Outside wall of the Esther Mahlangu HomesteadArt of Contemporary Africa
In February 2025, Google Arts & Culture with collaborators from Domestic Data Streamers, who've been researching synthetic memories, traveled to Dr. Mahlangu’s home in Mpumalanga to explore visualizing undocumented memories for cultural preservation.
A four-step process
STEP ONE
Gathering the Stories: Dr. Mahlangu and her family shared memories, which were recorded in detail.
STEP TWO
Identifying Key Moments: The team, led by curator Dr. Nontobeko Ntombela, identified four key narratives that were central to her origin story.
STEP THREE
Translating to Prompts: These oral stories were carefully crafted into descriptive prompts for Google's advanced image and video generation models.
STEP FOUR
Generating the Visuals: Using the prompts, the model generated a collection of "synthetic memories"—still images that would serve as the foundation for the videos.
Technology behind the memories
To bring the static images to life, the project utilized Veo, Google DeepMind’s most capable video generation model.
Jason Baldridge, a Research Scientist at Google DeepMind, explains the leap from a still image to a moving one: "We use those images as the initial frame. Then, with Veo, instead of just doing text-to-image, we do image-to-video." This final step transforms the generated still "photograph" into a living memory, adding motion, depth, and emotion to the scene.
In the following video, watch as Esther Mahlangu shares stories from her childhood, alongside a look at reconstructed memories brought to life with VEO.
Memories of home
You are all set!
Your first Culture Weekly will arrive this week.