{"id":111702,"date":"2023-11-22T21:10:14","date_gmt":"2023-11-22T19:10:14","guid":{"rendered":"https:\/\/prohoster.info\/blog\/novosti-interneta\/predstavlena-sistema-sinteza-video-stable-video-diffusion"},"modified":"2023-11-22T21:10:14","modified_gmt":"2023-11-22T19:10:14","slug":"predstavlena-sistema-sinteza-video-stable-video-diffusion","status":"publish","type":"post","link":"https:\/\/prohoster.info\/sq\/blog\/novosti-interneta\/predstavlena-sistema-sinteza-video-stable-video-diffusion","title":{"rendered":"Sistema e sintez\u00ebs s\u00eb videos Stable Video Diffusion \u00ebsht\u00eb paraqitur","gt_translate_keys":[{"key":"rendered","format":"text"}]},"content":{"rendered":"
Kompania Stability AI publikoi modelin e m\u00ebsimit t\u00eb makinerive Stable Video Diffusion, q\u00eb mund\u00ebson gjenerimin e videove t\u00eb shkurtra n\u00eb baz\u00eb t\u00eb imazheve. Modeli zgjeron mund\u00ebsit\u00eb e projektit Stable Diffusion, i cili m\u00eb par\u00eb ishte i kufizuar n\u00eb sintez\u00ebn e imazheve statike. Kodi i mjeteve p\u00ebr trajnimin e rrjetit nervor dhe gjenerimin e imazheve \u00ebsht\u00eb shkruar n\u00eb gjuh\u00ebn Python duke p\u00ebrdorur kuadrot PyTorch dhe \u00ebsht\u00eb publikuar n\u00ebn licenc\u00ebn MIT. Modelet e trajnuara tashm\u00eb jan\u00eb t\u00eb hapura n\u00ebn licenc\u00ebn leht\u00ebsuese Creative ML OpenRAIL-M, q\u00eb lejon p\u00ebrdorimin p\u00ebr q\u00ebllime komerciale. <\/p>\n
Ka\u00ebrkuar p\u00ebr ngarkim jan\u00eb dy opsione t\u00eb modelit: SVD (Stable Video Diffusion) p\u00ebr t\u00eb gjeneruar 14 \u043a\u0430\u0434\u0440\u0430 me rezolut\u00eb 576×1024 mbi nj\u00eb imazh statik t\u00eb caktuar dhe SVD-XT p\u00ebr t\u00eb gjeneruar 25 \u043a\u0430\u0434ra. \u00cbsht\u00eb e mundur gjenerimi i videove pa l\u00ebvizje ose me nj\u00eb rrotullim shum\u00eb t\u00eb ngadalt\u00eb t\u00eb kamer\u00ebs, me nj\u00eb gjat\u00ebsi jo m\u00eb shum\u00eb se 4 sekonda. Kontrolli i drejtp\u00ebrdrejt\u00eb i modelit mbi baz\u00ebn e nj\u00eb p\u00ebrshkrimi tekstual n\u00eb gjuh\u00ebn natyrale nuk mb\u00ebshtetet akoma, por mund t\u00eb p\u00ebrgatitet fillimisht imazhi burimor duke p\u00ebrdorur modelin e vjet\u00ebr Stable Diffusion 2.1 dhe m\u00eb pas ta transformoni at\u00eb n\u00eb video duke p\u00ebrdorur modelin SVD. <\/p>\n
Cil\u00ebsia e videos p\u00ebr momentin nuk ofron realiz\u00ebm t\u00eb p\u00ebrsosur fotografik dhe nj\u00eb p\u00ebrshkrim t\u00eb garantuar t\u00eb fytyrave dhe njer\u00ebzve. N\u00eb performanc\u00eb, modeli i hapur i propozuar tejkalon alternativat pron\u00ebsore nga kompania Runway dhe Pika Labs. Modeli mund t\u00eb adaptohet leht\u00ebsisht p\u00ebr t\u00eb zgjidhur detyra t\u00eb ndryshme, p\u00ebr shembull, mund t\u00eb p\u00ebrdoret p\u00ebr formimin e figurave voluminoze.
\r\n \r\n
<\/div>\r\n <\/div><\/center> <\/p>\n
S\u00eb aty, mund t\u00eb p\u00ebrmendim publikimin e mjet\u00ebve t\u00eb m\u00ebsimit t\u00eb makinerive Video-LLaVA, q\u00eb mund\u00ebsojn\u00eb krijimin e nj\u00eb p\u00ebrfaq\u00ebsimi vizual unik t\u00eb objektit, formuar n\u00eb baz\u00eb t\u00eb p\u00ebrdorimit t\u00eb fotografive dhe videove t\u00eb objektit gjat\u00eb trajnimit. Sistemi mund t\u00eb p\u00ebrdoret, p\u00ebr shembull, p\u00ebr t\u00eb njohur pranin\u00eb e t\u00eb nj\u00ebjt\u00ebve objekte n\u00eb imazhe dhe video. Kodi \u00ebsht\u00eb shkruar n\u00eb Python dhe shp\u00ebrndahet n\u00ebn licenc\u00ebn Apache 2.0. \n Burimi: opennet.ru<\/a> <\/p>","protected":false,"gt_translate_keys":[{"key":"rendered","format":"html"}]},"excerpt":{"rendered":"
![\"Luaj](\"https:\/\/img.youtube.com\/vi\/G7mihAy691g\/hqdefault.jpg\")
\r\n