How to use the DALL-E 2 neural network

What is DALL‑E

DALL‑E is a neural network from OpenAI that generates images based on text queries and can also create image variations based on user-suggested examples.

The tool is designed based on the model GPT‑3, who was additionally trained to create graphically people, animals, as well as a large number of other objects. Also, the service is able to combine unrelated concepts to form new ideas.

The first version of DALL‑E was released in 2021, and a year later, the developers introduced an improved version of DALL‑E 2, which supports higher image quality and advanced options for converting ready-made pictures.

The system defines queries in hundreds of languages, including Russian. But the main language for effective commands is English. First, the neural network recognizes the text entered by the user and generates a simple sketch from it, then converts it to the final version of a small resolution and then scales it up, adding details.

What the DALL‑E 2 neural network can do

As mentioned above, the model from OpenAI can not only generate images on demand, but also create projects by mixing different concepts. The neural network connects two frames and forms another one from them.

Using the Outpainting function, you can draw background Images. It is interesting to complement finished paintings or photographs, developing new stories on them. It is also possible to change the composition, edit the shadows and texture of the picture, add and remove objects in the frame.

One of the features of DALL‑E 2 is working with downloaded files. Using a user-friendly minimalistic interface, it's easy to add elements to photos, generate fantastic landscapes or increase the detail of pictures.

What are the disadvantages of DALL‑E 2

At the moment, DALL‑E 2 is not a perfect instrument. To get good results, you need to experiment, try different queries and tweak the parameters. The neural network periodically produces funny or even frightening frames with distorted objects and incorrect arrangement of elements.

The system is especially often mistaken if you give it long requests and giving too many specific details. It doesn't define professional terms very well, and also gets confused by using negations or exclusions instead of direct indications of what should be in the frame.

How to use DALL‑E 2

You can start using the system on the OpenAI website at section DALL E 2by pressing the Try DALL‑E 2 button. If you do not yet have an account on the site, then you can get it through your email address or existing Google or Microsoft profiles.

If you're trying to connect from Russia, you'll need a VPN and a virtual number to sign up for an account. You can read more about creating an account on the OpenAI website in material about ChatGPT.

Once logged in, you should have 35 free image generation credits for a month. Each next month, 15 attempts are given. You can replenish the balance in your personal account using the Buy credits button. You will have to pay a minimum of $15 for 115 requests. Unfortunately, users from Russia cannot replenish the balance directly now.

There are several options for solving the problem: ask a friend with the appropriate access to the payment system to pay the bill or provide an account for a while, find and buy a ready-made OpenAI profile through online trading platforms or use other services on DALL‑E base.

The most convenient alternative Microsoft image generator. This tool runs on DALL‑E 2 algorithms and gives almost the same results for free. But it has no mode. picture editing, completing the frame and other useful features. The analog can be used only to generate images, and then correct the result with other programs. You will need a Microsoft account and a VPN to use it.

It is better to make a request for generation both on the DALL‑E 2 website and in the Image generator in English, but in Russian the system also produces good results. The Surprise me button, which is located next to the input line, will automatically issue an unusual request for testing the neural network.

To get a quality result, use queries that clearly describe the desired frame, but without unnecessary details. Enter the image type in the input line. It can be a portrait, a watercolor picture, a pencil sketch, and the like.

Also specify the lighting option in the frame and style. For example, a realistic display, as in a comic book, or a specific style of a famous artist. It is desirable to supplement this with an approximate level of brightness.

At the end of the line, you can add the context of what is happening. For example, that the frame contains not only the lunar rover, but also the moment when it moves against the background of the Earth. Or a stage where plush hares fight aliens.

The request does not have to include all the items, the result can still be interesting, but the success rate will be noticeably lower. Help build a successful team specialized tools, such as Promptomania.