Yini enikezwa yi-MAI-Image-1 uma kuqhathaniswa ne-DALL·E, i-Midjourney kanye ne-Stable Diffusion?

I-MAI-Image-1 sekuyindlela entsha yokubheja yeMicrosoft ukubusa insimu yobuhlakani bokwenziwa obukhiqizwayo obusetshenziswa ezithombeni. Lo modeli, owakhiwe yinkampani ngokuphelele, ufuna ukunikeza okunye kwawo kunezinhlelo zezinkampani zangaphandle ebezilokhu zisetshenziswa kuze kube manje, ngokugxila okucacile kwikhwalithi yesithombe esingokoqobo, isivinini, kanye nokusetshenziswa kwangempela kulabo abakha okuqukethwe nsuku zonke.

Kunokuba kube isilingo esilula, I-MAI-Image-1 ifika ihlanganiswe ngokuphelele ohlelweni lwe-MicrosoftI-Bing, i-Bing Image Creator, kanye ne-Copilot kakade bathembele kule njini ukuguqula izincazelo zombhalo zibe izithombe ezinemininingwane. Ngaphezu kwalokho, imahhala kubasebenzisi abaningi emhlabeni jikelele, ngaphandle kokukodwa okubalulekile: i-European Union, lapho ukutholakala kwayo kuye kwahlehliswa khona ngenkathi inkampani ilungisa isevisi ngokwezidingo zomthetho.

Kuyini i-MAI-Image-1 futhi kungani ibaluleke kangaka ku-Microsoft?

I-MAI-Image-1 iyimodeli yokuqala yokukhiqiza izithombe eyenziwe ngaphakathi yi-Microsoft, eklanyelwe ngqo ukukhiqiza imiphumela yesithombe esingokoqobo kusukela ezicelweni zombhalo. Kuze kube manje, inkampani ibithembele ezixazululweni ezifana ne-OpenAI's DALL·E ukuze inike amandla amathuluzi ayo okubuka; noma kunjalo, imibiko ivele izinkinga zokudala izithombeFuthi ngalokhu kwethulwa, kuthatha isinyathelo esibalulekile sokuzimela okukhulu kwezobuchwepheshe.

Ngokusho kweMicrosoft uqobo, Imodeli iqeqeshwe kumasethi edatha akhethwe ngokucophelela Lezi zithombe zibuyekezwa ochwepheshe bokudala ukuze kugwenywe imiphumela ejwayelekile noma ephindaphindayo evame ukutholakala kwezinye izinjini. Umgomo ukunikeza izithombe ngezinhlobonhlobo ezibonakalayo, ukuguquguquka kwesitayela, kanye nenani elicacile elisebenzayo emikhakheni ehlukahlukene, kusukela kubaklami nabathengisi kuya kubadali bokuqukethwe kanye nama-ejensi.

Le nkampani ifingqa ifilosofi yalo msebenzi ngokuthi I-MAI-Image-1 iklanyelwe ukuletha "ukuguquguquka kwangempela, ukuhlukahluka kokubona kanye nenani elisebenzayo"Lokhu kusho ukuthi izithombe azibukeki kahle nje kuphela, kodwa futhi ziwusizo emikhankasweni, ezingcebweni zokuhlela, ezinkundleni zokuxhumana, ezethulweni zezinkampani, noma ezintweni zomkhiqizo lapho isici sezithombe sibalulekile khona.

Ngaphezu kwalokho, iMicrosoft yayifuna lokho Imodeli iphendula ngokushesha futhi ivumela ukuphindaphinda okusheshayo.Isivinini sokukhiqiza singenye yamandla ayo. Le nkampani ithi inhlanganisela yekhwalithi nokusebenza ivumela abasebenzisi ukuthi basuke emcabangweni wokuqala baye esithombeni esikhangayo ngesikhathi esifushane kakhulu, bese belungisa umsebenzi wabo nge amanye amathuluzi okudala afana ne-ComfyUI.

Ingasetshenziswa kuphi futhi kanjani i-MAI-Image-1

Enye yezinzuzo ezinkulu ze-MAI-Image-1 ukuthi itholakala mahhala. kubasebenzisi abaningi kakhulu. I-Microsoft ifake lo modeli ezinkundleni zayo eziningana ezibalulekile, ngakho-ke asikho isidingo sokufaka noma yini eyinkimbinkimbi noma ukuba nehadiwe ekhethekile ukuze uqale ukuyihlola.

