The AI innovations are flying around you in 2024, with one tool rapidly succeeding another. To transform the world of event interaction, the PresentersWall team proudly introduces its latest generative AI application: Wordcloud-to-Image. This feature not only adds an innovative layer to audience interaction during events but also increases the fun factor. Why? You'll read everything about it in this article.
A New Bridge Between Words and Images
At the recent MAK event 'Let's talk about stress', an initiative aimed at the challenges students face today, colleague Jerre had the honour of sharing the stage with prominent speakers such as Erik Scherder and Tim Senders. The highlight for us? The launch of the new PresentersWall feature: Wordcloud-to-Image. An AI technique that transforms words submitted by the audience from a word cloud into an image. Here, the AI tool uses all the submitted words as input to automatically write a prompt for generating an image. This feature offers a unique view of the submitted content by providing a visual representation of their thoughts and ideas.
Visualising stress
What does a 'stress-free evening' mean for students? What do their biggest stressors look like? These questions were answered by the students and captured in a word cloud, after which our AI application transformed the audience's submitted words into powerful images. The result was astonishing: on the one hand, an image of serenity and relaxation, and on the other, a darker image illustrating the complexity of stressful student life.
The technique behind the innovation
To realise the AI tool, a number of steps had to be taken. In addition to integrating with the latest version of GPT-4 and Dall-e-3, it was necessary to ensure that the submitted answers are acceptable to the AI. Thus, swear words are automatically filtered. But how does the tool deal with creatively written swear words, or words with a spelling mistake? By double prompting, a method in which the AI itself writes the prompt for the action to be performed, we get a pragmatic solution to a complex problem. Additionally, the AI tool focuses on the most submitted answers so that the image becomes as concrete as possible.
Image Generation with Dall-e-3
The most recent image generation API from OpenAI, Dall-e-3, is used for generating the images. This is mainly because models of comparable quality do not have an available API and because Dall-e-3 requires little guidance to produce "beautiful" images. So, there's no need to worry about unwanted results.
Swear Word Filter with AI
Offensive content such as swear words is undesirable as they could potentially create error messages with Dall-e-3 itself. To solve this, an AI filter has been created: a GPT-4 prompt that checks all submitted words for "child-friendliness." If GPT-4 labels a word as 'unsuitable,' it is removed from the list and not sent to Dall-e-3.
Input into Dall-e-3
People in the audience input one or more words. To translate this into a prompt for Dall-e-3, all input from the audience is first collected. Then, all this input is combined into one prompt, with the most common terms being labeled as 'the most prominent.' Finally, each prompt is given a style of choice, and the AI tool is allowed to apply its magic.
Join the Innovation
PresentersWall invites everyone to experience this new Wordcloud-to-Image functionality. If you also want to experiment with this innovative feature or are curious about how AI can change the way we communicate, please contact us and discover the power of Wordcloud-to-image for yourself.