With Dall-E, OpenAI helped pave the way for generative artificial intelligence that turns text into an image. Now there is much more competition, but version 3 of the service is still viable.
In various tests found on the web, comparing it with Adobe Firefly and Google ImageFX, we found that Dall-E 3 often did the best job with realistic and engaging images and almost always the best with surreal fantasies. It’s a bit tricky, but it’s very likely to give you good, usable results on the first try, especially if you’re looking for AI hallucinations that are fun instead of failures.
Dall-E has also been the best at encouraging the user to go crazy and explore what is possible. We are sure that there are designers, artists, programmers, and others who are able to realize their visions, but not everyone is capable. In this case Dall-E is the right solution for you.
Dall-E encourages a kind of exaggerated prompt engineering, in which people submit paragraphs of text, something between a vignette and a short story, the kind of prompt that is rejected as too long by some competitors. Look at this collective vision of Kansas settlers dreaming of an era of abundance after conquering nature and Native Americans. It is an image generated from a 186-word prompt. It is a form of computer-amplified creativity that is fascinating, and Dall-E is the best tool for this work I have ever tried.
Dall-E 3 is available only through the ChatGPT Plus premium service at $20 per month, which also gives you access to a more responsive version of the ChatGPT chatbot and OpenAI’s useful GPT Store with customized versions of its artificial intelligence tools. You can try the earlier Dall-E 2 for free if you want to get a taste of what is possible, but the results are not as good.
OpenAI states that it can use content submitted to Dall-E 3 to improve model performance, that it shares content with a select group of “trusted service providers,” and that it does not sell data or share content with third parties for marketing. You can also submit a privacy request to have OpenAI stop training on your data or delete your account. For more details, see OpenAI’s general privacy FAQ and main privacy policy.
Below is our review.
How to test AI image generators
To review AI image generators, the only method is a hands-on approach. The goal is to determine how good they are compared to the competition and what purposes they are best suited for. To do this, you give the AI suggestions based on real use cases, such as rendering in a particular style, combining elements in a single image, and handling longer descriptions. Finally, you evaluate the result eventually by assigning a grade to each test category.
How good are the pictures and how well do they match the suggestions?
ChatGPT is the best of the text-to-image artificial intelligence tools when it comes to producing useful, entertaining and believable results. It still makes a lot of mistakes, like a pickleball player whose racket sticks out of his head instead of his grip, but the results make you want to dig deeper, not close the browser tab. It does a better job with dynamic scenes, with contacts and interactions between different subjects, and with moods.
ChatGPT is an instrumental part of Dall-E. It magnifies your requests, adding smooth prose to give drama to the results. It also allows a conversational style of use: you can ask for an image, then an edit without having to resubmit the entire request.
ChatGPT’s language technology also enables it to process long and elaborate requests. It turns out that advanced word handling capabilities are useful for advanced image handling capabilities.
This helps Dall-E 3 outperform rivals such as Adobe’s Firefly and Google’s ImageFX when it comes to turning your prompt into what you want by correctly assembling multiple elements. For example, Dall-E 3 was the only AI image generator that managed to create a dragon flying over a castle, breathing fire and clutching a fluffy white sheep in its claws. Of course, the sheep is gently cradled by the dragon, but this perhaps depends on OpenAI’s norms against violence.
ChatGPT Plus subscribers have access to at least 10 GPT with Dall-E-based custom logo generators that have been fine-tuned for this task and are available in the GPT Store.
In many cases the details of the images are not particularly accurate. For example, in the image of the overwhelmed dog sitter, a cat, a two-headed dog, and various other issues are seen.
How engaging are the images?
Very engaging. Since E 3 he has always produced vivid, attention-grabbing images. Even when there were problems!
Dall-E 3’s maximalist linguistic approach, however, can sometimes be undesirable. When an image of a doctor and a patient surrounded by medical equipment was required, there were a dozen monitors tracking heartbeat and breathing data. One of the computers had about 100 keys on the keyboard.
People can also seem a little crazy with their emotions. A request to make a frustrated person stand behind a box of cleaning supplies produced a couple of people who looked more enraged than frustrated and one who was downright demonic.
Is it possible to improve the results?
The text interface of Dall-E 3 is conversational. Unlike Adobe’s Firefly, there are no buttons for image styles or parameters. One can get used to its colloquial style, but this long-time user of image editing software likes buttons and sliders.
You can ask for images to be widescreen, portrait, or landscape, and the artificial intelligence will do it. But when you start with a new image request, it sometimes reverts to the default square format. More than once you get a square image that you like, but you can’t ask to expand exactly that image. You can do that with Photoshop’s generative expansion feature, if you want to go that route.
How fast are images generated?
Good things come to those who wait. Dall-E 3 often takes 20 or 30 seconds to produce a single image. This delay can compromise the interactivity of ChatGPT’s style of operation, but the results are truly beautiful.
Conclusion
Dall-E 3 is an impressive tool that can provide some creative fun and does a useful job of image creation. Like all text-image generation tools, it is prone to error, but Dall-E 3 offers the best results among the rivals I tested. You’ll have to decide for yourself whether the relative quality-and the best version of the ChatGPT chatbot-is worth more than 20 euros a month.
