Get the newsletter
Sony · AI / ML · Filed Dec 29, 2025 · USPTO published Jun 25, 2026 · Jun 25 release

Sony Patents AI That Sharpens Video Detail Without Being Fooled by Shifting Colors

AI video upscaling usually assumes that pixels change because something moved. Sony's new patent tackles a trickier problem: what happens when pixels change color for reasons that have nothing to do with movement, like flickering lights or shimmering water?

Sony Patent: AI Super-Resolution for Texture-Changing Video, figure from US 2026/0179183 A1
Figure from the official USPTO publication.
Publication number
US 2026/0179183 A1(US20260179183A1)
Applicant
Sony Group Corporation
Filed · Published
Dec 29, 2025 · Jun 25, 2026
Inventors
Kenichiro Yokota, Hisashi Kobiki, Shoichi Ikenoue
Status when we published
Waiting for an examiner
Document
21 claims
US classification
382/299
Parent application
Continuation of PCT/JP2024/022662 (filed Jun 21, 2024)(PCTJP2024022662)

What Sony's texture-aware video upscaling actually does

Imagine watching a video of a campfire or a neon sign reflected in wet pavement. The texture of those surfaces is constantly shifting color, even when nothing is moving. Most AI systems that sharpen or upscale video footage get confused by this, because they're trained to track movement and use it to fill in missing detail.

Sony's patent describes a system that specifically spots those "color change" pixels, the ones that are shifting due to texture effects rather than actual motion, and tells the AI to treat them differently. Instead of letting those unstable pixels corrupt its frame-to-frame memory, the system replaces them with a neutral placeholder before passing information forward.

The result is an upscaling process that can still make low-resolution video look sharper, even when the scene contains objects like fire, water, or fabric whose surface appearance is constantly in flux. It's a targeted fix for one of the more stubborn blind spots in current video AI.

How the model flags and neutralizes color-change pixels

The system works by processing a sequence of video frames through several stages. First, each low-resolution input frame is enlarged into an intermediate frame with more pixels. Those intermediate frames are then fed into a machine learning model that outputs a final high-resolution estimated frame.

The key innovation is in how the model carries memory between frames. It maintains cumulative feature information (a running summary of what it has seen across all previous frames) to help it reconstruct fine detail. But if a pixel is changing color because of texture effects rather than movement, including it in that running summary would pollute future frames with bad data.

To prevent this, the system identifies color change pixels by comparing texture information across frames and flagging pixels whose color shifts regardless of object motion. Before that frame's data gets folded into the cumulative memory, those flagged pixels are replaced with a fixed placeholder value.

The model is trained specifically on data that includes these replacements, so it learns to produce sharp estimated frames even when portions of the input are intentionally blanked out. The training pipeline mirrors the inference pipeline, which helps the model generalize to real-world footage with unpredictable surface textures.

What this means for AI upscaling in cameras and TVs

For Sony, which makes cameras, TVs, and image sensors, this kind of fix matters across multiple product lines. AI upscaling is already a selling point on high-end televisions, and it's increasingly used in professional video production. The gap this patent addresses is real: footage of fire, water, foliage, or any reflective surface is notoriously difficult for frame-interpolation and super-resolution systems.

For you as a viewer or videographer, the practical upside would be fewer artifacts and sharper detail in exactly the scenes that currently look the worst after AI processing. Whether this ends up in a TV firmware update, a camera processor, or a post-production software tool, it targets a concrete limitation rather than just a marginal quality improvement.

Source: US 2026/0179183 A1 at the USPTO (PDF). Full patent text and figures are in the official publication.

Get our take in your Top Stories

Liked this breakdown? Add Patentlyze as a preferred source on Google, and our plain-English take shows up more often in your Top Stories the next time Sony patent news breaks.

There are more where this came from

We read every patent application Big Tech publishes and send you the ones worth knowing. Plain English, free, every week.

More in AI image & video

All 223 AI image & video patents →