THE text to image or applications able to convert a text into an image, were one of the revelations of the year 2022. Artificial Intelligence, traditionally associated with activities such as robotics, computer vision, language recognition or the behavior of characters in video games, this time tackled art and aesthetics, areas in which we would not necessarily have imagined that it could interfere. The surprise was there.
After having tested most of the applications available to date, we have selected 10 of the most outstanding, whether by the quality of the images, the creativity they demonstrate, their originality or the ease of use. We’ve also eliminated some of them for reasons of initial price – those that appear here can all be tried for free, with the possibility of obtaining privileged access if one is willing to pay usually reasonable sums.
From and 2
Give honour where honour is due, Give her was the first application offering to generate images from a sentence, and this, as of January 2021. However, this first version was similar to a prototype. The images that could be created had an “unfinished” look.
From April 2022, Dall.e 2 succeeded it and this time, the quality of the images offered was convincing. The choices of colors, decorations, shadows and lights have sometimes been worthy of artistic works.
It turns out that Dall.e 2 is a production of OpenAI, the creators of ChatGPT. So Dall.e 2 also relies on the GPT engine. The remarkable point is that OpenAI says it has succeeded in coding in its AI the characteristics that make an image seem pleasant to us to look at. This AI has thus been trained to compare the aesthetic quality – human judgment factor par excellence – of multitudes of images. An advantage of relying on GPT is that the application understands queries typed in French.
What struck users of Dall.e 2 was the ability to imagine totally surreal scenes and see the application respond to requests such as: “ A kangaroo with dreadlocks dressed in Louis XV style shaking hands with an cosmonaut. Frida Kahlo painting style ».
Dall.e 2 can also take an existing photograph and replace or complete the decor. By force of circumstance, the results are sometimes unexpected and disappointing, but it is usually enough to restart the generation of images to sooner or later achieve an honorable creation.
Stable Diffusion
Stable Diffusion quickly marked by its ability to generate more sophisticated images than those of Dall.e 2. The company that created this tool, Dreamstudio, has powerful servers and therefore the generation of images is rather fast. Above all, Stable Diffusion is impressive in terms of the visual richness of the images generated, with a quality worthy of what a graphic designer, draftsman or even a high-level photographer could have produced.
To generate an image, it is necessary to start from a prompt (description) which should ideally be very long. The interface can be intimidating at first when trying to influence the various options. The easiest way is to try various settings and see what comes out of them or to draw inspiration from creations already made by copying their prompt, which is possible from galleries such as Lexica.art. The result rarely corresponds to what we expected at first glance and therefore we quickly get into the habit of specifying “negative prompts”, basically, of also specifying what we do not want to see appear in the image: no skyscrapers, no mauve color, no out-of-frame character…
Stable Diffusion can also start from an existing image and propose variations while offering to adjust various parameters.
Several sites offer to make the large-scale use of Stable Diffusion more manageable, in particular dreamlike.art, invoke.ai, image.ai or stability.ai. Most require fairly quickly to buy credits but the price is very reasonable. Thus, from dreamlike.art, for 15 dollars, one can very quickly produce hundreds of images while indicating the desired output format. The ease of operation from such a site is very likely to seduce you.
Mid journey
Mid Journey is another application that produces images of undeniable quality from complex sentences. The criticism that some may make of it is that it is necessary to use it to go through the Discord server linked to this application and to intervene in one of the forums dedicated to “newbies”. From there, you can type your request and then have to wait while you see the generation of images requested by other users of the same service. It therefore happens that it is difficult to find his little ones. The alternative is to choose a paid version.
For each proposal, MidJourney offers 4 images that we see gradually being composed on the screen. We can dwell on one of them which will serve as a model for the generation of 4 new images. Once you have located a visual that you want to download, it is then possible to enlarge it. All this is not necessarily simple at first sight and it may be wise to consult a tutorial.
MidJourney also has the ability to produce the design of a given themed website and this service is of often astounding quality.
If we disregard the need to use this interface, which is not always practical at first glance, the fact remains that MidJourney is stunning in terms of the images it can produce, with a very stylized aesthetic bias. It’s not uncommon to see this app produce studio-worthy visuals almost every time.
