Testing AI-Tools: Designs AI

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.


Designs.ai is an AI-based design tool by the com­pany Inma­gine Lab that includes logo, video and graph­ics cre­ation. The pro­gram also includes the option of hav­ing a text gen­er­at­or cre­ate texts via an input prompt or of set­ting already writ­ten texts to music with an audio gen­er­at­or. Users can choose from over twenty lan­guages — includ­ing Ger­man­and Eng­lish — and sev­er­al speak­er options for each language.

The struc­ture and user design of the logo and design tool is remin­is­cent of the well-known design soft­ware Canva. For logos, design pro­pos­als are cre­ated on the basis of con­tent input. This option is miss­ing in the Design­maker, where tem­plates can only be selec­ted from exist­ing ones and then be edited. For addi­tion­al input, users can ask for sug­ges­tions on which col­ours and fonts go well together.

Bild 1: Screenshot der Startseite von Designs.ai nach Login
Image 1: Screen­shot of the Designs.ai homepage after login.

For par­tic­u­larly good res­ults, espe­cially for the copy­writer, prompts that are for­mu­lated in as much detail as pos­sible are a must-do. The more inform­a­tion the tool has about the tar­get group, the inten­ded use, as well as keywords that should appear in the text, the bet­ter the res­ult will be. For this reas­on, it is worth com­bin­ing the use of Designs.ai with offer­ings such as Per­plex­ity AI or the Bing Chat­bot, which can provide addi­tion­al inform­a­tion for prompts. With RATH from Kanar­ies Data (a data ana­lys­is and visu­al­iz­a­tion tool), data visu­al­iz­a­tions can also be cre­ated for inform­a­tion mater­i­als with Designs.ai.

Inma­gine Lab mar­kets Designs.ai spe­cific­ally to graph­ic design­ers and com­mu­nic­a­tion depart­ments. Com­pared to oth­er AI tools, Designs.ai is still largely unknown — at least if you com­pare the company's fol­low­er num­bers on social media with those of OpenAI, the com­pany from which Chat­G­PT ori­gin­ates, or the online design tool Canva.

The AI behind the application

Arti­fi­cial intel­li­gence gen­er­ates sev­er­al design vari­ants and sug­ges­tions on how they can be fur­ther pro­cessed from the inform­a­tion that users give to the respect­ive tool. Machine learn­ing is used in the Designs.ai tools to gen­er­ate texts based on a prompt (sim­il­ar to Chat­G­PT, but without the chat func­tion) and to set them to music.

Designs.ai is powered by "Inma­gine Brain", an AI product from Inma­gine Ltd. The soft­ware is avail­able to pur­chase for integ­ra­tion into your own web­sites and online offer­ings. The web­site gives no indic­a­tion of the data on the basis of which "Inma­gine Brain" was trained. It is merely poin­ted out that past search quer­ies of users are eval­u­ated in order to carry out future search quer­ies more pre­cisely and that Nat­ur­al Lan­guage Pro­cessing and Image Ana­lys­is are used in the eval­u­ation pro­cess — accord­ing to a note in the web­site sec­tion "Ser­vices". The Designs.ai FAQs explain that the design tools are based on a lan­guage mod­el that can also eval­u­ate images and videos for ele­ments they con­tain and assign rel­ev­ant text pas­sages in a script.

The rhetorical potential of the tool

Designs.ai prom­ises on its home page, that it is pos­sible to cre­ate logos, videos and more with­in two minutes. Can the pro­gramme keep this promise?

The answer: It depends. The high­er your own stand­ards are and the more com­plic­ated your require­ments, the longer it will take to pro­cess them. The text cre­ation tool reacts quickly to text prompts and adjust­ments to the type of text and adjust­ments to the level of com­plex­ity are pos­sible, but it often takes sev­er­al attempts before the prompt for the text gen­er­at­or is con­ver­ted into exactly the right text.

Image 2: Screen­shot from the copy­writer with prompt and AI-gen­er­ated text.

As you can see in the example, the text out­put by the gen­er­at­or is logic­ally struc­tured and con­sist­ent in terms of spelling and punc­tu­ation — even if the desired part on the chem­ic­al reac­tion was skipped by the AI. In addi­tion, the text is phrased very gen­er­ally (which was also the case with oth­er prompts) and does not go into depth in regard to con­tent. And just like it hap­pens with oth­er text gen­er­a­tion tools as well, the accur­acy of the con­tent is not guar­an­teed, which forces the user to check the con­tent reg­u­larly. In addi­tion, the text gen­er­at­or does not spe­cify the sources for its inform­a­tion and Inma­gine Ltd. has not yet dis­closed the train­ing data either.

On a pos­it­ive note, the adapt­a­tion to the tar­get group "sixth grade" works tone­wise and with the dif­fi­culty level of the text. How­ever, depend­ing on the prompt input, repe­ti­tions occur in the text — as it did in the example above with the "dome-shaped struc­ture" — so there is no get­ting around revis­ing the text. In addi­tion, the text gen­er­at­or can­not access inform­a­tion from the Inter­net. As a res­ult, more recent events and more com­plex inform­a­tion can­not be auto­mat­ic­ally imple­men­ted by the AI and must be included in the prompt. The lin­guist­ic tone with­in a text remains at a com­par­able level in each case. There is little room for per­son­al­iz­ing the texts or imple­ment­ing one's own writ­ing style; writers have to adapt this in a second, manu­al step. And all this is addi­tion­al to the fact that the text gen­er­at­or can­not be used in a chat-like man­ner, but a new prompt is needed each time.

