MCN 2026

Visiting an art museum often heavily relies on visual perception. For patrons who are blind or low vision, this dependence significantly impedes their experience. By conveying an artwork’s composition and content through a descriptive textual representation, museums can help close this gap. However, most museum collections are too vast to manually develop descriptions for each object, inspiring the question, “how can AI help streamline the writing process?”

Over the past two years, a collaborative, inter-museum working group of US-based art museums has been exploring using generative AI to provide image descriptions and alt- text for artwork. This session will explore how two of these museums, the Cleveland Museum of Art (CMA) and the National Gallery of Art (NGA), implemented comprehensive workflows to make visual descriptions available online for every primary image in their collections.

This session will dive into the technical, practical, ethical, and conceptual challenges inherent in developing generative AI pipelines in production and sharing this content with the public. We will share key decision points and processes put in place to instill confidence both internally and externally in the generated descriptions and maintain that confidence as new images become available and are described moving forward.

We will discuss in depth our respective approaches for developing infrastructure to develop, test, and integrate the descriptions with internal systems, and the strong collaboration with teams across each museum throughout the process. Emphasizing the importance of thoughtfully and responsibly sharing these descriptions, we will also talk through our final steps for publishing the descriptions for the public and continuing to gather feedback post-production.

Finally, each museum will share their next steps on image descriptions and their broader uses of AI that have developed out of this work, including broader cross-collection discovery efforts and accessible in-gallery experiences.