What if AI can illustrate your dreams?

The Dream Tapestry was created for the "Shape of Dreams" exhibition, showcasing works by Salvador Dalí and fellow surrealists. To foster a personal connection, we invited visitors to share their dreams, transforming their experience into an intimate encounter with the art. Our user-centered design focused on the target audience, aged 40-60, who might be new to generative AI. We aimed for a seamless, intuitive experience, starting with a kiosk guiding them to describe their dreams via a mobile session. An hourglass animation indicated submission time. After sharing, visitors received a link to download their dream, displayed on a large tapestry video wall. DALL-E then merged these dreams into a communal image, completing the experience with a final link and an invitation to view their contribution in the artwork.

Our journey began with a web application that generated AI-driven illustrations based on a curated artist list from the exhibition. Initially, AI-generated images were limited to low resolution (512px x 512px), so we used an AI upscaler and DALL-E's outpainting feature to enhance them to 1080p. I integrated this prototype into TouchDesigner via a webhook, enabling the generation of six unique QR codes per session for real-time display across four computers, safely storing visitor submissions in a dedicated database. We created two installations to refine the user experience: a quarter-scale simulation in San Francisco and the official installation in St. Petersburg, Florida. The official setup featured twelve 65" Sony OLED TVs, while the test used 27" 4K HP monitors, all connected to three Datapath FX4 video wall controllers managed by a Windows 11 computer with NVIDIA. Apple M1 Mac Minis operated the kiosks, programmed to turn on and off with museum hours to prevent screen burn-in.

Dalí and his contemporaries' artwork primarily reflects 20th-century Western European lives as depicted in artwork from that time, leading to AI outputs that often featured white men. This bias made it difficult for diverse visitors to connect with the depictions of dreams. To enhance representation, we included American artists to the dataset, like Jacob Lawrence who highlighted the African American experiences during the same era as Dalí. We also encouraged visitors to provide details about race and gender in their dream descriptions for more distinctive results. Recognizing the educational potential of generative AI, we collaborated with the museum to address these biases openly. Instead of hiding the dataset's limitations, we worked with museum docents to educate visitors on the cultural and historical biases in art and their impact on AI outputs. This approach sparked discussions about the intersection of art, history, and technology, fostering awareness of AI training data influences.

The Dream Tapestry project demonstrated the power of generative AI to create immersive experiences in a cultural space. It transformed an exhibition into a platform for visitors to engage with surrealist art and share their dreams. This project pushed my team and me to navigate complex technical challenges while emphasizing the importance of inclusive innovation. The result was an interactive experience that connected visitors to Dalí’s legacy and sparked meaningful discussions about technology, representation, and the future of art. Leading this project reinforced my commitment to creating impactful, accessible experiences that redefine interaction with art and technology.