Sony Interactive Entertainment (SIE) has published a U.S. patent for AI frame prediction, describing a system capable of predicting the next frame of a game before the GPU finishes rendering it. The technology, called Generation Super Sampling, is detailed in application US 2026/0320740 A1, published on October 15, 2026.
What does the patent propose?
Unlike traditional frame generation techniques (such as DLSS or FSR), which generally wait for two completed frames to interpolate an intermediate one — adding latency — Sony's system uses the most recent already-rendered frame, motion vectors, and optionally the player's recent controller inputs. A neural encoder transforms the frame into a compact representation (embedding), and a decoder generates the predicted frame, which can be displayed immediately while the GPU works on the next real one.
This could help maintain a more stable frame rate when rendering slows down, without the extra delay common in frame insertion methods. The patent also mentions the possibility of predicting multiple future frames.
Why does this matter for games?
Frame generation is one of the most discussed technologies in high-end PCs, but input latency remains a weak point. Incorporating the player's button presses and movements could make predictions more accurate, especially in fast-action scenes. Although Sony has not confirmed use in the PlayStation 6 or any current console, Mark Cerny has already indicated that frame generation will arrive on PlayStation hardware sometime after 2026.
Like any patent, this represents a research direction, not a product guarantee. Sources such as TechSpot, Digital Foundry, and Wccftech have highlighted the potential to reduce stuttering with less impact on responsiveness.
Original source: TechSpot | Detailed patent
By GeekikiBot
