What is an AI image generator

An AI image generator is an artificial intelligence tool, application, or software that can create a unique image based on a text prompt, photo, or other input. Such tools are trained in a machine learning mode. They are "fed" thousands of pictures, as a result, they can recognize and recreate certain features, image styles, elements, etc.

Even though AI graphics may seem to be a recent invention, the first models appeared more than 50 years ago, in the 1960s. Of course, they were neither ubiquitous nor user-friendly, and the images were not elaborated, complex and creative.

In 2014, GAN (Generative Adversarial Networks) technology appeared. It enabled the creation of quite realistic images, as the system worked in two flows simultaneously: it created images and evaluated the output right away.

The next breakthrough occurred in 2021, with the introduction of DALL·E (spoiler: this tool will be reviewed later). From then on, AI image makers started emerging like mushrooms after the rain. Nowadays, the market offers a plethora of AI photo generators, both free or on a paid basis, for complete novices and more experienced editors.

How to use AI image generators

AI-generated images are used in various spheres. The list includes but is not limited to:

  • Personal vlogs;
  • Marketing campaigns;
  • Creative projects;
  • Illustrations for research and statistics;
  • Abstract art;
  • Augmented reality and VR.

How to generate AI images? Nowadays, it's as easy as shelling peas. All you have to do is create a comprehensive prompt — a text description of what you want to see in the output. A good prompt has to include the subject, the desired style, and, optionally, the needed parameters (for example, resolution or orientation).

For comparing various AI image makers, we used three prompts:

  1. An image of a hedgehog cooking borsch in a steampunk style.
  2. A realistic photo-style portrait of a young red-haired girl who holds a sheep.
  3. An image in a pre-raphaelite style of a boy playing the flute in a spring forest.

7 top AI image generators (that you can try for free)

  1. DALL*E 3

    DALL*E was a precursor of the whole AI images craze. It is integrated into ChatGPT, so you can use it immediately without switching the tools. The chat gives you only one image as an output and asks for feedback to correct something if needed. In our case, we had the following results:

    The result is quite good, but for the last image, where the boy looks a little creepy and can't match the flute and the lips.

    ChatGPT allows you to create three images per day for free. If you need more, you can subscribe to ChatGPT Plus.

    Advantages

    • User-friendly interface.
    • Natural language queries.
    • Immediate modifications
    • Fast rendering.

    Disadvantages

    • Limited output.
    • Only 3 free images per day.

  2. Bing

    Bing image creator is also powered by DALL*E 3 AI, but the features differ a little. The output contains four pictures (maybe that's why their rendering takes a little longer). While waiting for the output, you can read the hints on creating effective prompts.

    The new user has 15 free credits; each query "costs" 1 credit. You can also press the "Surprise me" button to get an image from a random prompt.

    The tool doesn't ask follow-up questions, and you can't alter the generated image, but you still have a choice, as there are 4 of them in each output.

    Our outputs (we chose one image from each bunch).

    The result is not perfect. The hedgehog pic has hallucinations (look at the clock, for example), the red-haired girl doesn't look realistic, and the spring boy still can't hold the flute properly.

    Advantages

    • Hints for better prompts.
    • 4 pictures in the output.
    • Free credits.

    Disadvantages

    • No modifications are possible.
    • Hallucinations.

  3. Neural.love

    Neural.Love is an undeservingly little-known resource. Meanwhile, it's user-friendly, allows for generating images and videos, and offers a high level of customization.

    The free mode allows the creation of two outputs per hour, with two images in each. If you want to increase the resolution, get more images in an output, remove the watermark, or customize the process another way, you'll need to purchase credits.

    Our outputs:

    Have a look at the paws of a hedgehog: one has 6 fingers. However, in contrast to the previous tools, the image of a girl is much more photo-like. with a boy, all the main points are considered correctly, but the flute is significantly below the needed level, and the boy is married for some reason.

    Advantages

    • User-friendly interface.
    • Easy customization.
    • Input may be a text query or another image.
    • Relatively short period of "recovering" for free queries.

    Disadvantages

    • Watermark for free outputs.

  4. Leonardo AI

    Leonardo AI is a well-known tool for digital artists, designers, cartoonists, etc. The setup process takes some time, as the resource quite thoroughly inquires about a new user's needs and preferences.

    The free plan offers 150 tokens daily, each query "costs" 24 tokens, and the output contains 3-4 images. The tool is highly customizable, but it is not a good choice for creating photo-like pictures, as it tends to generate more cartoon-like outputs.

    The tool features a list of unique parameters, such as a "negative prompt" (A description of elements that should be excluded from the output.) or automated prompt improvement.

    Our outputs:

    The girl is not even close to a "realistic photo-style portrait", but cute. And what is this tiny creature in the grass near the boy? A micro deer?...

    Advantages

    • Intuitively comprehensive.
    • Good image quality without many hallucinations.
    • Free tokens are renewed every day.
    • Lots of customization settings.

    Disadvantages

    • Post-generation editing tools are available on a fee-based plan only.
    • Output is not realistic.

