolution 1024×1792 and 1792×1024, costs $0.120 / image

It's unclear which model is inside ChatGPT, but it's likely to be DALL-E 3 Standard.

Generate Images from Simple Prompts

Now that we got the technical details out of the way, it’s time to have fun and generate awesome images for your content with DALL-E 3.

Abstract Picture

The previous model was terrible at generating abstract images of landscapes or resonate with the input’s mood. In this example, DALL-E 3 generated beautiful images that could easily become desktop backgrounds.

It looks stunning, there are no missing limbs or odd sky patterns, and it evokes a sense of peace and tranquility: exactly what we asked.

Photorealistic Image

DALL-E 3 is slightly behind some competitors in generating lifelike scenarios or photorealistic images. We encountered instances of unusual eyes and unrealistic scenes where people appeared artificially inserted, but most were satisfactory.

In the example, it's evident that the image is AI-generated. However, it doesn't detract from its goal: depicting a realistic family moment

16:9 cover photo

DALL-E 3 tried to create a cover image for this article. Though somewhat eccentric, it accurately rendered the number 3, provided relevant context, and tailored the prompt to match the purpose. Unlike Midjourney, the images display a more diverse color palette.

Images with Text

That's correct: DALL-E 3 has writing capabilities. While it's not always accurate and has its share of errors, persistence can yield impressive results.

Generate a Book Cover

Making book covers with AI isn't a walk in the park right now. Basically, these models just don't have enough data to get the text right. It's often better to whip up the design with AI and add on the text with something like Canva.

That said, DALL-E 3 is showing great promise. Sure, it goofed up "Artificial Intelligence" and isn't nailing the whole alphabet yet, but the design is getting close—about 80% there.

Social Media Design

Using AI for social media posts might not be your best bet. Tools like Canva can probably get you cooler designs quicker.

But, we had to see what DALL-E 3 could do with its templates. In our test, it returned a really adorable robot with a speech bubble saying "hi" – no typos, and it looks natural in the pic. The design? Totally Insta or LinkedIn worthy, and it'd even work as a blog's open graph image. Took us about 10 tries, though.

DALL-E 3 went a bit text-crazy at first, but in the end, we got something pretty usable.

Fancy Font Styles

The model can write in all sorts of font types (and you might even attempt to create a custom font family with it).

In this example, we added the title “World of Mushrooms” to the image. As you can see, the text fits perfectly within the environment, and genuinely looks like the beginning of a story or a video.

Chat With Your Images

In their introductory video, OpenAI demonstrated the capabilities of ChatGPT by asking it to create a charming "super-duper sunflower hedgehog" for a bedtime story for a child. DALL-E 3 brought the hedgehog to life, and the conversation continued with the model, resulting in a series of images: Larry the hedgehog, Larry's home, stickers, and even a tale highlighting Larry's best attributes. The sequence concludes with a peaceful image of Larry asleep.

Although this might not directly relate to business applications, it reveals features absent in other AI models. The ability to converse using images is a distinctive feature of ChatGPT, thanks to the integrated language model. Here, users aren't navigating through a separate platform or Discord channel to create visuals. Instead, they're crafting images within the ChatGPT environment, rapidly emerging as a top choice for creative Generative AI.

Now, drawing from Larry's narrative, let's adapt it to resonate more with creative businesses.

Create Stories from Your Generations

We asked ChatGPT DALL-E 3 to generate an image of a robot jumping on bar graphs. It’s not exactly jumping on bar graphs, but close enough.

Here’s the story:

“In the vibrant city of Datascape, where numbers and analytics come to life, there's a unique annual event called the "Bar Leap Challenge." Participants, primarily AI entities, showcase their agility by navigating the ever-changing terrains of data trends. The chart isn't just a representation of numbers; it's a playground that tests adaptability and precision. Our robot mascot, Byte, a prodigious jumper from Techton Inc., is making his debut. With each leap, Byte isn't just overcoming a bar; he's decoding complex data points, demonstrating that understanding data can be both a science and an art. This moment captures Byte mid-jump, embodying the spirit of innovation and playful mastery in the world of AI.”‍

Naturally, we had to ask to see one scene from the Bar Leap Challenge.

This is not something we would publish on our website as-is, other than for this article to showcase its potential. But it’s definitely something that gets the creative flow going. It can help generate impressive, and unique, marketing campaigns that your competitors are not running.

DALL-E 3 can tell stories, understand images, and help you come up with very creative scenarios. It’s brainstorming on steroids.

Elaborate on Details

‍

DALL-E 3 in ChatGPT uses GPT-4 to generate text. You can use it to expand on any topic you want, or ask to describe something tricky in an image. 

It can go into granular details, and you can use these to generate a story. Additionally, you can use the image description to generate more accurate variations.

