With the "Deus Ex Mach­ina? — Test­ing AI Tools"-series we want to show you dif­fer­ent tools, that aim to sim­pli­fy writ­ing, design and research by using Arti­fi­cial Intel­li­gence. More on the "Deus Ex Machina?"-series can be found here.

"Help­ing people become more cre­at­ive" is the motto of Ideogram.ai, one of many online image gen­er­a­tion tools that have recently emerged on the inter­net. But how does Ideogram.ai dif­fer from Adobe Fire­fly, Stable Dif­fu­sion, Mid­jour­ney, and oth­er pro­viders of sim­il­ar services?

Ideogram.ai is the first AI that can gen­er­ate images with text. The advant­age: instead of manu­ally over­lay­ing text on an exist­ing image, both the image and text are gen­er­ated in a single pro­cess. For each prompt entered into the search bar, you receive four images. One of these four gen­er­ated images is selec­ted as the cov­er photo and dis­played to all users on the Ideogram.ai homepage. Click­ing on one of these pho­tos allows you to view the altern­at­ive ver­sions as well.

The web­site allows users to down­load not only their own images but also those of oth­er users, as long as they are using the free ver­sion. In addi­tion to the free access, there is also a basic ver­sion, which costs $7 USD per month, a plus ver­sion for $16 USD per month, and a pro ver­sion for $48 USD per month . The sub­scrip­tions mainly dif­fer in the num­ber of prompts that can be used per day and the qual­ity of the down­loaded images, meas­ured in pixels.

Ideogram.ai offers vari­ous formats and styles for cre­at­ing images. The font can also be cus­tom­ised as desired. This allows even a short prompt to be refined to achieve res­ults that bet­ter match the desired styl­ist­ic dir­ec­tion. For those who like to exper­i­ment, there is also the "Lucky Style" fea­ture, which gen­er­ates a style randomly. 

To con­vert a text into writ­ten form instead of visu­al, it needs to be enclosed in quo­ta­tion marks. This can also be observed in the example prompts used for this article. 

Ideogram.ai was co-foun­ded by former employ­ees of Google Brain, UC Berke­ley, and the Uni­ver­sity of Toronto, Mohammad Nor­ouzi, Chit­wan Saharia, Wil­li­am Chan, and Jonath­an Ho. With a start­ing cap­it­al of $16.5 mil­lion raised through seed fund­ing, the tool was launched in August 2023. Ideogram.ai offers a wide range of artist­ic styles, from lifel­like cine­mat­ic images to graf­fiti and com­ic or anime styles. The res­ults can look like this: 

It must not be for­got­ten, how­ever, that the AI is still very young — and also improv­able. The premise of gen­er­at­ing images with text sounds prom­ising. Nev­er­the­less, it should be noted when using the applic­a­tion that the AI is still in its early stages and may make some errors. In par­tic­u­lar, the tool does not always handle long prompts appro­pri­ately, espe­cially when a lot of text or com­plic­ated words are used. In a self-exper­i­ment, some funny, but unfor­tu­nately also 'use­less' images were cre­ated. The res­ults of the remix func­tion, where new images can be cre­ated based on already cre­ated images, are par­tic­u­larly often unsatisfactory. 

Shortly after its ini­tial release, Ideogram.ai already had 1.1 mil­lion users, and by early Octo­ber 2023, the AI had cre­ated around 80 mil­lion images. Users some­times have to wait up to 30 seconds or longer for the desired images to be cre­ated, as the serv­er is often over­loaded. Frus­tra­tion with this is evid­ent in sev­er­al prompts used by users, which cre­ate images of pro­test­erts hold­ing signs such as "YOU NEED MORE SERVERS".

Since Feb­ru­ary 2024, Ideo­gram 1.0 has been avail­able free of charge to all users. A new fea­ture of this enhances ver­sion is the so-called Magic Prompt, which can be toggled on and off as desired. The Magic Prompt is essen­tially an exten­ded ver­sion of the user's own prompt, which is auto­mat­ic­ally added by the AI dur­ing the cre­ation of the 4‑image series. 