Leonardo
If MidJourney has long been considered a sort of Rolls-Royce of text-to-imageIt seems like Leonardo be able to snatch this place from him. It is necessary, as with many applications, to register and wait for an invitation from the application. The usual wait time here is weeks or months due to Leonardo’s success and we had to push hard to get ours sooner than expected in these tests.
If we had to describe in a simple way what makes Leonardo’s strength, we could sum it up in one sentence: a quality similar or even superior to that of Midjourney, from an extremely simpler interface, which promotes greater flexibility. In fact, it is from a simple website that you type your request. What to say ? Leonardo produces visuals of exceptional quality with a correct probability of obtaining a rendering that corresponds well to what was requested.
Undoubtedly, when you browse the gallery of images produced with Leonardo, there is something to be amazed. Especially since the application also offers very useful aids in case you lack style inspiration. Moreover, Leonardo is also based on a model similar to Stable Diffusion and therefore, with prompts but also “negative prompts” to specify what you do not want.
Leonardo also has advanced tools, for example modifying only part of an image with enormous flexibility.
NightCafe
NightCafe offers an interface that facilitates the creation of images. Once the prompt is defined, we must first select a predefined algorithm. NightCafe relies on various engines, in particular Dall.e 2 or Stable Diffusion, but also other lesser-known tools such as Clip or VQGAN which allow very artistic creations. This tool also invites us to choose from various predefined styles: artistic, digital animation, manga… It is possible in passing to mix several styles. It is thus enough to be guided and the creation of an image can seem rather long. Some of the results we obtained were more than honorable and some really beautiful.
Among the original points of NightCafe is the fact that the application encourages the possibility of creating images by involving friends, which is able to “boost” creativity.
Maze Guru
The images from Maze Guru are of appreciable quality and creativity. A simplified version of this tool is accessible from the Maze Guru website. However, it would appear that one must first register on the Maze Guru Discord forum to access the web tool.
It is possible with Maze Guru to generate fictional fantasy visuals, but also a version of yourself as a cyborg, Pixar character or other. A large number of very aesthetic models are available and all you have to do is click on the visuals to choose them one by one. As much to know, if you choose the Disco model, the images can take a very long time to produce.
If you have the courage, you can operate from Maze Guru’s Discord server which works on the same principle as MidJourney: generation of 4 images, variations on one of them, enlargement… Let’s face it: all this It’s not easy to learn, although after a few minutes of practice you should be able to master this approach. And besides, we take a liking to seeing the creations of other Internet users being created before and after our own, because it results in a regular source of inspiration.
Deep Dream Generator
Deep Dream is a tool that aims to be simple: it is therefore based on a single page in which the criteria for the desired image are defined — it can be integrated into an existing photograph. If you click on “Edit”, Deep Dream offers various guides. For example, lists of artists (Picasso, Michel Ange…), effects, lighting, photographic style… Advantage: by default, this tool creates images in landscape mode and with good definition.
Deep AI
If your primary desire is to produce images that look like paintings, then Deep AI is able to meet this desire. Once the text of your choice has been introduced, you can choose a pictorial style, and DeepAI generates a suitable visual. If the result does not suit you, you can choose another type of paint. The options are limited to the choice of the shape of the image (square, rectangular…). However, this type of tool is well suited to those who do not want to get lost in technical details.
Broader art
Broader art is another tool that highlights its ease of use. It is therefore intended to be practical with an interface based on icons and models such as portrait, album cover, cartoon character with examples from which one can draw inspiration. Originality: it is possible to use this tool to create images of fictitious characters which will be composed of “parent” characters or even landscapes. Artbreeder also offers the creation of images from prompts.
Patterned
Looking for a design for your brand logo, for your website. Well, that’s exactly what Patterned produced from the indicated text. And the results are usually attractive. They are free of rights and usable at will, and some may even find it appropriate to resell them. Only 10 designs are offered for free, and this credit is quickly exhausted, after which, it is necessary to pay a small fee (10 dollars for 300 designs per month with the possibility of canceling at any time). Other tools of the same type exist and some such as Poly might also meet your needs.
