The best AI art generators in 2023

The biggest AI art generators we have discussed so far require payment, at least initially. The open source Stable Diffusion model is the largest AI art generator available for free. The possible downside is that it requires a bit more effort. You can run Stable Diffusion on your own computer, but you will need a powerful machine and some installation knowledge. Alternatively, you can run it online through various methods, including using Google collabs for free, or you can use the model more conveniently but for a fee through Stability AI's web tool DreamStudio.

Either way, Stable Diffusion is a powerful and adaptable AI image generator that can compete with DALL-E 2 in terms of precision and the variety of styles it can produce. It has the advantage in some styles and offers more flexibility than DALL-E 2 in terms of aspect ratio (although this can cause problematic artifacts), the ability to increase the resolution to 2048x2048, the option to add a negative warning, and the ability to use the seed number to limit the randomness of image generation.

Other cool features include a depth-image tool that can translate the sense of depth from a loaded composition into a rendered image. Although the user interface (UI) is not as intuitive as that of DALL-E 2 and you have to learn a few terms, it is an extremely versatile AI image generator. Developers are allowed to use the code for free in their own projects by downloading it from Github. If you want to use DreamStudio's out-of-the-box option, it costs $10 for 1,000 credits that can be used to create about 5,000 photos when using the default settings.

Midjourney is the program used to generate the "last selfie" photos that won a fine art competition and went viral last year. It initially had a more limited artistic range than some of the other AI art generators and was often used to create impressionistic fantasy landscapes. The AI image generator that has made the fastest progress, though, is Midjourney. Now at version 5.1 (or maybe higher by the time you read this) it can create incredibly realistic photos in a variety of styles.

The interface takes some getting used to because you have to use the instructions to ask a robot to create an image on the Discord social chat platform (enter '/imagine' to bring up a text prompt), but it doesn't take too long. In public channels, everyone can see your prompt and you can see everyone else's prompt, which can be a useful way to get an idea of how the model understands descriptions (on a $48/month 'Pro' account, you can create a photo privately with the '/stealth' command).

The biggest disadvantage of Midjourney is this. There is no longer a free version, so you have to subscribe to use it. It costs $8 per month for a basic subscription valid for up to 200 photos. 30$ per month will get you unlimited pictures. With more expensive subscription levels you get faster generations.

For those who are just interested in technology and want to play without the hassle of registering or buying credits, we believe Craiyon is still the best AI art generator. Visit the website, enter the information you want and Craiyon will start generating. The results can still be absolute nightmare material, especially since the human forms are often horribly distorted. It's a bit slow and you'll have to put up with ads unless you subscribe.

But that doesn't mean Craiyon is useless. It can offer intriguing surprises and is surprisingly varied, making it a fun tool for experimenting with concepts. It has a small number of features, though. Although you can now add negative suggestions, there's not much you can do in the app itself, as there are no in-app editing tools or image-to-image builder. Strangely, you can buy t-shirts with your generations on them, which can be annoying. Before OpenAI talks to them, remember that Craiyon is known as Dall-E tiny.

Adobe came rather late to text-to-image creation, but it made up for it in 2023 with what we believe is the greatest AI art generator for the majority of professional designers. Still in beta, Adobe Firefly is the most useful AI image generator we've seen to date, thanks to its connectivity with other Adobe Creative Cloud tools.

Since Firefly is trained with fewer resources than some of the models we discuss below, we discovered that it has certain limitations. However, we believe these disadvantages are more than compensated for by the variety, usefulness and adaptability of its editing tools, as well as its simple interaction with other Adobe products. Compared to most other top AI art producers, it also offers a much more user-friendly user interface.

The text effects model and the vector recolor model are just two of the many AI features available in Adobe Firefly for various applications. We found that the text-to-image tool makes it easy to quickly replace images and generate various versions of components. The Firefly-powered Generative Fill feature in Photoshop can create new objects in non-destructive layers with all the necessary properties, including shadows and reflections, to match the existing image, greatly speeding up the compositing process.

Adobe Stock assets and public domain materials were used to train Firefly. Some Adobe Stock contributors are unhappy with Adobe's strategy because the company seems to be in no hurry to establish a payment plan. We are therefore hesitant to call it an "ethical AI art maker", even though it provides more moral and legal security than alternatives.

