I-Stability AI ishicilele imodeli yokufunda yomshini ebizwa ngokuthi I-Stable Video Diffusion engakwazi ukukhiqiza amavidiyo amafushane ezithombeni. Imodeli inweba amandla ephrojekthi ye-Stable Diffusion, ngaphambilini ekhawulelwe ekuhlanganiseni izithombe ezimile. Ikhodi yokuqeqeshwa kwenethiwekhi ye-neural kanye namathuluzi okukhiqiza izithombe ibhalwe nge-Python kusetshenziswa uhlaka lwe-PyTorch futhi ishicilelwe ngaphansi kwelayisensi ye-MIT. Amamodeli asevele aqeqeshiwe avuliwe ngaphansi kwelayisensi ye-Creative ML OpenRAIL-M evumelayo, evumela ukusetshenziswa kwezentengiso.
Kunezinketho ezimbili zamamodeli ezitholakalayo ukuze zilandwe: I-SVD (I-Stable Video Diffusion) yokukhiqiza amafreyimu angu-14 anokulungiswa okungu-576x1024 okusekelwe esithombeni esimile esinikeziwe kanye ne-SVD-XT yokukhiqiza ozimele abangu-25. Kungenzeka ukukhiqiza ividiyo ngaphandle kokunyakaza noma ngokuzungezisa ikhamera okunensa kakhulu, okuhlala isikhathi esingaphezu kwemizuzwana emi-4. Ukulawulwa kwemodeli eqondile okusekelwe encazelweni yombhalo wolimi lwemvelo akukakasekelwa, kodwa ungakwazi kuqala ukulungisa isithombe sangempela usebenzisa imodeli endala ye-Stable Diffusion 2.1 bese usiguqulela kuvidiyo usebenzisa imodeli ye-SVD.
Ikhwalithi yevidiyo ayinikezi i-photorealism efanelekile kanye nokunikezwa okulungile okuqinisekisiwe kobuso nabantu. Mayelana nokusebenza, imodeli evulekile ehlongozwayo ingaphambi kwama-analogue obunikazi avela ku-Runway kanye ne-Pika Labs. Imodeli ingashintshwa kalula ukuxazulula izinkinga ezihlukahlukene, isibonelo, ingasetshenziswa ukwenza izibalo ezintathu-ntathu.
Ukwengeza, singakwazi ukuqaphela ukushicilelwa kwekhithi yamathuluzi yokufunda yomshini we-Video-LLaVA, ekuvumela ukuthi udale ukumelwa okubonakalayo okubumbene kwento, okwakhiwa ngokusekelwe ekusetshenzisweni kwezithombe kanye nokuqoshwa kwevidiyo kwezinto ngesikhathi sokuqeqeshwa. Uhlelo lungasetshenziswa, isibonelo, ukubona ubukhona bezinto ezifanayo ezithombeni nakumavidiyo. Ikhodi ibhalwe ngePython futhi isatshalaliswa ngaphansi kwelayisensi ye-Apache 2.0.
Source: opennet.ru
