Reconstructing Pelé’s “lost” goal

Going from the pitch to pixels

Recreating this goal required a combination of practical filmmaking and our most advanced AI models: Veo, Gemini Omni and Nano Banana Pro.

First, our crew shot live-action footage right on the grass of the Rua Javari stadium, using heavy leather balls and period-accurate uniforms. This physical foundation was then fed into our models to begin the digital transformation. We focused on three core technical experiments:

• Character replacement: Accurately mapping Pelé’s likeness and his classic number 10 kit onto a modern stunt player.
• Environment restyling: Transforming the modern stadium to match the cloudy weather and architecture of that specific day.
• Generating the ambiance: showing how fans watching the match at the stadium and listening to the radio broadcast at home experienced the moment.

Pelé’s original cleats from 1959. Archival photographs and artifacts served as the basis for all AI-generated scenes, ensuring historical accuracy across the film.

Raphael Herrera, photographer at Javari Stadium in 1959

Angelo Agarelli and Vicente Romano Netto, Juventus’ fans and eyewitnesses of the goal at Javari.

Crowds packing Javari Stadium during a Juventus match

Balancing photorealism with performance control

While generative models excel at photorealism, the extreme athletic choreography of a legend like Pelé presents a unique challenge. To solve this, we used Performance Control — an approach based on Veo 3 that extracts precise 3D geometry and motion from a modern stunt player to drive video generation. By pairing this with complementary workflows that use Nano Banana Pro and Gemini Omni, we generated a final video that seamlessly brings together the stadium architecture, field conditions, Pelé’s likeness and the dynamic play.

Starting with the original live-action video (top-left), we break the scene into separate, editable layers. We capture the athletes’ exact 3D motion (top-right), isolate them from the scenery (bottom-left), and generate a clean background without them (bottom-right). This allows us to modify the players and the environment independently.

To streamline editing, VFX, and video generation, Gemini Omni and Veo isolated actor footage, extracted the background and generated 3D blue-mesh player movement representations.

Performance Control creates editable 3D blue mesh renderings from an input video, modifiable using reference images.

Building a hybrid post-production pipeline

To achieve the final polish, we built a hybrid pipeline that combined AI generation with traditional visual effects (VFX). Using custom internal tools, we further refined the AI-generated shots with Gemini Omni and Nano Banana Pro, relying on archival imagery to ensure every detail was accurate. The workflow then moved to traditional VFX for tasks like ball compositing, grain integration, and rigorous color balancing. To ensure the generations looked as period accurate as possible, we ran the digital output through a filmout machine, capturing the distinct look and feel of 1950s cinema.

“The past remains a vast land, where many stories lie dormant, most of them not about famous goals or kings. What this reconstruction taught me is that they, too, might one day see a new light.”

By Anita Lucchesi

Brazilian Historian, UERJ & Arka

Museu Pelé, Santos SP, Brazil

On August 2, 1959, Pelé scored the most beautiful goal of his career: three consecutive “sombreros” without the ball touching the ground. Yet, the legendary ‘Gol da Rua Javari’ was never caught on film. We worked alongside historians, sports journalists, football legends, and Pelé’s family to reconstruct this lost moment and share it with the world.

This project was created in full partnership with the Pelé Brand, the official managers of the Pelé estate.

Rooted in education and cultural preservation, the project heads to the Pelé Museum this year. To achieve this, teams from Google DeepMind used models like Gemini Omni and Veo to transform historical fragments into moving imagery, filmed on the original pitch with authentic uniforms and a vintage ball.

By making the memory of this goal more accessible, it stands as a tribute to inspire a new generation of fans during the first World Cup without him.

The goal is on display at the Museu Pelé in Santos, Brazil.

Learn more at https://blog.google/innovation-and-ai…

No dia 2 de agosto de 1959, Pelé marcou o gol mais bonito da sua carreira: três chapéus consecutivos, sem deixar a bola tocar no chão. No entanto, o famoso “Gol da Rua Javari” nunca foi filmado. Em um trabalho em conjunto com historiadores, jornalistas esportivos, lendas do futebol e familiares do Pelé, reconstruímos esse momento antológico do futebol para o mundo todo ver.

Esse projeto foi criado em parceria com a marca Pelé, administradores oficiais do espólio do Rei Pelé.

Com foco em educação e preservação cultural, o projeto segue para o Museu Pelé este ano. Para conseguir isso, equipes do Google DeepMind usaram modelos de IA como Gemini Omni e Veo, para transformar fragmentos históricos em filmes, usando filmagens no campo original com uniformes e bola da época.

Tornar a memória desse gol acessível, é uma forma de homenagem, para inspirar as novas gerações de fãs de futebol, na primeira Copa do Mundo sem o Rei Pelé.

O “Gol da Rua Javari” está em exibição no Museu Pelé, em Santos.

Special thanks to:

Flávia Kurtz, Walter Dias, Pepe, Neymar Jr., Marta, Mauro Beting, Vicente Romano Netto, Angelo Agarelli, Anita Lucchesi, Paulo Monteiro, Santos F.C., C.A. Juventus, NY Cosmos, Museu Pelé, Museu do Futebol

Reconstructing Pelé’s “lost”

Reconstructing Pelé’s “lost”

Reconstructing Pelé’s “lost”

• Eyewitness Accounts: Researchers interviewed historians, sports journalists, members of the Mooca community, and fans who were at the original stadium.
• Archival Data: The team collated over 2,000 historical records, using scale models of the Rua Javari stadium, diagrams, and archival photographs to ensure exact spatial and temporal accuracy.
• AI Technology: Google utilized frontier models including Gemini Omni to interpret historical data and Veo to generate the video, relying on a technique that grounded the AI in historical facts rather than inventing new content.
• Live-Action Shoot: Live footage was shot on the pitch at the original stadium using period-accurate uniforms and heavy leather balls. [3, 4]

• Read about the process and watch the mini-documentary on the Google DeepMind Blog .
• Learn about the behind-the-scenes preservation efforts via Google’s Official Announcement. [2, 3, 4]

If you’d like, I can:Detail the exact movements and technical execution of the three sombreros.Provide more information on how the Veo AI model transforms historical memory into video.Let me know if you want to explore a specific part of this AI reconstruction!

The History of the World in 20 Minutes

. How Did Ancient Humans Travel the World?

. Nonprofits empower a nation of problem solvers | Episode 4

The Real Geological History of India

History of the birth of Rail Transport – The Evolution of Railways

. Mystery of Titanic: How The Unsinkable Ship Sink?