Just over a month ago, I received the long-awaited invitation to use DALL-E 2, an OpenAI tool that allows you to create arts in various styles from phrases, using algorithms and artificial intelligence.
The DALL-E 2 is very powerful, and in addition to creating original images, it also allows you to upload real images to create variations of them. In addition, it is also possible to insert new characters, animals, items or objects in your photos, always with very surprising results.
With the possibility of generating images using others in the center, it is possible to create the infinite zoom effect, and in this post I will show you some examples and also teach you how to create an image on top of another, the initial step to create your animations.
In this post, I’m going to show you what you can do with the DALL-E 2, but for those who want to go (much) further, a great tip with lots of information on how to create better images is the DALL-E 2 Prompt Book, a free digital book created by Guy Parsons, available on the DALL-Ery GALL-Ery website.
A brief history of the DALL-E
In January 2021, its first version was released, simply called DALL-E. The name is a mix between the eternal master of surrealism, Salvador Dali, with the title character of the already classic Pixar film WALL-E.
In July of this year, OpenAI released the beta version of DALL-E 2, when we talked about the app for the first time here on the site. The beta version is available to everyone, just join a waiting list that is available on the website, through which I received my invitation.
How does DALL-E 2 work?
In short, the artificial intelligence of DALL-E 2 has been trained with millions of real images and also works of art by different artists, painters or sculptors, as well as different types of materials.
When the user types a sentence, or uploads an image, and clicks the button to generate the image, the DALL-E 2 encoder maps the text (or photo) to identify what it is. A model then maps these items or terms with images that represent the semantic information. And finally, an image decoder generates the visual representations of this information.
How to use DALL-E 2 and what are the prompts
Upon being invited and signing up, you receive 50 credits that can be used to generate a prompt. But, what is a prompt? It’s usually a sentence, but it can also be another image, as long as it doesn’t show a human face, which is not allowed by the app’s terms.
You can use commas to enter picture styles and new details, as I did in the example below.
Each prompt generates four variations (eg the four robots in the screenshot above), and spends one of the available credits. If you liked any of the variations, or maybe all of them, you can generate new alternatives in the same style.
In the image above, I asked DALL-E 2 to create an astronaut running from the Sun on the planet Mercury. I liked the four options, but if I had to choose just one, it would certainly be the first one, as I found its movement interesting.
So, I made some variations of this first image, to show you what the result of this feature looks like, and I must say that I liked all of them more than the original. Often you’ll have to spend several tries (and credits) until you find exactly what you’re looking for.
Another possibility is to create variations of your images, always with very curious effects. In the photo above we see three alternate versions of my girlfriend’s dogs, their multiverse versions if you will.
In addition, it is also possible to play with the photos, inserting completely new things, which is one of the most fun things about the app. It is quite easy to make a mask to erase specific areas of the image to be occupied by what will be generated by the DALL-E 2.
In the image above, I deleted a part of my original photo of a nighttime São Paulo and asked DALL-E 2 to insert Godzilla destroying buildings there. I liked both versions so much that I couldn’t choose just one. It’s amazing to see how faithful it is to the original lighting in the photo.
It is worth mentioning that this type of effect with an original photo is impossible in its competitor Midjourney, which even allows you to upload images, but only to generate new prompts, and not to interact with them in such a creative way.
Some examples of prompts done in DALL-E 2
In the image above, my prompt was a 3D rendering of a capybara in a spacesuit on top of a tall building on an alien planet, and I asked for the digital art style to be applied.
I also had fun creating images in the style of different artists. In the image above, for example, prompts of Doctor Who and his TARDIS in Monet and Picasso styles (line drawing), plus the Sgt. Pepper’s by the Beatles painted in the style of Van Gogh.
The image above shows a steampunk alien, with two versions generated from the same prompt. I liked using the term steampunk, so I also created the eye below, in yet another attempt to re-brand our site.
In the image below, I asked DALL-E 2 to create a spacecraft approaching a high-tech satellite in the year 3000, with Saturn’s aurora appearing in the background.
I also approved the result, which was very high-tech with these neon lights, so I generated 3 more variations of this theme, which you can see below.
Using images as a central part of a new image
One of the most interesting features of the DALL-E 2 is being able to create an image using another as a central point, as in the three images above, one created from the other. In the image below, I made one more version, still using the central image as a base.
So, using this effect, it is possible to edit a video with infinite zoom between the arts. Yes, it’s a good job, but the result is simply amazing, at least in my opinion.
Creating an image in DALL-E 2 with another image in the center
To make one image from another, you’ll need a program like Photoshop, but any other competitor (online or otherwise) will do the trick. The first step is to generate an image in DALL-E 2, which in my example is the image of the computerized eye.
Then, just open this art in an image editor and reduce the size by 50%, that is, to 512 pixels, leaving the image area centered, with the rest of the area transparent. Save this image to your computer.
The next step is to upload this image with the transparent border to the DALL-E 2, and then click on Edit Image. DALL-E 2 will ask you to click on a part of the image to make the mask.
Click on the blank side. Then, in the text field, create a prompt to specify the details that will be next to the original image.
Choose the variation you prefer, and save the image. Repeat the process in Photoshop or similar app to make it 50% full and the edges transparent.
Going back to DALL-E 2, upload the image, and think of a new prompt. You can repeat this process as many times as you want, and thus create a video with infinite zoom going from one point to another completely different.
50 credits for free plus 15 per month
In the DALLE-2 it is not possible to generate images with higher resolution, as in the rival Midjourney, so they all have the same size, 1024 x 1024 pixels. Even if you run out of 50 credits, you can continue using the app for free, as OpenAI will give you 15 credits that can be used in a month. Anyone who needs more can buy 115 credits for $15.
Yes, its competitor Midjourney manages to present more artistic results, so to speak, but the DALL-E 2 remains unbeatable in its task of recreating the real world, or something very close to it. Speaking of Midjourney, if you haven’t read my text about it, just click here.
In addition, I also recommend listening to our Sync podcast on the topic, in which we talk about both DALL-E 2 and Midjourney.
Have you watched our new videos on YouTube? Subscribe to our channel!