Once you are sat­is­fied with the text, you can con­tin­ue to use it via the graph­ics cre­ation func­tion or the video cre­ation tool. When cre­at­ing a video, the text is entered first (for this exper­i­ment, the ice vol­cano text from the first example is the basis for the video), which can then be read out by an auto­mat­ic­ally gen­er­ated voice-over. After that, you select the func­tion the video is sup­posed to serve from a vari­ety of cat­egor­ies, e.g. edu­ca­tion, advert­ising, and oth­ers. For the ice vol­cano text, the gen­er­at­or sug­ges­ted the top­ics con­struc­tion industry or loco­motion as suit­able cat­egor­ies — neither of which fit the text, of course. After the selec­tion of lan­guage and voice, the pro­gram cre­ates a video pro­pos­al with already selec­ted sequences and back­ground music.

Image 3: Screen­shot from the Video­maker after gen­er­at­ing a first video pro­pos­al based on the text input.

The tool pro­ceeds in such a way that with each sen­tence a new clip is shown. As you can see from the clips shown in the lower sec­tion, the AI is guided by cer­tain keywords in the sen­tence and searches for match­ing clips — which, how­ever, do not always match the over­all con­text of the video. As you can see from the clips dis­played at the bot­tom of the screen, the tool focused on words like "Jupiter" or "water" for the first three clips, and on "pro­cess" for the fourth — which is how the pro­gram deemed the clip with the sticky notes to be suit­able for the top­ic "ice vol­ca­noes". So here, too, manu­al re-adjust­ments are necessary.

Bild 4: Screenshot aus dem Designmaker. Das Mock-Up der Vorlage wurde beibehalten, Text, Farben und Bild wurden manuell ausgetauscht
Bild 4: Screen­shot from the Design­maker. The mock-up of the tem­plate was kept, text, col­ours and image were replaced manually.

Adding a mood or even emo­tions to the products cre­ated with Designs.ai through images, music or cer­tain word­ing is a pro­cess that has to be con­trolled manu­ally to a large extent. Based on the design and video tem­plates, suit­able tem­plates can be found depend­ing on the occa­sion and addi­tion­al sug­ges­tions on how these can be expan­ded. Unlike the Video­maker, how­ever, you do not give the AI any text input when cre­at­ing designs before select­ing the tem­plates. For videos, there is the option of adding music and adjust­ing the font and col­or selec­tion as desired. For the ice vol­cano video, for example, a track entitled "Space" was added, remin­is­cent of the music from doc­u­ment­ar­ies made for plan­et­ari­ums. This was done to illus­trate the pur­pose of the video (to con­vey inform­a­tion in a simple and clear way) and the fas­cin­a­tion for the topic.

Usage in science communication

Designs.ai names "edu­cat­ors" as a poten­tial tar­get group: "Sup­ple­ment les­son plans with enga­ging slides, posters and videos to inspire your stu­dents" it reads on the homepage. Teach­ers are not the company's only tar­get group, however.

For sci­ence com­mu­nic­at­ors who are act­ive on social media, there is a wide range of social media tem­plates that can be used to cre­ate enga­ging visu­als. Oth­er tools can be used for cre­at­ing visu­ally appeal­ing know­ledge dis­tri­bu­tion pro­jects, for example by cre­at­ing logos, or to cre­ate a con­sist­ent look for campaigns.

Reviews of Designs.ai praise its ease of use and vari­ety of offer­ings. Through the con­nec­tion to the stock photo com­pany 123RF, users have a wide selec­tion of image and video mater­i­al that can be included in their own projects.

One cri­ti­cism is that the text-to-voice pro­cessing often still sounds unnat­ur­al and that Designs.ai is so far only sup­por­ted web-based and not as an app. In addi­tion, the web­site is partly buggy and slow in execut­ing the AI-sup­por­ted design and writ­ing pro­cesses. Since only the free ver­sion was tested here, the num­ber of prompts and tem­plates that could be used was limited.


"Cre­ate logos, videos, ban­ners, mockups with AI in just 2 minutes": Designs.ai wants to keep this prom­ise even for users with no pre­vi­ous exper­i­ence.  An advant­age of the tool are the step-by-step instruc­tions for the indi­vidu­al design pro­cesses. How­ever, this ini­tially res­ults in gen­er­ic-look­ing products that often fol­low the same visu­al for­mula. So, if you want a unique res­ult, you def­in­itely have to spend more than two minutes.

It would cer­tainly be desir­able to train the AI to pay more atten­tion to the entire con­text of a text, espe­cially when cre­at­ing videos, instead of just read­ing out indi­vidu­al sen­tences. That would shorten the cre­ation pro­cess enorm­ously. The option of includ­ing one's own voice-over in the Video­maker would also be a sens­ible exten­sion. This would make the res­ult­ing videos sound more per­son­al and bet­ter adap­ted to the indi­vidu­al needs of sci­ence com­mu­nic­at­ors. And of course, it would be par­tic­u­larly desir­able to dis­close the train­ing data for the AI and its origin.