DALLE 2 is widely regarded as one of the most significant technological innovations of the 2020s, and it unquestionably sparked the current generative AI boom.
However, since its inception in 2022, other picture generators like as Stable Diffusion and Midjourney have been producing increasingly remarkable AI art.
It’s felt like DALLE has been sliding behind for the past six months or so.
But, with the release of DALLE 3, that has changed.
What exactly is DALLE 3?
DALLE 3 is the most recent release of OpenAI’s AI art generator. I
t’s a significant improvement over DALLE 2, both in terms of how you use it and the quality of what it can output.
It can once again compete with all of the other AI picture generators on the market.
The most significant change is that DALLE 3 is no longer available as a standalone app, at least for the time being.
It is instead linked with ChatGPT. This makes it easier to use, but it still has a few oddities, which we’ve learned to anticipate from any AI-powered application.
So let’s get started.
First, register for ChatGPT Plus
DALLE 3 is currently exclusively available to ChatGPT Plus customers. Sign up for a ChatGPT account and then click Upgrade to Plus at the bottom of the left sidebar to gain access to it.
Choose the $20/month ChatGPT Plus subscription option, add your payment information, and you should be all set.
So far, DALLE 3 appears to have the same limit as GPT-4: 50 requests per three hours.
In theory, this implies you could produce thousands of photos per day.
That’s a lot more than you’d receive with an AI image generator like Dream Studio (which employs Stable Diffusion) or Midjourney, though pricing isn’t consistent across the board.
And I imagine that if you start cranking through that many prompts on a regular basis, OpenAI will have something to say.
How to Begin with DALLE 3
To get to DALLE 3, go to ChatGPT and then hover over GPT-4 at the top of the screen. Select DALLE 3 Beta from the dropdown menu.
Following that, using DALLE 3 is identical to using ChatGPT.
You have the option of entering a standard text prompt or interacting with it using natural language.
Each question will provide you with four different photographs from which to chose.
You can try out:
“A watercolor interpretation of a dancing flamingo amidst a desert oasis.”
“A digital art piece illustrating a robot playing the violin in a futuristic city.”
“A surrealistic depiction of a mermaid conversing with a desert cactus under a starry sky.”
“A vintage photograph of a French baker crafting pastries in an old-world kitchen.”
DALLE 3 accomplishes something unique in that it automatically improves and iterates on your prompts.
It also employs a slightly different prompt for each of the four images produced.
For example, when I entered “A watercolor interpretation of a dancing flamingo amidst a desert oasis.”
DALLE 3 actually used the following prompts:
“Watercolor painting of a graceful flamingo dancing with its long legs and curved neck on the sandy shores of a desert oasis, surrounded by tall palm trees casting shadows.”
“Watercolor depiction of a vibrant sunset over a desert oasis, where a lone flamingo showcases its dance moves, creating ripples in the water.”
“Watercolor artwork of a dreamy desert landscape, where amidst the calm waters of an oasis, a flamingo is caught in a dynamic dance pose with its feathers fluttering.”
“Watercolor portrayal of a serene desert oasis scene, with a backdrop of sand dunes and cacti, and at its heart, a dancing flamingo reflecting in the clear waters.”
Simply click on any image to discover what DALLE 3 used as a prompt. What it tried to generate will be displayed in the Prompt box.
Hover over an image to download it, then click the download symbol in the top-left corner.
How to Use ChatGPT to Control DALLE 3
The best part of using DALLE 3 is that you can respond to natural language inquiries via ChatGPT. If there is something you don’t like or want to be highlighted with the varied outcomes, you can simply request it.
I have tried for example:
Requesting additional versions of a single photograph
Changing the perspective of each image
Changing the position of the subject in each photograph
Changing each image’s aspect ratio
Increasing or decreasing the number of people in each photograph
Adding, deleting, and changing subject details such as color and size
Adding and deleting background information
Displaying the created items on gallery walls
Unfortunately, rather of making direct adjustments, DALLE 3 now generates a new prompt based on your requirements and then generates a new set of images.
When the differences between the two photos are minimal and exactly what you want, it seems like magic.
However, DALLE 3 will occasionally throw out what you loved about a particular image.
Working with DALLE 3 to fine-tune the prompt that delivers you exactly what you want is still lot easier and more efficient than relying on trial-and-error, as you had to use DALLE 2.
It also helps that DALLE 3 keeps jazzing things up and creating more fascinating and evocative prompts for you.
How to Get the Best Out of DALLE 3
While DALLE 3 is still in beta, it is possible to achieve excellent results. DALLE 3 in particular excelled at creating sketches, paintings, and other types of artwork, rather than photorealistic photographs.
Here are some pointers to help you achieve the greatest outcomes.
Provide specific prompts
Even though DALLE 3 makes it easier to utilize simpler prompts by extrapolating a lot of things for you, if you want a specific image, include a lot of specifics in your prompt.
For instance, the screenshot below began with my prompt:
“A meticulously crafted sculpture of an Egyptian Mau as a sorcerer, commanding a flight of mythical beasts in a fierce sky battle against a horde of menacing gargoyles. The Mau dons a mystical cloak and waves a crystal staff, as it meows incantations to its squadron. The skies are turbulent, lightning crackles through the ominous clouds, the scene is nothing short of cataclysmic. Cool and ghostly hues dominate the scenario. The suspense lingers, will the feline sorcerer triumph?”
Not bad, right?
DALLE 3 comprehends numbers and positions
Although it is still possible to overload DALLE 3 with an absurd number of details in your prompt, it is far more difficult than it was with DALLE 2.
While it isn’t perfect, DALLE 3 has a far better knowledge of things like numbers and the arrangement of various items within your image.
You can, for example, ask it to generate anything in the foreground or on the left side of the image, and it will almost certainly do it.
Similarly, if you ask it for a certain quantity of something, it will almost always get it right.
Request modest changes
When you ask DALLE 3 to make adjustments depending on one of its results, it can sometimes make significant changes to the original request.
If you want it to keep things more consistent, tell it to make “subtle variations.”
While this does not prevent it from creating whole new photos, I discovered that it changes the initial prompts less.
It’s a lot to get 50 requests every three hours
I tested DALLE 3 extensively over the course of two days to compose this post, and I never reached the limit.
Take your time telling it what to do and going through each image. You’re unlikely to hit the cap without making a concerted effort.
Have fun and experiment
Seriously, the only way to truly understand what DALLE 3 is and isn’t capable of is to experiment with it.
ChatGPT handled several requests that I expected it to struggle with, but it also completely messed up what I thought were trivial improvements.