A Sony patent for frame generation was published on October 15. The proposed method attempts to predict subsequent frames based solely on a single freshly rendered real frame, displaying the predicted result ahead of time while the next real frame is still being rendered.
The patent, titled "Generation Super Sampling," bears application number US 2026/0320740 A1 and lists Sony engineer Denis Dupeyron as the inventor. An encoder neural network first converts the rendered frame into a compact representation, which a decoder then uses to predict the next frame. Motion vectors and recent controller inputs can also serve as auxiliary information to help the system anticipate the player's next moves.
The application argues that some existing techniques require obtaining two real frames before interpolating between them, thereby increasing wait times and latency. Sony's approach instead attempts to predict forward from a single real frame, aiming to alleviate stuttering in high-load scenarios without compromising visual quality, while also supporting the prediction of multiple frames at once. The document also proposes a simpler non-AI fallback path that moves pixel blocks using motion vectors.
The relevant model can be self-trained using video streams by comparing its predictions against subsequently generated real frames, eliminating the need for manual labeling. However, the application provides no latency or image quality test data. The patent is currently pending review and makes no mention of the PlayStation 6 or any specific product, so it remains unconfirmed whether this technology will be used in the next-generation console.