Reconstructing Pelé’s lost goal

Reconstructing Pelé’s lost goal

More

Descriptions:

Google DeepMind presents a short documentary on using AI to reconstruct Pelé’s self-described most beautiful goal — a header scored on August 2, 1959 against Juventus at the Rua Javari stadium in São Paulo — for which no video footage exists. The project combined historical research, eyewitness interviews, and multiple AI systems to produce a visual reconstruction of a moment that predates Brazilian television broadcast.

The technical pipeline involved two primary AI components. Gemini Omni processed archival photographs and historical imagery to generate period-accurate scenes of the Moaca neighborhood stadium, crowd attire, uniforms, and equipment — including leather balls that absorbed rainwater and became significantly heavier, a detail consistent with eyewitness accounts of wet conditions that day. A Performance Control Model handled markerless motion capture: input video of live actors is processed to produce 3D skeletal mesh renderings, which are then used to transfer motion onto digitally reconstructed players without requiring traditional capture suits.

The goal itself involved Pelé flicking the ball over four successive defenders — three sombreros plus the goalkeeper — without it touching the ground. The production brought together local Brazilian actors, a film crew, surviving eyewitnesses, and football historians to validate period accuracy. DeepMind frames the project as a demonstration of AI-assisted cultural preservation, specifically its potential to bring pre-broadcast historical moments to audiences who have never had the chance to see them.


📺 Source: Google DeepMind · Published July 14, 2026
🏷️ Format: Showcase

1 Item

Channels

1 Item

Companies