A girl looks at her futuristic smartphonein the background there is a cyberpunk city: this scene does not come from a film or a video game, but was generated by artificial intelligence, starting from a textual request.
To simplify, we have written in English the sentence with which this article begins (a girl looks at her futuristic smartphone), we hit Enter and waited for something second: the result is what you see above.
Yes, because creating an image starting from a text thanks to artificial intelligence (the so-called TTI, Text to image), is now an extremely widespread practice. Originally, it was From-E 2, the Open AI project we wrote some time ago. In recent months, other platforms have emerged, with similar functioning, which have definitively democratized access to this technology. In other words, if Dall-E 2 (as well as Imagen, which is from Google) remains a research project, which few can access and invite only, that same technology is now available to virtually anyone, in other forms. So that, even, an AI-generated image went on to win an art contest in Colorado (video).
If the process behind AI image generation is extremely complicated, creating photos or drawings, for a common user, is quite simple. From a technical point of view, that’s enough know the services and understand how they work Basic. We have identified 3 of the most popular: here’s how they work, how to access and some tips for writing requests.
Meta’s idea: improve AI by using AI to improve Wikipedia
by Emanuele Capone
Midjourney is among the most popular research projects dedicated to the generation of images through text. It is an algorithm which, starting from a description, is able to generate an image. To use it, it is necessary use Discord (how does it work?): yes, yes subscribe to this server and then join one of the #newbie channels, any one. At that point, you need to enter the command / image, followed by the description of what you want to generate. The image will appear, after a few seconds, in the chat. Each user has 25 free trials available, before being forced to subscribe: plans start at $ 10 a month.
Stable Diffusion (which we used to generate the opening photo) is among the latest born: an artificial intelligence project that has made a lot of talk about itself just released to the public, at the end of August. Unlike the others, the Stability.ai model is completely open source: anyone can download it, install it on their computer and create dedicated applications. Stable Diffusion works similarly to Midjourney, but it has a specific web platform, which is called Dream Studio. Operation is intuitive: just enter a text, wait a few seconds and look at the result. Also on Stable Diffusion the trial is free, with 200 credits that are loaded upon registration and that allow the generation of 100-150 images, depending on the degree of difficulty. In this case, there is no real subscription: on the other hand, there is the possibility of buying credits from time to time (one for just over one euro).
Pixelz.AI is an application, of which there is an effective version for smartphones, which uses the Stable Diffusion algorithm and Dall-E 2 processing to generate images. The app, available for Android and iOS, has a fairly simple interface, which allows you to choose the type of image to create (artistic, realistic or with the Stable Diffusion algorithm) and then to generate it in a similar way to the others. Pixelz.AI is also free in small doses: it offers 10 credits a day to all users, which usually allow the creation of 3 images. For those who want something more, it is possible buy credit packs (50 cost € 2.99).
Is it possible to fall in love with an artificial intelligence?
by Francesco Marino
Instructions for use: how to write requests
It all sounds extremely easy, and indeed it is, at least in purely practical terms. You log in, write the request and the artificial intelligence does the rest. But that’s not quite the case. The challenge with TTI is being able to talk to AI; in other words, to be able to write requests that can produce the desired results.
Prompting (that’s what it’s called) is a real science, which has a lot to do with the relationship between man and machine. TO starting from the Dream Studio blog, let’s try to give some basic tips, useful for getting started. An important note: currently, all TTI systems work from English. This means that all requests must be produced in this language, but those who have difficulty can get help from Google Translate.
Always start from a simple word, able to summarize in the most suitable way what you want genre. A panda, a skeleton, a dog: the backbone of the request must always start from a clear idea. That does not necessarily have to be something concrete, but it may very well be a state of mind.
Beside this basic concept it is always important specify the style of the image you want to generate. And then add specifics like “a photo of” or “a painting of”. It’s a lot of fun, for example, to play with the styles of the artists: above is the Champion’s League final as seen by Vincent Van Gogh, generated with MIdjourney.
At this point, it is useful to add whatever specifications I may help artificial intelligence understand even better the request: they are very fashionable, and are effective, adjectives such as beautiful, realistic or very detailed.
To get some ideas and learn the basics of AI conversation, you can take a look at Lexica, a prompt search engine from Stable Diffusion. Just enter a word to see all the images created with that keyword and the requests of other users. And be inspired.