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.
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.
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.
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.
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.