[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"branding":3,"analytics":7,"article-split-detects-ai-edited-video-without-any-training":10,"sections":40},{"siteName":4,"siteTagline":5,"publisherName":4,"contactEmail":6},"The Revision","Tech news, decoded.","editor@therevision.news",{"gaMeasurementId":8,"adsenseClientId":9},"G-ZW2MV82GYR","ca-pub-8533917693782264",{"article":11},{"id":12,"slug":13,"title":14,"dek":15,"body_md":16,"tags_json":17,"published_at":18,"created_at":19,"updated_at":20,"status":21,"review_note":22,"review_notes":23,"image_url":22,"persona_id":22,"persona_name":22,"section":30,"tags":31,"sources":35,"feedback":39,"feedback_at":22,"cost_usd":39,"total_tokens":39},3774,"split-detects-ai-edited-video-without-any-training","SPLIT Detects AI-Edited Video Without Any Training","A new training-free detector uses spatial and temporal signal analysis to flag AI-generated or partially edited video at an extremely low false positive rate.","A research tool called SPLIT can identify AI-generated and partially edited video without ever being trained on labeled examples.\n\nDeveloped by academic researchers and released on arXiv, SPLIT — short for Spatial Patch-Level Incoherence and Temporal Roughness — works by analyzing patch tokens from a frozen vision encoder, requiring no fine-tuning or labeled data. It computes two signals: Two-step Temporal Roughness (TTR), which contrasts one-step and two-step feature variations to catch non-smooth patch trajectories, and Local Spatial Motion Incoherence (LSMI), which flags spatially inconsistent temporal changes via gradients of a feature-space motion field. The two signals are fused multiplicatively with gamma correction to sharpen the boundary between real and fake at strict decision thresholds. The researchers also argue that standard metrics like AUROC obscure real-world performance, so they propose a service-aligned protocol — Fake Recall at a fixed false positive rate — that better reflects what a live moderation system actually needs.\n\nThe real-world framing matters. Most detection benchmarks optimize for aggregate accuracy, which lets a model look good while still flagging one in twenty authentic videos as fake — a rate that would be operationally catastrophic for a platform. SPLIT targets a false positive rate of 0.1%, and across three benchmarks (FakeParts, GenVideo, and ViF-Bench) it outperforms both supervised and training-free baselines at that threshold. The researchers also report that SPLIT is robust to post-processing with negligible computational overhead, meaning it holds up against common evasion attempts like compression and blurring without meaningfully slowing down a pipeline.\n\nThe code is public on GitHub, which invites scrutiny — and it will need it. Training-free detectors carry a structural vulnerability: as generative models improve, the artifacts SPLIT hunts for may shrink or shift, and a system that requires no retraining also has no built-in mechanism to adapt. That is a feature today and a liability the moment the next video model ships.","[\"ai\",\"video-detection\",\"deepfakes\",\"computer-vision\"]","2026-07-07T04:00:00.000Z","2026-07-07T08:13:26.305Z","2026-07-07T08:13:29.134Z","published",null,[24],{"id":25,"reviewer":26,"round":27,"reason":28,"status":29},"editor-r1","editor",1,"The article describes Temporal Roughness as 'Two-step Temporal Roughness (TTR)' in the source but simplifies it without noting the two-step (one-step and two-step feature variation contrast) mechanism — a minor inaccuracy — but more critically, the article omits the source's claim that SPLIT is 'robust to post-processing with negligible overhead,' which is a concrete, verifiable spec that adds newsworthy value; the final paragraph's skepticism about generative models being optimized to fool SPLI","resolved","ai",[30,32,33,34],"video-detection","deepfakes","computer-vision",[36],{"name":37,"url":38},"arXiv cs.AI","https:\u002F\u002Farxiv.org\u002Fabs\u002F2607.02886",0,{"sections":41},[42,46,51,56,61,66,71,76,81,85,90,94,99,104],{"name":43,"slug":30,"count":44,"latest_published_at":45},"AI",2590,"2026-07-16T04:00:00.000Z",{"name":47,"slug":48,"count":49,"latest_published_at":50},"Security","security",294,"2026-07-15T19:59:48.000Z",{"name":52,"slug":53,"count":54,"latest_published_at":55},"Deals","deals",179,"2026-06-29T20:02:07.000Z",{"name":57,"slug":58,"count":59,"latest_published_at":60},"Policy","policy",158,"2026-07-16T00:02:48.000Z",{"name":62,"slug":63,"count":64,"latest_published_at":65},"Hardware","hardware",122,"2026-07-14T19:46:26.000Z",{"name":67,"slug":68,"count":69,"latest_published_at":70},"Consumer Tech","consumer-tech",93,"2026-07-13T13:20:48.000Z",{"name":72,"slug":73,"count":74,"latest_published_at":75},"Software","software",70,"2026-07-13T19:52:25.000Z",{"name":77,"slug":78,"count":79,"latest_published_at":80},"Science","science",66,"2026-07-10T10:29:37.000Z",{"name":82,"slug":83,"count":84,"latest_published_at":18},"Dev Tools","dev-tools",59,{"name":86,"slug":87,"count":88,"latest_published_at":89},"Gaming","gaming",41,"2026-07-09T04:00:00.000Z",{"name":91,"slug":92,"count":88,"latest_published_at":93},"Startups","startups","2026-06-29T20:55:50.000Z",{"name":95,"slug":96,"count":97,"latest_published_at":98},"General","general",29,"2026-07-10T22:28:58.000Z",{"name":100,"slug":101,"count":102,"latest_published_at":103},"Reviews","reviews",20,"2026-06-24T12:00:01.000Z",{"name":105,"slug":106,"count":107,"latest_published_at":108},"How-To","how-to",6,"2026-06-16T09:00:00.000Z"]