Reconstructing Pelé’s “lost” goal

Source: Google Date: 2026-07-14 URL: https://blog.google/innovation-and-ai/models-and-research/google-deepmind/reconstructing-peles-lost-goal/

Summary

Google DeepMind recreated a famous unfilmed 1959 Pelé goal using Veo (with Performance Control for 3D motion capture), Gemini Omni for scene isolation, and Nano Banana Pro for editing, combined with live-action filming at the original stadium and traditional VFX (character replacement, period-accurate restyling). Built with a historian’s archival research and Pelé’s family; the recreation now displays at the Pelé Museum in Santos, Brazil.

Implications

Low signal / noise for the tracked threads as a news item, but worth a note as a generative-video capability showcase: Veo’s Performance Control being used for motion-accurate reconstruction (not just prompted generation) is a more concrete capability demo than typical marketing pieces, relevant only tangentially to agentic engineering via Google’s pattern of using flagship creative demos to exercise the same model stack (Gemini, Veo, Nano Banana) that ships in its agent products.