WISHING WELL is an interactive, AI-powered digital art installation developed for Take Over Festival 2024, Festspielhaus Baden-Baden. It explores the fundamental question: What can art look like when it’s shaped through real-time interaction with AI? And perhaps more intriguingly, who really holds the reins—artist, audience, or the algorithm itself?
In WISHING WELL, visitors step into the creative process through a playful, multi-layered AI system, that helps them to visualize what they whish to see. The experience begins at an input screen, where users are invited to choose how to complete a simple prompt: “Show me a picture of…” The system then presents four AI-generated options for completing the sentence. This approach leans into the core idea of Large Language Models (LLMs)—not just generating content, but enabling the user to make a choice, flipping the usual AI-creation narrative. By doing this, visitors actively shape the visual content of the artwork.
This curated selection of options not only puts creative power in the hands of the audience but also makes the process of generating prompts for the AI video generator quick and intuitive. In a festival setting, people often don’t want to spend too much time crafting complex prompts. So, we employed AI to simplify and streamline the experience, keeping it engaging and spontaneous.
Once the prompt is chosen, the rest unfolds: an AI video generator brings visual forms to life, inspired by the movement of the visitor’s silhouette in front of a motion-sensing camera. Meanwhile, a generative AI soundtrack completes the multi-sensory experience.
The Multi-Part AI Process
To achieve the seamless interplay of AI technologies, WISHING WELL relies on several components working in harmony:
- Prompt Selection with an LLM
When visitors face the input screen, an LLM generates four variations to complete the phrase “Show me a picture of…”. This design choice reflects a return to LLMs’ core use case—creating linguistic suggestions—but here, it transforms the visitor into a co-creator. Visitors simply choose an option, which adds an element of chance and intuition to the creative process. - Prompt Translation for Enhanced Imagery
Since the installation primarily operates in a German-speaking environment, we employ a secondary LLM to translate these selected prompts into English. This is done in real-time to harness the richer visual results that English language prompts yield in our chosen image generator, StreamDiffusion. - AI Video and Sound Generation
As the visitor’s silhouette is captured in front of a sensor camera, it dynamically shapes a visual representation generated by the AI. This real-time interaction results in abstract, evolving images that blend human input and algorithmic creativity. A responsive AI soundtrack complements the visuals, creating a holistic experience.
Reimagining Creative Roles
By combining these AI elements, WISHING WELL rethinks the role of artists and viewers, giving them shared ownership over the artwork’s creation. This project moves the focus from a finished piece to the creative process itself, offering a fresh perspective on digital art in an AI-driven world.
WISHING WELL was created by CYLVESTER, the artistic duo formed by Max Schweder and Dr. Tobias Hartmann, in collaboration with digital artist Alexander Rechberg (PlexNoir). CYLVESTER has been redefining the digital art scene for over a decade. WISHING WELL made its debut at the Take Over Festival at the Festspielhaus Baden-Baden.
Credits:
- Concept & Realization: CYLVESTER (Max Schweder & Dr. Tobias Hartmann) and Alexander Rechberg (PlexNoir)
- Festival Debut: Take Over Festival, Festspielhaus Baden-Baden
- Technologies: Large Language Models, AI-driven image generation (StreamDiffusion), AI-generated soundscapes
Other Showings:
- Kunsthalle Weishaupt, Ulm ,21.09.2024
- Orangerie Theater, Cologne, 11.10.2024
Here, an altered, more accessible, inclusive and also audioreactive version was shown at the first iteration of 4.0_audiovisuelle_installation in Cologne