One component of Adobe Creative Cloud is Firefly. It has its own browser-based tools and Adobe has started to incorporate some of its features into Adobe Express and Photoshop (beta). Text-to-vector, text-to-pattern and text-to-stencil tools are among the upcoming additions.

With the exception of Adobe's special text effects tool in Firefly, text is one of the biggest drawbacks of even the best AI art makers. While tools like Midjourney have begun to solve the problem of human figures having six fingers, written language is still often represented as alien hieroglyphics.

All in all, DeepFloyd IF's exceptional text generation capability is what makes it what it is. The Stable Diffusion team has actually produced this AI art generator and it is also free and open source. For certain styles, the images themselves can at times outperform stable diffusion (specify "Polaroid" at the command prompt and the results really look like a Polaroid). However, the ability to instruct at the command prompt to add text to the image is what stands out the most. For example, you might want Lisa Simpson holding a sign that says "Creative Bloq" or a bar in the middle of the desert at night with a neon "open" sign.

We discovered that it is quite random, as with all AI image generators, and the words sometimes still sound partially distorted. Whatever the reason, the model often adds a second "q" to our name, making it "Creative Bloqq". The ability to add text to collages, textiles, stained glass windows and more, in addition to signs, is the best text processing feature we've seen in an AI image generator. This will be the stuff of meme makers' dreams. Hugging Space is now demoing Deep Floyd IF on a research-only, non-commercial license.

DALL-E 2 is, in our opinion, the easiest and most straightforward of the more sophisticated tools, making it the best AI art generator for individuals who want to start studying the technology. It is well known, hence the many tutorials available, simple to use and capable of producing amazing results, including detailed photos with a high level of realism. The start is quite fast, the primary text-to-image tool is simple to use and we discover that the image creation process is quite fast.

We found that DALL-E 2 can create clear visuals that often match the prompt description, including features such as camera lenses or creative styles. If you want more customization, this text-to-image builder may be a bit too restricted. There are now restrictions on the resolution and aspect ratio of the 1024x1024 canvas, and there's no way to add negative prompts or the various precise instructions that Midjourney allows. On the other hand, the inpainting and outpainting editing capabilities provide a great deal of versatility for replacing specific picture elements or "cropping" an image outside of its original frame.

The best part is that DALL-E 2 offers consumers a free monthly top-up of 15 credits, as well as enough free credit to get you started. They don't go too far, but at least they give you a chance to try something out before you have to pay for additional credit.

The biggest AI art generators we've discussed so far all produce 2D graphics from text instructions. But SHAP-E, a more recent invention from the creators of DALL-E, does the same thing in 3D. It's the best 3D AI art generator we've seen so far, and the assets can be exported as .glb files for viewing in Microsoft Paint 3D or Blender, or converted to STL files for 3D printers, but they are by no means the most stunning 3D assets.

Shap-E is designed to allow game artists to generate objects for virtual environments and architects to choose the sizes, materials and colors of buildings. While not quite there yet, it will no doubt be interesting to watch, given how far AI art generators have come in the last year. You can get the model weight, inference code and examples on the Shap-E GitHub website. You can install the code using the Python pip command; an NVIDIA GPU and a powerful CPU are required. Hugging Face has an online demonstration.

CF Spark is a little more than an AI image generator. It is a component of Creative Fabrica, a website where you can buy digital assets, including fonts, images and handmade products. The AI art creator is available in a completely free version, but if you pay a membership fee, you can profit from the sale of your creations.

In terms of the variety of capabilities it provides, we consider the generator to be among the largest AI art generators. You can create text, patterns, translucent PNGs, coloring pages and text-to-image models. It also comes with a tool to create prompts for AI tools; yes, it is an AI tool. The community feed is very active (people have created over five million photos as of this writing). Here you can view the work of others for inspiration, their tips and seeds, and try "remixing" their art.

The downside is that there is no great degree of control over the generations and only a limited number of predefined artistic styles (kawaii can be cute and some illustrative styles can be effective, but this generator is not great at photorealism and the 'hyperrealism' style is not very hyperrealistic). Generating images using the free plan can also be slow (paying for a subscription allows you to generate images faster and run multiple generations simultaneously in the queue). CF is always adding more features, so some of these may change.