[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"branding":3,"analytics":7,"article-orbitquant-cuts-ai-image-and-video-model-costs-without-retraining":10,"sections":34},{"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":24,"tags":25,"sources":29,"feedback":33,"feedback_at":22,"cost_usd":33,"total_tokens":33},3548,"orbitquant-cuts-ai-image-and-video-model-costs-without-retraining","OrbitQuant Cuts AI Image and Video Model Costs Without Retraining","A new quantization method shrinks diffusion transformer inference overhead without needing calibration data for each new model or modality.","Researchers have found a way to compress image and video AI models without the usual retraining tax.\n\nOrbitQuant is a post-training quantization method that compresses diffusion transformers - the model architecture behind tools like FLUX.1 and CogVideoX - without needing a fresh batch of calibration data every time a new checkpoint or modality appears. The trick is a mathematical rotation called a randomized permuted block-Hadamard transform, which reshapes activation values so they cluster predictably regardless of what the model is generating or at what timestep. That predictability lets a single codebook handle the full range of inputs, where prior methods had to refit for each scenario. The team tested it across four models spanning both image and video generation and reported state-of-the-art results at several low-bit settings.\n\nQuantization is one of the most practical levers the field has for cutting inference costs, and diffusion transformers are notoriously expensive because they run the same model dozens of times per output. A method that transfers across modalities without per-model tuning could matter a lot as video generation models grow larger and more compute-hungry. The push to W2A4 - two-bit weights, four-bit activations - is notable: that level of compression usually produces unusable output.\n\nThe catch is that \"usable\" is doing a lot of work in that last sentence - the paper does not define a specific quality threshold, and production teams will want to run their own evaluations before treating OrbitQuant as a drop-in solution.","[\"ai\",\"machine-learning\",\"diffusion-models\",\"quantization\"]","2026-07-03T04:00:00.000Z","2026-07-03T08:25:47.616Z","2026-07-03T08:25:50.581Z","published",null,[],"ai",[24,26,27,28],"machine-learning","diffusion-models","quantization",[30],{"name":31,"url":32},"arXiv cs.AI","https:\u002F\u002Farxiv.org\u002Fabs\u002F2607.02461",0,{"sections":35},[36,40,45,50,55,60,65,70,75,80,85,89,94,99],{"name":37,"slug":24,"count":38,"latest_published_at":39},"AI",2590,"2026-07-16T04:00:00.000Z",{"name":41,"slug":42,"count":43,"latest_published_at":44},"Security","security",294,"2026-07-15T19:59:48.000Z",{"name":46,"slug":47,"count":48,"latest_published_at":49},"Deals","deals",179,"2026-06-29T20:02:07.000Z",{"name":51,"slug":52,"count":53,"latest_published_at":54},"Policy","policy",158,"2026-07-16T00:02:48.000Z",{"name":56,"slug":57,"count":58,"latest_published_at":59},"Hardware","hardware",122,"2026-07-14T19:46:26.000Z",{"name":61,"slug":62,"count":63,"latest_published_at":64},"Consumer Tech","consumer-tech",93,"2026-07-13T13:20:48.000Z",{"name":66,"slug":67,"count":68,"latest_published_at":69},"Software","software",70,"2026-07-13T19:52:25.000Z",{"name":71,"slug":72,"count":73,"latest_published_at":74},"Science","science",66,"2026-07-10T10:29:37.000Z",{"name":76,"slug":77,"count":78,"latest_published_at":79},"Dev Tools","dev-tools",59,"2026-07-07T04:00:00.000Z",{"name":81,"slug":82,"count":83,"latest_published_at":84},"Gaming","gaming",41,"2026-07-09T04:00:00.000Z",{"name":86,"slug":87,"count":83,"latest_published_at":88},"Startups","startups","2026-06-29T20:55:50.000Z",{"name":90,"slug":91,"count":92,"latest_published_at":93},"General","general",29,"2026-07-10T22:28:58.000Z",{"name":95,"slug":96,"count":97,"latest_published_at":98},"Reviews","reviews",20,"2026-06-24T12:00:01.000Z",{"name":100,"slug":101,"count":102,"latest_published_at":103},"How-To","how-to",6,"2026-06-16T09:00:00.000Z"]