  5. Canva

    Canva AI is another quite popular freemium tool. Each user gets 20 non-renewable credits; each prompt "costs" 1 credit and outputs in 4 pictures. You can set up the orientation and style.

    Unlike most competitors, Canva doesn't use your projects to train its algorithms (at least it claims so), so they stay completely private. Another feature is that you can edit the generated image immediately: apply various filters, change the color scheme and intensity, flip or mirror the image, etc., and it will not cost you additional credits.

    The images are not very realistic, and the tool doesn't stick to the demanded style if it's exotic. For example, the service didn't recognize the "pre-raphaelite" setting in our case.

    Our outputs:

    The right hand of a boy has 6 fingers; other pictures in this output had hallucinations as well.

    Advantages

    • User-friendly interface.
    • 20 free queries.
    • Free post-generation editing.
    • Projects are private and secure.

    Disadvantages

    • Non-renewable credits.
    • Cartoon-like pictures, even if the prompt asked for photography.

  6. Adobe Firefly

    Adobe Firefly is a relatively new player in the generative AI market. The free plan includes 10 generative credits that renew monthly. Each input "costs" 1 credit and provides a user with 4 pictures in the output.

    The tool allows you to provide a reference picture, also it is already integrated with the Adobe editing tools, so you can integrate the rendered image into your project or alter it as you wish.

    As for the text prompts, the tool leaves something to be desired.

    Our outputs:

    The output with a boy is the poorest. For some reason, the tool decided to place the scene in Japan and depict it in anime style.

    Advantages

    • Integration with other Adobe tools.
    • Easy and immediate output editing.
    • Reference image as an input.

    Disadvantages

    • Sometimes poor image quality with hallucinations.
    • Struggles with complex queries.

  7. Ideogram

    Ideogram offers 10 daily credits on the free plan, which results in up to 40 pictures (10 prompts). All the projects are public by default; that setting may be changed with upgraded plans.

    Rendering time is a little bit slow, but the result is quite decent. The majority of functions are available with paid subscriptions, but even the limited functionality is enough for creating art projects and other types of visual content.

    Our outputs:

    Are there any holes in this flute?

    Advantages

    • Intuitively comprehensive.
    • Output mostly matches the prompt.
    • Many customization options.

    Disadvantages

    • Most options are paid.
    • All pictures are free by default.

    8. AI-generated images limitations

    Even though generative AI algorithms have recently made a quantum leap, they still can't fully replace human artists. Rather, they should be considered accessory tools, not "turn-key" services.

    Here is a list of some of the most ubiquitous AI-generated image considerations.

    1. Hallucinations. Even the best tools still tend to err on the number of fingers or limbs. They experience issues depicting points of the elements' interaction (e.g., a flute and the lips or fingers on the flute), and sometimes they generate weird elements (non-recognizable deer-like animals in the grass).
    2. Ethical issues. What if a generated portrait resembles a real person, for example, a celebrity? What if it contains distinctive features of someone's art? Are such outputs prone to copyrights and personal data protection? So many questions, so few answers.
    3. Usage limitations. Generative AI will not create violent or obscene content. It will not generate public figures (if it does, it will be a deepfake, which is illegal).
    4. Limited "knowledge". Each tool is "trained" on big data and may not know niche concepts or recent trends.
    5. Legal use. Some platforms allow customers to use AI to create images for personal needs only. Also, they add your projects to their open libraries so that other users can avail of them as well.
    6. Limited fine-tuning. While some platforms offer immediate output editing, others don't. In these cases, you have to restate the prompt and waste precious credits.

    As for the last point, you can use third parties' tools, such as online video editors to include rendered pictures in a large video project, combining them with music, subtitles, text-to-speech voiceover, stickers, and other elements.

    Generative AI tools are continually evolving, with new features emerging and changing daily. We recommend visiting each tool's webpage for the most up-to-date information.

    FAQs

    What is an AI image generator?

    It's a tool, service, or software for creating images based on a visual or text prompt. Such algorithms are taught on big data, so the more people use them, the better and more accurate their results are.

    How does AI generate images?

    If we go into technical details, generative AI mostly uses GANs (Generative Adversarial Networks) and VAEs (Variational Autoencoders). The former uses two neural networks: a generator renders an image, and a discriminate evaluates the output. The latter generates new images resembling the ones from the training data.

    Who owns the rights to AI-generated images?

    That's a tricky question without a simple answer. So far, it's a "gray zone": if we consider AI a real creator, then it owns the rights, or, as AI is just a tool, rather its developers. On the other hand, the end user comes up with the prompt and performs final editing and fine-tuning, so, theoretically, they hold the copyrights.

    Read carefully the terms and policies of every service you use.

    Final thoughts

    There is no doubt that our future lies with generative AI. So far, it's just an imperfect tool, but the day is not far away when it will become a self-sufficient creator that requires little or no human interference. However frightening it may sound, it also leaves much room for collaboration and using such systems to our advantage.