Generate Variations

As always, offering more context leads to improved results. While the initial image was satisfactory, this version conveys the same idea with enhanced detail and atmosphere.

Use DALL-E 3 Together With Other ChatGPT Add-Ons

Update: everything here still holds true, but you won't need to access separate interfaces anymore. All Add-Ons now work in the same GPT-4 environment.

Earlier in the article, we delved into the ChatGPT ecosystem. So, how does this benefit you with DALL-E 3?

With the introduction of plugins, ChatGPT evolved beyond a simple text generator, influencing various industries. Now, with DALL-E 3 integrated into the platform, you can leverage its capabilities alongside other features.

This could mean generating a graph with Advanced Data Analysis starting from a mockup you generate with DALL-E 3, taking a story and turning it into a 1,000 words article, or compress/convert the file format of DALL-E images.

Process Images with Advanced Data Analysis

When you generate images with generative AI, chances are they’re going to be huge. This is the case with the previous image in the article as well: over 650KB! 

That’s way too much to publish on a website. In the example, we asked Advanced Data Analysis to compress it down to 50-60 KB.

It was unable to meet the requirements, but it returned the download of the compressed image with a weight of 70KB. That’s 89.28% of the original image. No tools, no websites required.

Apply Filters

You can use Advanced Data Analysis (ex ChatGPT Code Interpeter) to apply lots of different filters to your image.

In the example, we applied greyscale to the previously generated book cover. ChatGPT also provided a downloadable link to take this offline.

‍DALL-E 3 Vs Midjourney

DALL-E 3 and Midjourney are both amazing models for image generation that will likely continue to be popular in the months ahead.

However, there are definitely things each excels at. For example, DALL-E 3 is definitely better at text. But what about faces or hands?

Let’s find it out in this comparison. Both models received the same prompt for each test.

Abstract Images

Prompt: Abstract Image of a peaceful lake. Dimensions 16:9

DALL-E 3 Output

Midjourney Output

Images with Text

Prompt: peaceful image of a lake. Add a text on top. The text is "LAKE". Ensure it blends naturally in the picture. Dimensions 16:9

DALL-E 3 Output

Midjourney Output

Images with Faces

Prompt: generate an image of a realistic couple. Make sure the result is hyper realistic and show their face in full. Dimensions 16:9

DALL-E 3 Output

Midjourney Output

Winner

Overall, DALL-E 3 provides more variety in the outputs and requires way less time when designing the prompt. Something very basic can get you a gorgeous output. It’s also much better at text, while Midjourney excels at faces.

What’s best for you depends on the feature you need. Midjourney has been around for longer and offers image blending, masks, and much more. DALL-E 3 only offers image generation for the time being, but it's the clear winner if you need to inlude texts in your designs.

Limitations & Safety

OpenAI states the safety guidelines for DALL-E 3 the first time you prompt it. There are also other limits embedded to the ChatGPT platform as a whole (not directly related to DALL-E 3).

Violent Images

DALL-E 3 won’t generate violent images. It’s a limitation for clear reasons, but might not be ideal for people that need it to generate stories.

Public Figures

DALL-E 3 will likely generate images about public figures, but it won’t engage in conversations about them. This is to avoid inappropriate usage of the platform, something Character.AI is already struggling with.

Content Ownership

OpenAI clearly states they take no ownership of the output. All images are yours to use however you please. However, please note AI can generate copyrighted material, and it’s your responsibility to avoid publishing it.

For example, Midjourney generated the Amazon logo in multiple instances. While the image is “technically” pixel-level unique, you likely don’t want to debate that in a court against Amazon.

Lack of the File Uploader

You can’t upload images into DALL-E 3. This means Larry’s story is not actually that easy to replicate. When trying to use the same persona/mascotte in other images, you’ll face very clear limitations.

In the future, the model will likely include the file uploader currently available for Advanced Data Analysis.

Lack of Image Blending

You can’t blend multiple images together, nor move things around reliably. DALL-E 3 will try to edit the output based on your requests, but it usually ends up generating something entirely new or different.

Conclusion

ChatGPT DALL-E 3 is leaps ahead of its predecessor, and can now be used in business settings and applications.

Thanks to the great work at OpenAI, prompting DALL-E 3 is a breeze and doesn’t require any modifiers. The AI will be smart enough to enhance the prompt and generate great results that match your query.

Additionally, the fact DALL-E is now in ChatGPT expands the use cases to nearly all marketing roles and most engineering roles. Anyone that promotes any kind of product will benefit from ChatGPT and the new integration.

Make sure to enable the beta, and start creating great images with DALL-E 3 and ChatGPT Plus!

If you like these tools, and want to step up your AI game for actual business use cases, make sure to take a look at Akkio: your AI assistant. Akkio can help with data visualization, no-code machine learning, and so much more. Try it out for free today - no credit card required.