Raul Crist
AI image generators A beautiful childlike face of a redhead woman with a parrot on her face looks like a photograph. A famous painter who paints himself. A Nike shoe with a jellyfish. All these ideas can be created in a few seconds using image generators on the web. The image generators with the names Dall-E, Midjourney, or Stable Diffusion are all based on artificial intelligence (AI) that can create such impressive images.
New opportunities
Completely realistic photos in a specific art style are also possible, as well as infographics on specific topics or web design ideas. For many professions, image generators could be an important addition to the design.
The new technology, which is currently experiencing a massive boom, can help graphic designers to come up with new ideas for logos or artworks, architects design interiors, and fashion designers to create patterns.
This is how AI generators work
But how do computer-generated images actually come about? Most of the AI models mentioned use “neural networks” for this. These neural networks were previously trained with a gigantic amount of existing images. About thousands of pictures of a cat were entered into the computer system. The AI image used these photos to learn exactly what a cat’s characteristics are in order to then imitate them. Most AI image generators are based on several different databases.
Many of these images, which the neural networks use to learn, come from the internet. Companies such as OpenAI are specifically looking for images and the associated text descriptions that the machines can use to learn independently. The better the image description, the better the database and thus also the images generated by the AI.
Tying out made easy
So it makes sense to try different image generators to find the one that best suits your needs. The new AI image generators are not only reserved for professional users. Anyone can try it out online for free.
One of the image generators that are free to try is called Dall-E. It’s from the same company as the popular chatbot ChatGPT, which writes texts for you. OpenAI was one of the first companies to publish an image generator, and many other image generators followed. In addition to Dall-E, there is also Midjourney and Stable Diffusion.
There are only organizational hurdles lurking on the way to trying it out: With Dall-E and Stable Diffusion you have to create your own account, and with Midjourney a Discord account is a prerequisite for use. Unfortunately, with Dall-E you also have to enter your phone number.
This is how short texts become pictures
All three AI image generators mentioned convert short text inputs into images. For example, if you enter “black cat sitting on the sofa”, you will see 4 images of such a display. The photos the generators spit out mostly look believable. With some objects, you can see that they have not been executed cleanly. For example, the cat’s demarcation from the background could be indistinct, or the whiskers could be missing.
In the illustration, we tried the following input: “black cat sitting on a pink couch, color, living room background”.
3 tips for the best “prompts”
The more information you enter as a so-called command (“ prompt ”), the more accurate the results you get. The term prompt comes from prompt engineering. The developers of the AI models describe the text input field as a “prompt” in order to elicit images from the AI. So that you don’t have to experiment with “prompts” forever, or get disappointed because the picture doesn’t live up to expectations, here are a few tips:
- In general, it is advisable to describe everything that you want to see in the picture as precisely and in detail as possible. It can be quite long phrases that you can enter. Some generators like Stable Diffusion also have an input field where you can specify what you do not want to see in the image.
- In addition to a long, detailed description of what you want to see, a “prompt” can also contain certain additional information. For example, you can enter art styles separated from the main phrase with a comma, such as pop art or comic, or you can enter the names of well-known artists to get works in their style. If you want to get motifs that are as photo-realistic as possible, you can also use words like “ hyper-realistic ” to tell the AI that you want the picture to look as real as possible. The imagination knows no limits.
- If you cannot imagine exactly what you could enter as a prompt, you can use what other people use as a guide. This can be done, for example, via the “Lexica” search engine. Beginners can click through the pictures there. If you want to create AI-based images, you will find a lot of inspiration here.
Not for nothing
If you want to use the tools professionally, frequently, and at high quality or speed, you have to pay money for them. All AI image generators charge certain fees if you use them regularly. One of the reasons for this is that it costs companies computing power – and therefore money and energy – every time new images are generated.
With Dall-E you can try 50 times for free, with Midjourney it is 20 and with Stable Diffusion, it is 100, whereby 4 images are always generated at once. With Dall-E you can then buy additional credits, with Midjourney there is a subscription model, and with Stable Diffusion, there is also a monthly membership.
What else is there to consider?
In most cases, you can use the pictures that you have created with AI privately, for example on your own website or on Instagram. However, you should first read the general terms and conditions (GTC) of the image generators. Especially if you want to use the works for business. There is no copyright for AI images, but the respective companies behind the image generators can set license terms for use. From a legal point of view, however, there are also some gray areas
around AI images, which primarily affect the training data and thus the companies behind the AI image generators. Since the training may also have used works that were protected by licenses or works by artists who had explicitly not allowed this, there could still be some court cases to clarify certain things. Both Getty Images and some artists have already filed their first lawsuits.
