{"slug": "qualcomm-patents-a-way-to-skip-repetitive-video-frames-before-ai-watches", "title": "Qualcomm Patents a Way to Skip Repetitive Video Frames Before AI Watches", "summary": "Qualcomm has filed a patent for a method that uses caption comparison to decide which video frames a large multimodal AI model processes, skipping redundant frames to cut token counts. The system generates a caption for the whole clip and one per frame, scores their similarity using perplexity, and drops frames whose captions match the clip caption; in a worked example, scores of 0.4, 0.4 and 2.4 for \"crouching,\" \"crouching\" and \"running\" let the running frame contribute six times more tokens than a crouching one. Qualcomm's filing targets phones, wearables, AR/VR headsets, vehicles and robots, where a 900-frame 30-second clip can otherwise produce millions of visual tokens the filing calls a performance bottleneck.", "body_md": "# Qualcomm Patents a Way to Skip Repetitive Video Frames Before AI Watches\n\n[Get the best of each week in your inbox, free →](#get-weekly)\n\nQualcomm's patent decides which video frames an AI should bother with by comparing two sentences: one describing the whole clip, one describing a single frame. If they say about the same thing, the frame gets skipped.\n\n## What Qualcomm's caption-comparing video filter does\n\nImagine you ask your phone's AI assistant, \"What happened in this 30-second clip?\" The clip has 900 frames, and most of them show a person crouching in nearly the same pose. A typical AI would chew through every one of those frames, which is slow and burns battery for very little new information.\n\n[Qualcomm](https://patentlyze.com/qualcomm/)'s filing describes a shortcut. First, the system writes a short **caption for the whole clip**, such as \"a man rises from a crouch.\" Then it writes a quick caption for individual frames. If a frame's caption says roughly the same thing as the clip's caption, the frame is *redundant* and gets left out.\n\nFrames that say something different, like \"a man is running,\" are kept, and the more different they are, the more of their detail the AI gets to see. The point is to feed the AI less, but more useful, material.\n\n## How two captions decide which frames get dropped\n\nThe patent covers a method for deciding how much of a video a large AI model actually needs to look at. Video AI models (often called large multimodal models, meaning they handle pictures and text together) break each image into **visual tokens**, small chunks of numbers that stand in for what the image shows. One frame can produce hundreds or thousands of tokens, so a video can reach millions, which the filing calls a performance bottleneck.\n\nHere is the flow the claims describe:\n\n- A video encoder turns the clip into one set of tokens, and a video-captioning model turns those into a text description of the whole clip.\n- An image encoder turns individual frames into their own tokens, and an image-captioning model writes a caption for each frame.\n- The system measures how similar each frame caption is to the clip caption. The filing uses **perplexity** , a score for how surprised a language model is by a piece of text.\n- If a frame's caption is too similar (the score does not pass a threshold), zero of its tokens are used. Otherwise some are kept, and the AI processes the clip tokens plus the chosen frame tokens.\n\nThe description gives a worked example: scores of 0.4, 0.4 and 2.4 for \"crouching,\" \"crouching\" and \"running.\" It says the running frame could contribute six times more tokens than a crouching one. It also notes you could check only every tenth or 30th frame.\n\n## Why this matters for AI video on phones and cars\n\nIf an AI can skip frames that add nothing, answering questions about your videos gets cheaper to run. That matters most on devices with limited power, and the filing lists phones, wearables, AR and VR headsets, vehicles and robots as places the method could run. Qualcomm makes chips for many of those, so efficiency tricks like this fit its business.\n\nThe filing also argues that comparing captions picks frames that matter for the *meaning* of a scene. Comparing raw image data, by contrast, can favor frames that merely look different. For questions like \"what did the person do?\", that choice is the useful one. Older approaches fixed the number of tokens in advance, while this one adjusts to each video.\n\nQualcomm's seventh entry in our [AI image and video](https://patentlyze.com/ai-image-video-patents/) coverage since May follows earlier applications on [filling in missing photo areas](https://patentlyze.com/patent/qualcomm-pinch-zoom-out-past-your-photo-edge/) and [fewer steps in AI video](https://patentlyze.com/patent/faster-ai-video-generation-device/).\n\nOn the ship-path question, this one sits close to the finish line. The claims describe software steps only: encoders, captioning models, a similarity score and a threshold. There is no new chip or sensor required, and the filing says it can run on a phone, a watch, a headset or a car.\n\nWhat has to exist first is a video AI model running on the device, plus two captioning models to do the comparing. That adds work of its own, since writing captions for frames costs computing time. The filing does not say how much that overhead eats into the savings, so the real-world gain is an open question.\n\nThe shortest route to a product is a feature inside a video-understanding app trimming frames before the main model runs. It reads as a practical efficiency fix rather than a showpiece, and that makes it the sort of idea that could get used without anyone noticing.\n\nGet our take in your Top Stories\n\nLiked 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 Qualcomm patent news breaks.\n\n[Add Patentlyze on Google](https://www.google.com/preferences/source?q=patentlyze.com)One click.\n\n### There are more where this came from\n\nWe read every patent application Big Tech publishes and send you the ones worth knowing. Plain English, free, every week.\n\n## The drawings\n\n10 drawing sheets from US 2026/0360166 A1 · click any drawing to enlarge\n\n## Show the other 2 sheets\n\n[Browse all the drawings 10 drawing sheets](#all-drawings)\n\n    Want this weekly breakdown for a company we don't cover?\n    [Patentlyze Pro →](https://patentlyze.com/pro/?src=post)\n\n**Source.** Full patent text and figures from the\n\n[official USPTO publication PDF](https://image-ppubs.uspto.gov/dirsearch-public/print/downloadPdf/20260360166).", "url": "https://wpnews.pro/news/qualcomm-patents-a-way-to-skip-repetitive-video-frames-before-ai-watches", "canonical_source": "https://patentlyze.com/patent/qualcomm-skipping-repeat-video-frames-ai/?utm_source=rss&utm_medium=referral&utm_campaign=all", "published_at": "2026-11-13 02:56:56+00:00", "updated_at": "2026-11-13 03:24:04.059294+00:00", "lang": "en", "topics": ["artificial-intelligence", "large-language-models", "computer-vision", "ai-chips", "ai-infrastructure"], "entities": ["Qualcomm", "Patentlyze"], "also_reported_by": [], "alternates": {"html": "https://wpnews.pro/news/qualcomm-patents-a-way-to-skip-repetitive-video-frames-before-ai-watches", "markdown": "https://wpnews.pro/news/qualcomm-patents-a-way-to-skip-repetitive-video-frames-before-ai-watches.md", "text": "https://wpnews.pro/news/qualcomm-patents-a-way-to-skip-repetitive-video-frames-before-ai-watches.txt", "jsonld": "https://wpnews.pro/news/qualcomm-patents-a-way-to-skip-repetitive-video-frames-before-ai-watches.jsonld"}}