The filing, titled “Generation Super Sampling,” was published on October 15 as US 2026/0320740 A1 and lists Sony engineer Denis Dupeyron as the inventor. It describes an encoder neural network that compresses a rendered frame into a compact embedding, then a decoder network that uses that embedding to predict the next frame.
One Frame In, the Next Frame Out
Most current frame generation inserts a synthetic frame between two frames that have already been rendered. Sony’s design predicts forward from a single frame instead, and the generated image appears on screen while the next real frame is still being drawn. That ordering is what the patent leans on to reduce input delay, and it could help smooth out performance hitches in demanding scenes without sacrificing visual fidelity.
The filing also covers a few extras:
- Optional inputs such as motion vectors and recent controller data, which help the model anticipate where the player is heading
- Prediction of several frames at once
- A simpler non-AI fallback that shifts blocks of pixels based on motion vectors
- Self-supervised training on the video stream itself, comparing predictions to actual frames, so no human labeling is needed
What the patent does not include is any latency or image quality benchmark, so how well it works in a real game is an open question.
Where It Would Fit on PS6
Frame generation would be a logical follow-up to the PlayStation 5 Pro’s PSSR upscaler, extending Sony’s AI-assisted rendering pipeline. Leaks already point to the PS6 targeting 4K at 120 frames per second in most games. Multi-frame generation could complicate that on standard 120 Hz TVs, and some industry observers worry developers may lean on it in place of optimizing their games.
The filing also joins a growing pile of Sony AI patents, including bosses that learn from player reactions, consoles that repair themselves, AI help for players with physical limitations, and stream viewers controlling an in-game cursor. Like those, it shows where Sony is exploring, not what will ship.
Hashlytics Take
Predicting forward is a smarter answer to the latency problem than interpolation, but it trades one weakness for another. A system that guesses the next frame has to guess wrong sometimes, and those misses show up as visual artifacts exactly when the game changes fast: a hard camera cut, a sudden dodge, an explosion. Feeding in controller data is Sony’s attempt to narrow that gap, and the patent says nothing about how often the guess fails. So the number to watch if this reaches hardware is not the frame rate it hits, but how it looks and feels in the half second after the player does something unexpected.
Follow Hashlytics on Bluesky, Facebook, LinkedIn , Telegram and X to Get Instant Updates
