Sony has filed a patent application titled "Generation Super Sampling" that outlines its own approach to frame generation, with the PlayStation 6 appearing as the likely target despite no console being explicitly named. The filing, US 2026/0320740 A1, was published on October 15 and lists Sony engineer Denis Dupeyron as the inventor. It describes an encoder neural network that converts a newly rendered frame into a compact embedding, and a decoder network that predicts the next frame using that embedding. Optional inputs include motion vectors and recent controller inputs to anticipate player movement.
Unlike existing methods that insert a generated frame between two real ones—requiring both frames and adding delay—Sony's version predicts forward from a single frame. This aims to smooth out hitching in heavy scenes without lowering visual quality. The system can also predict multiple frames at once and includes a simpler non-AI fallback that shifts pixel blocks along motion vectors. The model trains on the video stream itself, comparing predictions to real frames, eliminating the need for human labeling.
However, the filing includes no latency or image quality numbers, and remains pending, so there is no guarantee the technology will ship. Frame generation would follow the PS5 Pro's PSSR upscaler, and leaks suggest the PS6 targets 4K at 120 FPS. Multi-frame generation is less relevant on 120 Hz TVs, and some worry developers might rely on it instead of optimizing games. Like Sony's other recent AI-related patents, this should be viewed as exploratory.
来源: TechPowerUp News · 由HeadlinesBriefing整理摘要