DALL-E is the brainchild of OpenAI, designed to generate images more efficiently than other AI-art models. Considering the hype surrounding it and that the public now has access to it, getting to know its pros and cons before investing time and potential money is a good idea.
Here’s what you can expect from DALL-E as you create AI images for your projects. There are lots of good and bad points to keep in mind.
The Pros of Using Dall-E for AI Images
DALL-E truly is a fast and powerful tool for generating art without having to hire an expert. Let’s break down the key advantages of using DALL-E to create AI images from text descriptions.
1. DALL-E Creates Images With Diffusion
Generative adversarial networks (GANs) were the best way for AI programs to learn and recreate images, but diffusion models have turned out to be more effective and impressive in their results.
This method basically creates a pattern of dots or noise that gradually develops into the desired image and can deliver AI images of higher quality. This doesn’t mean, however, that you always get the results you want, a matter discussed below.
When it comes to DALL-E’s technology, the right prompt and subject can deliver beautiful works of art, whether you need them to look like paintings or photos.
2. DALL-E Is Generous With Its Credits
A major advantage to joining the DALL-E family is that you get a lot more credits to use than other platforms.
Your first month comes with 50 free credits to spend on any of DALL-E’s image generation tools, while every subsequent month gives you 15 credits.
Note that free credits don’t roll over to the next month. Any left will expire, so you’re better off using them all while you can. But purchased credits are cumulative and quite the bargain. For example, $15 gets you 115 credits, more than enough to keep you going if you run out of your free supply.
3. DALL-E Responds Well to Prompts
Thanks to the advanced algorithms working in the background, you’re less likely to get results that are irrelevant to the prompt.
The images you get may not always be perfect, but they’ll reflect your description well enough. And the more detail you add to the prompt, the better DALL-E will understand what you want.
If you specify that your image should be abstract, realistic, or fantasy art, the AI will deliver to the best of its capabilities.
4. AI Image Creation Is Fast
Despite the complexity of its system, DALL-E works quickly. It doesn’t take more than a few minutes to get the images you asked for, no matter how detailed your prompt.
Whether the results are spot-on is a different issue, but at least you’ll be able to rely on the platform’s great processing power.
5. DALL-E Offers Lots of Image Variations
Each image request generates four variations. You can then select a single creation and ask for alternatives to that. DALL-E also lets you upload images and get their variations. Before you know it, you’ll have a wall of AI-made images, but there are additional tools that allow even more tweaking.
One way you can edit a generated or uploaded image is to erase parts of it you don’t like and request variations, which will then only alter the sections you deleted. Not only that, but you can upload an image and add a 1024 x 1024 generation frame around it—the only size available at the time of writing.
Your variation request will then fill the empty space within the frame, allowing you to create composite images and several versions of them.
A downside is that you can’t delete individual images in your collection. Your only option is to scrap all your existing images and start a new collection.
6. You Can Edit Images on DALL-E
DALL-E’s editor is in beta and its features are few, but they’re constantly improving and expanding to let you properly work on your AI images instead of having to create new ones until you find the best fit.
As already mentioned, the editor offers an eraser, image uploader, and generation frame, as well as tools for selecting and panning your image.
You can also download your creation, which is strongly recommended since DALL-E doesn’t save your editing projects yet. To keep working on the image, you just upload it.
The Cons of Using Dall-E for AI Images
DALL-E is great at generating stunning AI images, but it’s still in its infancy and has disadvantages that might put you off it for the time being.
1. Signing Up Can Take a While
Gaining access to DALL-E involves filling in a form with your details and waiting for OpenAI to review it and send an invitation.
It can take several weeks to hear back, so if you don’t have the time to spare, it’s not the best platform for you. You can always forget DALL-E and try Midjourney’s AI image generator.
2. The Artwork Quality Is Low for Some Prompts
If you ask for images that contain landscapes, human faces, or popular art styles, DALL-E is unmatched in terms of quality and potential for developing a masterpiece, regardless of whether or not you should consider AI-generated images real art.
But when it comes to unconventional requests that demand complex descriptions, like strange made-up creatures, DALL-E seems to struggle, and the images produced aren’t as good.
Since AI art generators learn from existing images, text, and other online materials, if your description doesn’t already correspond to enough relevant data for the algorithm to draw from, the outcome suffers.
In the case of DALL-E, it seems to work best with mainstream prompts. If the images you need are very unusual, you may need to refine your descriptions and spend extra time editing them.
Alternatively, try a different platform. Fortunately, there are many more free AI text-to-image art generators to explore that are constantly upgrading their services to rival DALL-E.
3. You Have to Specify What Style You Want
Other AI image generation services like StarryAI and NightCafe display a selection of styles to choose from, but DALL-E has no such thing.
If you want your images to reflect a particular type of art, like surrealism, cyberpunk, or Da Vinci’s style, you have to include it in your prompt.
In other words, you may have to do some research into artists and movements if you want an image with a specific visual effect. Otherwise, you could upload existing images with the style you want and point DALL-E in the right direction.
4. DALL-E Can Create Harmful Images
In its GitHub preview of DALL-E, OpenAI itself mentions problems related to harmful image generation caused by the masses of stereotypical or explicit data the algorithm gathers from the internet.
It’s not always the case now as OpenAI works on countering the issues, but when you request images of, say, a parent, nurse, or CEO, you’ll still notice patterns of gender or ethnic bias.
Besides stereotypes, DALL-E’s image creation sometimes uses explicit images that match your prompt, resulting in images that might be more sexual or gruesome than you wanted.
To avoid such outcomes, you could use more detail when describing the people or scenes you want DALL-E to visualize.
This small solution, however, doesn’t excuse the big problem of deep learning models like DALL-E getting ideas from disturbing things people share online. AI-generated images are changing art’s future, but they can also reflect humanity’s shortcomings.
5. DALL-E Is Still in Development
There are many advantages to using OpenAI’s art generator, from impressive images to plenty of free credits to play with, but DALL-E is still a work in progress.
Even though you can get lots of cool images out of it now, you may want to wait for its services to improve. With time, the editor should become a joy to use and any bad images may be fixed or less prominent.
Get to Know How Useful AI-Generated Images Can Be
Once you learn the ins and outs of DALL-E or any platform like it, think about how you can use the images you create. Some projects need more careful phrasing and editing than others, but AI models can level the field of art and its uses.
For example, you and your trusty AI artist could make your own illustrations, concept art, and even teaching materials without wasting time or money. That said, it’s also worth understanding the risks and limitations of AI image generation.