Magic Prompt: A play­ful and ador­able cat, stick­ing out its tongue with a mis­chiev­ous grin, is seen in the cen­ter of a vibrant and col­or­ful poster. The cat is sur­roun­ded by a vari­ety of objects, such as a rain­bow, a sun, and a smil­ing moon, with a cheer­ful and ener­get­ic atmo­sphere. The over­all design is eye-catch­ing and per­fect for a light­hearted and fun set­ting, poster | Mod­el: Ideo­gram 1.0

Ideogram.ai is par­tic­u­larly well-suited for graph­ic design­ers, social media con­tent cre­at­ors, or any­one look­ing to quickly cre­ate logos, posts, fly­ers, etc. The tool is easy to learn, and com­ing up with and typ­ing a prompt is not dif­fi­cult. Sut­dents and even school pupils can also bene­fit from it, as well las private indi­vidu­als who enjoy AI image gen­er­at­ors. How­ever, com­mer­cial use is pro­hib­ited. The abil­ity to choose from dif­fer­ent styles also allows for per­son­al­isa­tion of the res­ults. A prompt can thus take on dif­fer­ent forms:

Vari­ous col­ours and shapes are also no prob­lem for Ideogram.ai. While the select­able styles are avail­able with oth­er AI image gen­er­at­ors, the final designs dif­fer in type and design. Typ­ic­ally, the images do not depict faces or bod­ies dis­pro­por­tion­ately (i.e., provid­ing incor­rect image inform­a­tion, such as too many fin­gers on a hand). This raises the ques­tion of how far the remix func­tion can be pushed. For this self-exper­i­ment, the middle image of the orange cat 1 above, the so-called "par­ent" image (prompt: an orange cat say­ing "Hello", poster | Mod­el: Ideo­gram 0.2), is sent through the remix func­tion mul­tiple times. Unplanned, a black cat 2 joins the first with the remix. In the first remix, instead of an orange cat 1 and a black cat 2, an orange-black cat 3 appears. In the second remix, the style also changes and the image sig­ni­fic­antly loses qual­ity. Addi­tion­ally, anoth­er orange cat 5 grows out of the orange-black cat 4. In the third remix, the text changes from "Hello" to "Hello you." The res­ults can be seen here:

On the one hand, it is prob­lem­at­ic that the entered texts are often not dis­played cor­rectly. Anoth­er issue is that although there is cur­rently no copy­right on the images cre­ated by the AI, com­mer­cial use is still pro­hib­ited, as there may be copy­right on them in the future. This clearly reduces the use­ful­ness of the AI, espe­cially for graph­ic design­ers, who can there­fore use the tool more for inspir­a­tion rather than for cre­at­ing fin­ished products.

Can Ideogram.ai there­fore only func­tion as an idea gen­er­at­or or for
private enjoy­ment? Is this AI just a fun hobby, a nice dec­or­a­tion piece
for uni­ver­sity present­a­tions, or is there more to it?

On the oth­er hand, exper­i­ment­ing with sen­tences and words (the prompts) enhances one's cre­ativ­ity. Due to the addi­tion of text, prompts need to be rethought and adjus­ted com­pared to oth­er image gen­er­at­ors to achieve the desired out­come. The pre­vi­ously men­tioned remix func­tion also assists users who are not yet pro­fi­cient with craft­ing the right prompts. Addi­tion­ally, users can always look at oth­er users' prompts by brows­ing through the images on the homepage and view­ing the asso­ci­ated prompts. Until Feb­ru­ary 2024, the AI did not sug­gest example prompts itself. Thus, the (pass­ive) exchange with­in the com­munity has rather fostered the devel­op­ment of the AI. Now, the pre­vi­ously men­tioned Magic Prompts can assist users if desired. They make it easi­er to obtain cre­at­ive res­ults but also lim­it the users' own cre­ativ­ity. There­fore, it is import­ant that users can decide wheth­er to enable or dis­able the Magic Prompts.

