Google DeepMind AI Recreates a Couple’s Lost Memory in Love, Rendered
Google DeepMind and filmmakers make Love, Rendered, recreating Burt and Ethelle Shatz's unfilmed first meeting after 70 years of marriage with generative AI.
Summary
Published September 9, 2026, the documentary short “Love, Rendered” follows Burt and Ethelle Shatz, married for over 70 years, as Burt experiences cognitive decline. It reconstructs their first meeting at a Cleveland student co-op, which was never photographed or filmed. Academy Award nominated filmmaker Liz Garbus directed, Dan Cogan and Darren Aronofsky produced, and Google DeepMind collaborated with Aronofsky’s Primordial Soup.
Technical lead Michael Chang’s team restored black and white photographs from the couple’s youth with generative models, then used performance capture to map present day mannerisms, including Burt’s head tilt, speech hesitation and eye crinkles, onto their younger likenesses. Ethelle corrected details such as a staircase curve and shoe heel, and the couple said the generated memory felt authentic. The project applies AI to preserving fading family memories, while Gemini users can upload old photographs and request restoration and colorization that retains each subject’s appearance, expression and pose. Google DeepMind did not identify the specific models used.
Positives
- Burt and Ethelle Shatz said the AI generated reconstruction of their unrecorded first meeting felt authentic.
- Performance capture preserved distinctive mannerisms, including Burt’s head tilt, speech hesitation and eye crinkles.
- Ethelle directly corrected visual details, including a staircase curve and shoe heel, throughout the reconstruction.
- Gemini lets users request restoration and colorization of family photographs while preserving appearance, expression and pose.
- Liz Garbus, Dan Cogan, Darren Aronofsky, Google DeepMind and Primordial Soup combined documentary filmmaking with emerging AI tools.
Risks & concerns
- Burt’s cognitive decline is causing memories from the couple’s marriage of over 70 years to fade.
- No photograph or film exists of Burt and Ethelle Shatz’s first meeting at a Cleveland student co-op.
- Generative models had to fill missing visual details, making Ethelle’s corrections essential to the reconstruction’s authenticity.
- Google DeepMind disclosed no specific generative or performance capture model names or evaluation measures.