Résumé
CogVideo (2022) mooy xeetu bind-ci-wideo bi njëkka ubbeeku ci escale bu yaatu, ba noppi CogVideoX (2024) mooy xeetu xët wi gëna am kàttan bu bawoo ci Tsinghua/Zhipu AI. Dañu am solo ndax dañu def wideo yu baax yi ci loxo askan wi, du ci loxo liggéeyukaay yu mag yi kese.
CogVideo ak CogVideoX bokk nañu ci liggéey yiy wane ordinatër, yuy tekki wala defar ay meja yuñ mëna gis ngir jàngat, liggéey ak sos lu bees.
Plongeur bu xóot
CogVideo, bi ñu génne ci 2022, dafa tabax ci kaw CogView2 transformatëru bind-ci-nataal ba noppi jëfandikoo xeetu kadre yu bari, xeetu autoregressive ngir defar ay clip yu gàtt, nekk xeetu bind-ci-wideo bu mag biñ njëkka génne ci lu ubbeeku, di jàppale Chinois ak Angale. Ki koy wuutu ci 2024, CogVideoX, jëmmal bu mat sëkk la: dafay jëfandikoo autoencoder 3D ngir kompresse wideo ci espace ak temps, ginaaw ga Transformer Expert ak mébetu diffusion buy boole ay mbind ak ay token wideo yuñ boole. Modèle CogVideoX (ci dayo yu melni 2B ak 5B parametre) defar ay segond yu bari yu lëkkaloo, wideo yu am yëngu bu kawe ci dayo yu melni 720x480 ak jàppale nataal-ci-wideo ak wéyal wideo. Lu gëna am solo mooy, poid yi ak kode yi dañuy nekk ci mbooloo mi, di dundal vague de fine-tunes, jumtukaay ak gëstu ci askan wi.
Gis-gis xarala
VAE 3D bu CogVideoX dafay wàññi wideo bu ñor bi ci volume bu nëbbu bu dëgër, dagg limu token yi suko defee transformateur bi mëna modele ay toppalante yu gudd ci njëg yu yomb. Transformer Expert dafay jëfandikoo norm couche adaptive ba noppi boole mbind ak jetons visuel suko defee ñaari modalite yi di toppatoo seen biir directement, gëna baaxal njubluwaay mbind-wideo. Tàggat yaram buy yokk resolusioŋ yi ak guddaay yi, boole ci bind ay done yu jaar yoon, dafay jur yëngu bu gëna neex, gëna dëggu ci wàllu semantik.
CogVideo ak CogVideoX
CogVideo (2022) mooy xeetu bind-ci-wideo bi njëkka ubbeeku ci escale bu yaatu, ba noppi CogVideoX (2024) mooy xeetu xët wi gëna am kàttan bu bawoo ci Tsinghua/Zhipu AI. Dañu am solo ndax dañu def wideo yu baax yi ci loxo askan wi, du ci loxo liggéeyukaay yu mag yi kese. CogVideo ak CogVideoX bokk nañu ci liggéey yiy wane ordinatër, yuy tekki wala defar ay meja yuñ mëna gis ngir jàngat, liggéey ak sos lu bees. Ngir tabax xam-xam bu xóot, jàppal CogVideo ak CogVideoX ni xeetu liggéey, du benn man-man: leeral njariñ yi nga bëgg, leeral xalaat yi, ba noppi tàqale li sistem bi mëna def ci anam wu wóor ak li ba leegi soxla àtteb kàngam.
Ci jëf, ekip yu am doole yiy jëfandikoo CogVideo ak CogVideoX dañuy ekilibre njub ak dëggantaanu liggéey lu melni kalite done, faraasu leeral, ak méngoo etiketu. Dañuy bind kritër yu leer ngir am ndam, natt leen ci done yu dëggu ak def liggéey, ba noppi ñu baamtu ci anamu ñàkka mëna seetlu, du ci benn yoon benchmark wins. Mooy barab bi xam-xam theorie bi di soppiku nekk kàttan buy yàgg ci produit yi, ci politik yi ak ci liggéey yi.
Visual IA mën na otomatise saytu, gis ak etiketu liggéey ci eskaal. Ci jamano jooju, yelleefi nataal ak nangu mën na nekk risku yoon sudee fimu bawoo leerul. Xeetu jëf bi gëna dëgër mooy boole gaawaayu jàngat ak disipline nguur: doxal pilote, jàpp firnde, siiwal dogal yi, ak wéy di yeesal kaaraange gi ci anam wi ñuy doxalee, li jëfandikukat bi di xaar, ak sàrti sàrt yi di jëm kanam.
njeextalu pexe
Visual IA mën na otomatise saytu, gis ak etiketu liggéey ci eskaal.
Visual IA mën na otomatise saytu, gis ak etiketu liggéey ci eskaal. Ci jëfandikoo yu am kalite bu kawe, loolu dañu koy tekki ci sàrti liggéey yuñ mëna natt, ay peggu boroom, ak ay xew-xewu xoolaat yu bari suko defee ekip yi mëna yokk wóolu seen bopp ci barabu yokk lu jaxasoo.
Ekipu kreatif yi mën nañu defar konsept yu gëna gaaw te duñu def lu bari ci loxo.
Ekipu kreatif yi mën nañu defar konsept yu gëna gaaw te duñu def lu bari ci loxo. Ci jëfandikoo yu am kalite bu kawe, loolu dañu koy tekki ci sàrti liggéey yuñ mëna natt, ay peggu boroom, ak ay xew-xewu xoolaat yu bari suko defee ekip yi mëna yokk wóolu seen bopp ci barabu yokk lu jaxasoo.