It is worth men­tion­ing that the AI cur­rently only works in Eng­lish but can trans­fer text in oth­er lan­guages onto an image. The AI can repro­duce the words (cor­rectly or incor­rectly), but it evid­ently does not "under­stand" them.

Con­crete applic­a­tions in sci­ence com­mu­nic­a­tion are not about cre­at­ing graph­ics or example images, but about cre­at­ing images that can illus­trate con­cepts without claim­ing to rep­res­ent the truth. Ideogram.ai can provide vari­ous inter­pret­a­tions of a cross-sec­tion of the Earth:

Image	Prompt	Magic Prompt
Example 1	cross-sec­tion of the earth, scientifical	A cap­tiv­at­ing cross-sec­tion­al image of the Earth, depict­ing its vari­ous lay­ers, includ­ing the crust, mantle, out­er core, and inner core. The crust is shown with diverse land­forms, such as moun­tains, deserts, and forests. The mantle, with its dark­er hue, has con­vec­tion cur­rents that drive tec­ton­ic plate move­ment. The out­er core, com­posed of mol­ten iron and nick­el, is sur­roun­ded by the sol­id inner core. The Earth's core is an intense envir­on­ment, with extreme pres­sure and tem­per­at­ure conditions.
Example 2	Cross-sec­tion of the earth, cinematic	/
Example 3	Cross-sec­tion of the earth	A stun­ning cross-sec­tion of the Earth, reveal­ing its lay­ers in intric­ate detail. The crust, mantle, and core are clearly vis­ible, with the mantle being divided into the upper and lower mantle. The core con­sists of the dense inner core and the less dense out­er core, sur­roun­ded by the Earth's crust. The image is col­or-coded, with the crust in shades of green and brown, the mantle in shades of red and orange, and the core in shades of yel­low and gold. The Earth's mag­net­ic field lines are depic­ted in blue, spiral­ing around the mol­ten out­er core.

Espe­cially in this con­text, the new fea­ture of the Magic Prompt can be very help­ful if one does not yet have a clear idea of the end res­ult. The images are of high qual­ity, but some­times more fant­ast­ic­al and cre­at­ive than strictly sci­entif­ic. Depend­ing on the con­text and envir­on­ment in which one wants to use the graph­ics, they can be more or less useful.

Graph­ics such as a pie chart are not really feas­ible to cre­ate with Ideogram.ai. Using the prompt "pie chart, 65% like cats, 25% like dogs, 10% like fish" does gen­er­ate a pie chart, but not with the desired distribution.

Even after refor­mu­lat­ing the prompt to "pie chart with 65% cats, 25% dogs, 10% fish" and exper­i­ment­ing with and without the Magic Prompt, the desired res­ult is not obtained.

It seems that Ideogram.ai is not yet able to cre­ate dia­grams. Addi­tion­ally, it must be reit­er­ated that images from Ideogram.ai can­not be used com­mer­cially. There­fore, for example, they can­not be used by magazines engaged in sci­ence communication.

In sum­mary, Ideogram.ai stands out for its ver­sat­il­ity, sim­pli­city, and access­ib­il­ity. With some exper­i­ment­a­tion, users can achieve unique and beau­ti­ful image res­ults in the desired style. How­ever, using prompts inde­pend­ently (without Magic Prompts) requires prac­tice and patience, and users may not always achieve the desired res­ults. While Magic Prompts sim­pli­fy the pro­cess, they can also be mis­lead­ing and pro­duce res­ults that have little to do with the ori­gin­al inten­tion of the users. There­fore, the tool is par­tic­u­larly suit­able for those who are open to vari­ous out­comes and do not have a spe­cif­ic out­come in mind.

The cre­at­ors of Ideogram.ai are the first to take the ini­ti­at­ive to devel­op a text-to-image AI. This makes Ideogram.ai unique among image gen­er­at­ors (for now).