Skip to main content

On Saturday, February 10, 2024, the Library of Virginia hosted Envisioning Ancestors with AI, an image-generating workshop using historic records and free artificial intelligence tools to create imaginative depictions of people from the past. We’ve been spending a lot of time with a particularly interesting historic record known as “Free Negro Registers.” These register books contain detailed physical descriptions including skin complexion, height, age, and facial features of free Black and multiracial Virginians.

LVA is actively digitizing these “Free Negro Registers” (such as the Rockbridge County “Register of Free Negroes”, 1805-1831 shown to the right) as a part of the Virginia Untold project and we’ve been working with our volunteers to crowdsource their indexing. The relatively new access to text-to-image AI tools presented a unique opportunity to investigate what generative artificial intelligence means for archives.

What could the AI software produce with a 200-year-old description from our records?

What would it mean to have an image of a person of color who otherwise would never have had their photograph taken or portrait drawn?

Could we engage our patrons to investigate this process with us?

Duncan Henry is described as a “mulatto” man around 25 years old, 5’6” in height, having a large Roman nose and freckles with almost straight hair. How do you imagine Duncan Henry looked from this description? Hover over the image to see AI’s interpretation.

We decided to host an exploratory workshop that would include both instruction and hands-on experiences surrounding historic records and AI. We began by introducing different historic records from LVA’s collections that include physical descriptions of individuals. We focused primarily on documents found in the Virginia Untold collection and demonstrated how to use the Virginia Untold website to find resources.

After a short break, we jumped into the AI portion of the program. We offered an overview of the technology, discussed limitations and ethical concerns, and demonstrated three free AI text-to-image tools available to participants.

These images were produced by AI using the description of John Butler from the Fairfax County ``Register of Free Negroes``, 1822-1841. We can't know if any of these resemble the real John Butler, but such images may help us imagine him more completely.

Participants were encouraged to bring their own descriptions of ancestors, but we also provided descriptions from our historic collections. Nearly all participants came equipped with their own descriptions, fully prepared to create imaginative depictions of their unseen family members.

We quickly butted against the limitations of technology. One participant was trying to depict a darker skinned or Middle Eastern woman but the AI tools continually produced women who appeared lighter-skinned or Anglo-Saxon any time “Jewish” was included in the prompt, causing her frustration. Any traits to be included in the image must be explicitly stated as part of the text prompt; otherwise, the AI tools may introduce unwanted assumptions.

The participants also encountered difficulty producing images of people with mixed-race ancestry, such as a Black woman with blue eyes or someone with Black and indigenous ancestry. Some details, such as marks and scars, were recorded in the Free Registers but difficult to get the AI tools to produce. We also were not able to achieve the nuances of Black skin tones that are sometimes included in historic descriptions using these AI text-to-image tools.

Participant Spotlight

Imagining Jane

Viola Baskerville attended the Envisioning Ancestors workshop to produce a possible image of Jane Gentry Johnson, her great-grandmother, of whom no images survive. Jane’s mother was described as a Black and Native American woman with light-brown skin, long jet-black hair, and high cheekbones. She wore colorful beads in her hair and an ankle-length gingham dress, in a description from family sources. Viola used AI tools to combine this description of Jane’s mother with a photograph of Jane’s daughter to try to approximate what her great-grandmother could have looked like as a young woman. The text-to-image tools we used did not produce an image representing both Black and indigenous ancestry until her features were more fully described.

At the end of our workshop, we reflected with our participants:

“AI provides a more personal human dimension to a long-lost ancestor. Genealogy is more than just data; it’s telling a story, working towards a concept of a 3-D human being. AI creates that 3-D concept in your mind. It’s a more holistic picture, and a sense of being and belonging.” – Viola Baskerville

“The software still has its limitations, but I can see how this has terrific future potential for creating a realistic image of an ancestor that can put ‘skin on their bones.’”- Workshop Participant

Despite the limitations, this workshop provided a way for participants to connect more strongly with their ancestors while learning about historic resources and exploring new technology. Feedback from participants revealed that the deconstruction of the AI technology and the opportunity to learn more about the tools was the most popular aspect of the workshop. The recent explosion of AI image-generation tools means that we all need to become more savvy consumers; understanding technology is often the first step in digital literacy.

Sonya Coleman, Digital Engagement Coordinator and Lydia Neuroth, Virginia Untold Project Manager

Envisioning Ancestors with AI will be offered again on August 10 from noon-2 p.m. at the Library of Virginia as a hybrid event. At the time of this posting, in-person seats are still available. The event is free but registration is required.

Leave a Reply