Google DeepMind helped a 92-year-old couple 'see' the day they met, 70 years later
A new short film uses generative AI to rebuild a memory that was never photographed. The result is moving, and it raises real questions about what counts as a real memory.
Key points
- "Love, Rendered" is a documentary short directed by Oscar-nominated filmmaker Liz Garbus, made with Google DeepMind and Darren Aronofsky's Primordial Soup.
- The film recreates the day Burt and Ethelle Shatz met in a Cleveland student co-op more than 70 years ago, a moment never captured on camera or film.
- Ethelle Shatz sat with the technical team as a co-creator, correcting details like the shape of a shoe heel and the curve of a staircase.
- Google's team used two AI systems: image restoration on old black-and-white photos, and performance-capture models that mapped the couple's present-day mannerisms onto their younger selves.
- Google says anyone can restore old family photos in the Gemini app by uploading an image and asking it to restore and colorize while preserving faces and poses.
Burt and Ethelle Shatz have been married for more than 70 years. They met as students at a co-op in Cleveland. No one took a picture that day. Now Burt is living with cognitive decline, and that first meeting is one of the memories slipping away.
A new short documentary, "Love, Rendered," tries to give it back to them. The film was directed by Liz Garbus, an Academy Award nominee, and produced with Dan Cogan and Darren Aronofsky. It was made in partnership with Google DeepMind, the AI research lab inside Google, and Primordial Soup, Aronofsky's production company. Google AI published the behind-the-scenes account.
How do you rebuild a memory that was never recorded?
Carefully, and with the person who remembers it sitting next to you. The filmmakers leaned on reminiscence therapy, a clinical practice that uses old songs, photos and stories to spark memories in people with dementia. The problem here was different: there were no photos of the moment itself.
So Ethelle became a co-director of sorts. She sat with Google's technical team and corrected the details as they appeared on screen. The curve of a staircase. The shape of a heel on a shoe. Small things that made the scene feel like the real one she still carries in her head.
What did the AI actually do?
Google's team used two kinds of generative AI, meaning software that creates new images or video from patterns it has learned. Each did a specific job.
| Step | What it did | Why it mattered |
|---|---|---|
| Image restoration | Cleaned and colorised old black-and-white photos of Burt and Ethelle as young adults | Gave the recreations an accurate visual anchor |
| Performance capture | Mapped Burt and Ethelle's present-day gestures, such as a head tilt or a pause in speech, onto their younger likenesses | Made the recreated versions feel like them, not like actors |
The technical lead on the project wrote that combining the two let the team weave the couple's past and present together into a single scene. Burt and Ethelle told the filmmakers the result felt authentic.
Is this a real memory, or something new?
Honestly, it is something new that stands in for the real one. The film does not claim otherwise. Aronofsky, quoted in Google's account, compares the AI to a paintbrush or a hammer: a tool that does nothing until a human guides it. The memory belongs to Ethelle. The images are a reconstruction shaped by her corrections.
That distinction matters, because generative video has a shakier public reputation right now for deepfakes, meaning fake clips of real people made to deceive. Here, the subjects are willing participants, sitting in the room, approving every frame.
Can you try this at home?
Yes, in a limited way. Google says you can upload an old family photo into the Gemini app, its consumer AI assistant, and ask it to restore and colorize the image while preserving the appearance, expression and pose of the people in it. That is a long way from a full documentary recreation, but it is the same underlying idea: use AI to make faded pictures legible again.
For families sitting with an ageing parent or grandparent, it is a practical, low-stakes place to start. Bring the photos. Ask the questions. Let the person in the picture do the remembering.
Common questions
Is "Love, Rendered" available to watch?
Yes. Google links to the full short film in its published piece about the project, and the film is credited to Liz Garbus, Dan Cogan and Darren Aronofsky in collaboration with Google DeepMind and Primordial Soup.
Does using Gemini to restore family photos share them with Google?
Uploads to the Gemini app are governed by Google's standard privacy policy for the product. If you are restoring sensitive family images, check the current data-use settings in the app before you upload.
