You must have heard about it in recent weeks DALL-E 2a project by the artificial intelligence company OpenAI (participated among others by Elon Musk), which after months of closed-door testing is open to anyone who wants to use it, although it remains in beta.
After seeing and reading a lot about DALL-E 2, an AI that, it should be noted, is capable of creating original images from the written descriptionI was encouraged to give it a try to see if it was, as some have ventured, the early death of digital creators. All I can expect is that right now, not even close. But it’s still a very strange development that I have no doubt will come to change things, but in the future.
Specifically: DALL-E 2 can generate images based on what you describe to it, so the description must be accurate because everything you enter counts and the AI will interpret it in its own way. That thing has a crumb, yes. Once you enter the pages of DALL-E 2, you will see a lot of images, all of which have been created by the service, and all of which exude the same slightly tacky, if not repulsive, style of surrealism. The truth is, there is a little bit of everything, which is why the above description is important.
As for its capabilities, in addition to describing the image you want DALL-E to create for you, it is possible to make simple adjustments to the result, such as deleting image elements, generating variations or scaling based on z blocks frame original from 1024 × 1024 pixels… Because that’s another one: the size is not modifiable, at least in this first iteration. Not much, although the editor is also in development.
Otherwise DALL-E is a service free but limited: once you register, they give you 50 credits, each of which is used to generate images, that is, four images for the same description; once these credits are used up, 15 credits are given out per month. In addition, additional credits can be purchased.
Also, keep in mind that to use DALL-E, you need to create an account with the service and verify your phone number, which is a basic requirement.
Other DALL-E 2 limitationsthose of an “artistic” nature include recreations of real famous people, scenes that suggest violence or sex, illegal or highly controversial subjects and, very strangely, faces, an aspect that was supposed to be overcome over time by facial deformation. to prevent –deep pretense at first – it’s not clear what, but it continues to feel scary… with exceptions, as will be seen below.
And there’s not much to say about the DALL-E 2… or rather the opposite: there’s a lot to say about a technology that, we won’t deny, has a strong dose of innovation behind it. But if you’re more interested in its details, you’d better satisfy your curiosity straight from the source.
I leave you with some of my creations in DALL-E 2… or rather some of the creations that DALL-E 2 made based on my requests. To get an idea of what DALL-E 2 can achieve with a few hints (in choosing the exact words, which ingredients of the recipe, is the secret of this invention).
But first: here’s the DALL-E 2 website where you can sign up if you fancy trying it yourself.
Various examples of DALL-E 2 “genius”.
Here are some of the many requests I made for the AI in DALL-E 2 and my impressions of each one… as you see fit. On top of all that, each one can take about 10-20 seconds to generate
I started with something simple in my opinion, the first thing that came to mind… and the moñiga came out. After using it for a long time, I learned it copyrighted topics are sensitive and that to get closer to the picture I had in mind, seeing an epic picture of a battle between two characters in a good setting, I would have to be much more specific.
So I asked for something even simpler but less specific, like “the land of Hyrule” (because I had Zelda in mind). And the result was better… with a but. Here’s what one of these images looks like in higher resolution (remember: they’re 1024×1024 pixels; unless you start gluing the pieces together like a puzzle, DALL-E 2 won’t make you wallpapers amazing):
It looks more like a painting than a photo, but it works.
I wanted to see how far the copyright issue would go with a character as famous as Harry Potter and… let’s just say the result is just as disappointing, even if it seems more accurate. By the way, I tried it in Spanish and understands to perfection.
Since there aren’t two without three, I’ve finished the review with the image you see, and as always, the more detail the scene has, the closer it will be to what you’re asking for; so I consider it valid, based on the brevity of the description.
I tried to investigate the copyright issue a little more with what I intended to be a typical catalog image and… yes, the intention was captured, but from the title nanay, and worse, I ran into the theme for the first time “horror faces«…
Neither the uncanny valley nor the bagpipes: it’s somewhere between disgusting and creepy, not what DALL-E 2 does with faces…
So I went back with it more specifically and to that repeated horror, I tried again in much more detail…
And, well, let’s just say that at least this time DALL-E 2 didn’t give me back the monsters. About tastes already… And yes, it is normal that they limit perversions in the service that tries to make a collection; what i don’t understand is why it generates those horrible faces by default.
I continued my research and asked about AI stuff to see how small changes affected the outcome, for example with the visual style used.
And as you can see, it’s all a bit strange…
… But at the same time, he has a pass. Again, maybe I should have specified the style…
Since I have returned to more mundane themes and this century and things as they are, you will fool many with these images.
What if I provide more specific details in addition to dictating in Spanish? The truth is that I was surprised by these pictures, although things as they are, if you inspect them carefully, you can see the seams and they have a similar effect to the faces: the effect of rejection.
I took this a step further and asked him for something that you can find a million photos of on the internet, but what he got back seems fair to me, one came back. There is the best stop of the four…
No, I won’t go into the fact that it’s not a Valencian paella. Neither that nor what is seen in the other pictures. But the thing is, if anyone finds it tasty, I feel sorry for them… because it looks like mortar. In other words: either the rice has reached the point of mush, or the image has lost enough resolution to create this effect. I know what you’re thinking: what nonsense…; but no, because since DALL-E 2 is a tool made for advertising and for advertising… It makes me sick.
So I tried again, this time much more specifically.
Much better, right? As they say, the devil is in the details and these pictures, in the details…
What do you think? would you eat it Maybe I’ve lost track of reality, but it still feels inauthentic, even if you’re asking for it. With this example, I actually abandoned any “realistic” wishes for DALL-E 2, although I still had a few things left to test before completing my first session with the AI.
The first was to ask for a picture in a language other than English and Spanish, and since I got where I got to, I did it in Catalan. The understanding got me, but the result was again a little disappointing. Let’s just say that without a Tolkien-esque description of what you want to see, DALL-E 2 has little imagination.
Finally, I wanted to try creating variants from an external image, the DALL-E 2 feature they’re already pointing out could be controversial given that it could potentially be used to fake anything, but I wouldn’t worry too much about the result either (although there’s a gap between this and the Barcelona beach pic a section that is difficult to cover). That’s why I asked for a variant A walk on the beach by Joaquín Sorolla, one of those types of painting that appear to be done with brushstrokes, but when you look closely, you catch the details, the light… and realize that they are not…
And what DALL-E 2 brought back to me in a pseudo Soroll style, but in everything much dirtier, is… more monsters. Compare it if you want.
I wanted to call this entry something like «DALL-E 2, the artificial intelligence that Sorolla wanted to be, but couldn’t even get into p…». And I thought better of it. Because it may happen that DALL-E 2 is not yet close to the mastery of established painters, that it is not yet able to approach the know-how of a good illustrator, a good photographer… or that almost everything that comes out of its servers are the same small mistakes, and that if you’re not looking for an image with a touch of eccentricity and a special style, even if you mark the style you want, you won’t get anything really consistent. But as is clear, the emphasis is on yet.
That said, this is just the first part of my adventures with DALL-E 2. Once I learn how to get the most out of it, I may post more… or maybe not. Just in case, I’m going to say goodbye with one more picture to close the circle…
Source: Muy Computer
Alice Smith is a seasoned journalist and writer for Div Bracket. She has a keen sense of what’s important and is always on top of the latest trends. Alice provides in-depth coverage of the most talked-about news stories, delivering insightful and thought-provoking articles that keep her readers informed and engaged.