Liggéeyukaay yi mën nañu jëfandikoo siñaal nataal wala wideo yu jafewoon lool ci liggéey.
Liggéeyukaay yi mën nañu jëfandikoo siñaal nataal wala wideo yu jafewoon lool ci liggéey. Ci jëfandikoo yu am kalite bu kawe, loolu dañu koy tekki ci sàrti liggéey yuñ mëna natt, ay peggu boroom, ak ay xew-xewu xoolaat yu bari suko defee ekip yi mëna yokk wóolu seen bopp ci barabu yokk lu jaxasoo.
Doxal ci àdduna dëgg
Defar ab clip narratif bu gàtt ci benn prompt ci làkku chinois wala anglais di jëfandikoo ay poid yu ubbeeku
Soppi benn nataal buñ yebbi mu nekk wideo buy toxu jaaraleko ci CogVideoX nataal-ci-wideo
Defar bu baax modelu ubbeeku bi ci stil wala personage buñ personaaliseer ngir animation indie
Gëstukat yi dañuy méngale pexe yu bees yi ñuy defaree wideo ak baseline buñ mëna defaraat
Modèlu jëfandikoo
CogWideo ak CogWideoX ci jëf
Defar ab clip narratif bu gàtt ci ab prompt chinois wala anglais di jëfandikoo ay poid yu ubbeeku bu baax.
Defar ab clip narratif bu gàtt ci benn Chinois wala Anglais buy jëfandikoo ay poid yu ubbeeku bu baax. Ekip yi dañuy faral di am njariñ yu gëna baax suñu joxee ay threshold yu baax ci kanam, tëye yoonu escalation nit ngir jafe-jafe yi, ba noppi topp njariñu produit yi ak njëgu njuumte yi ci diir bu gàtt.
CogWideo ak CogWideoX ci jëf
Soppi benn nataal buñ yebbi mu nekk wideo buy toxu jaaraleko ci CogVideoX nataal-ci-wideo.
Soppi benn nataal buñ yebbi ci wideo buy toxu jaaraleko ci CogVideoX nataal-ci-wideo Teams yi dañuy faral di am njariñ yu gëna baax suñu leeralee threshold yu baax ci kanam, tëye yoonu escalation nit ngir jafe-jafe yi, ba noppi topp njariñu produit ak njëgu njuumte ci diir bi.
CogWideo ak CogWideoX ci jëf
Defar bu baax modelu ubbeeku bi ci stil wala personage buñ personaalise bu baax ngir animaasioŋ indie.
Fine-tuning model ubbeeku ci stil wala jikko buñ jagleel ngir animation indie Teams yi dañuy faral di am njariñ yu gëna baax suñu joxee ay threshold yu baax ci kanam, tëye yoonu escalation nit ngir jafe-jafe yi, ba noppi topp njariñu produit yi ak njëgu njuumte yi ci diir bi.
CogWideo ak CogWideoX ci jëf
Gëstukat yi dañuy méngale pexe yu bees yu ñuy defaree wideo ak baseline buñ mëna defaraat.
Gëstukat yi benchmarking pexe wideo-generation yu bees ak benn baseline ubbeeku buñ mëna defaraat. Ekip yi dañuy faral di am njariñ yu gëna baax suñu joxee ay threshold yu baax ci kanam, tëye yoonu escalation nit ngir jafe-jafe yi, ba noppi topp njariñu produit ak njëgu njuumte ci diir bi.
Risk yi ak balustrade yi
Yelleefi nataal ak nangu mën na nekk risku yoon sudee fi ñu bawoo leerul.
Performance model bi mën na wuute ci leeraay bi, demographie bi ak environmaa bi.
Njuumteg positive yi mën nañu dem te kenn duko seetlu fileek xool wuñu buntu wóolu sa bopp.
Roadmap ngir samp gi
Mandargal kritërium nangug njub, woowaat ak njëgu njuumte.
Mandargal kritërium nangug njub, woowaat ak njëgu njuumte. Japp jéego bu nekk ni buntu firnde: sudee mattul kritër yi, noppali génne gi, tëj bërëb bi, ba noppi nga yaatal jëfandikoo gi.
Saytu ak done yu méngoo ak anam yi ñuy liggéeyee dëgg.
Saytu ak done yu méngoo ak anam yi ñuy liggéeyee dëgg. Japp jéego bu nekk ni buntu firnde: sudee mattul kritër yi, noppali génne gi, tëj bërëb bi, ba noppi nga yaatal jëfandikoo gi.
Yokk jàngat nit ngir xam fu wóorul dara wala am njeexital yu rëy.
Yokk jàngat nit ngir xam fu wóorul dara wala am njeexital yu rëy. Japp jéego bu nekk ni buntu firnde: sudee mattul kritër yi, noppali génne gi, tëj bërëb bi, ba noppi nga yaatal jëfandikoo gi.
Toppal model drift bi nga baaxal ko ginaaw bi kamera bi wala done yi soppeekoo.
Toppal model drift bi nga baaxal ko ginaaw bi kamera bi wala done yi soppeekoo. Japp jéego bu nekk ni buntu firnde: sudee mattul kritër yi, noppali génne gi, tëj bërëb bi, ba noppi nga yaatal jëfandikoo gi.