Sony has filed a patent for a proprietary frame generation technology, potentially intended for the future PlayStation 6. The document, titled Generation Super Sampling (US 2026/0320740 A1), was published on October 15, with engineer Denis Dupeyron listed as the inventor.
The proposed system utilizes an encoder neural network to convert rendered frames into compact embeddings, while a decoder network predicts subsequent frames. The model can incorporate motion vectors and controller input data to anticipate player movement, aiming to smooth out performance hitches in demanding scenes.
Unlike traditional methods that insert frames between two existing ones, Sony's approach predicts forward from a single frame to reduce latency. The technology is capable of predicting multiple frames simultaneously and includes a non-AI fallback that shifts pixel blocks based on motion vectors.
The neural network is designed to train on the video stream itself by comparing its predictions against actual rendered frames, eliminating the need for manual labeling. As the patent remains pending and lacks specific performance metrics, there is no confirmation that this feature will be implemented in future hardware.