Empeleni, Ungafinyelela i-MAI-Image-1 ngenjini yokusesha ye-Bing kanye nohlelo lokusebenza olusemthethweni lwe-Bing.kokubili kuzinguqulo zayo zedeskithophu nezeselula. Ngaphezu kwalokho, ihlanganiswe ne-Bing Image Creator, isigaba esizinikele ekukhiqizeni izithombe ezisebenzisa i-AI, esisebenza njengendawo yokungena elula kulabo abafuna nje ukubhala incazelo nokuthola imiphumela ebonakalayo engalandwa.

I-interface yomsebenzisi ilula kakhulu: Umsebenzisi ufaka isaziso esichaza indawo, into, noma isitayela asifunayoIsibonelo, “isithombe esingokoqobo sehlathi ekuseni kakhulu elinenkungu ethambile” noma “ipuleti le-pasta elinesosi katamatisi elibonakala phezulu, ukukhanya kwemvelo.” Uma incazelo icacile futhi inemininingwane eminingi, amathuba okuthola isithombe esifanelana nalokho obekucabanga ayanda.

Ukuze ufinyelele lezi zinketho, udinga i-akhawunti ye-Microsoft kuphela, ngakho-ke noma ubani osevele esebenzisa izinsizakalo ezifana ne-Outlook noma i-Xbox izinhlelo zokusebenza ku-Windows 11 Ingahlanganiswa kalula. Lokhu kuhlanganiswa ne-ecosystem ekhona kwenza kube lula ukusetshenziswa kunoma iyiphi idivayisi exhunyiwe futhi kwenza ukusetshenziswa kwayo kube ngokushesha ezigidini zabasebenzisi.

Ikhwalithi yesithombe esingokoqobo, isivinini, kanye namafomethi ahambisanayo

Isithembiso esiyinhloko se-MAI-Image-1 ukuletha izithombe ezibukeka sengathi zingokoqoboIsuka ezitayeleni "ezidonswe kakhulu" noma ezikhiqizwe ngokusobala yi-AI, iMicrosoft igcizelela ukuthi le modeli yaklanywa ngokuqondile ukuze ibalekele izinto ezijwayelekile, igxile ezigcawini ezikhanyayo nezikhanyayo ezinezimo ezikholisayo.

Ezivivinyweni zangaphakathi kanye nokuhlolwa komphakathi, I-MAI-Image-1 ibonise ukusebenza okuncintisanayo uma kuqhathaniswa nezinye izinhlobo zokubhekiselaLe nkampani ithi uhlelo luphakathi kwamamodeli ayishumi aphezulu e-AI okuguqulwa kombhalo ube isithombe ku-LMarena, ipulatifomu yokubambisana eqhathanisa amamodeli ngokuvota kontanga okungaboni. Nakuba iMicrosoft inganikanga izibalo eziqondile noma ishicilele izilinganiso eziphelele, iqokomisa lokhu kubekwa njengophawu lokusebenza kwayo okuqinile.

Esinye isici esibalulekile isivinini sokuphendula. Ngokusho kwethimba lokuthuthukisa, I-MAI-Image-1 ingacubungula izicelo futhi ibuyise imiphumela ngokushesha kunamanye amamodeli amakhuluokuvame ukuba nzima futhi kuhambe kancane ukukhiqiza.

Ngokuphathelene nezici zobuchwepheshe zemiphumela, Izithombe ezikhiqizwe zingalandwa ngesisombululo esiphezulu samaphikseli angu-1.248 x 832Lesi yisisombululo esenzelwe ukusetshenziswa kwedijithali okuvamile: okuthunyelwe ezinkundleni zokuxhumana, izihloko zewebhu, izinto zokwethula, noma ama-prototypes okudala angaphinde asetshenziswe ngamanye amathuluzi.

Ngaphandle kwalokho, I-MAI-Image-1 isekela amafomethi ahlukahlukene esilinganiso sobukhulunjenge-1:1, 3:2, kanye ne-2:3, ezihambisana nalezo ezisetshenziswa amanye amamodeli athuthukile njenge-GPT-4o ngesilinganiso sesici esibonakalayo. Lokhu kwenza kube lula ukuhlanganiswa kwezithombe ezikhiqizwe emisebenzini ekhona, lapho lezi zinhlobo zezilinganiso zisetshenziswa njengezijwayelekile kuma-banner, ama-cover, izikhangiso, noma izithonjana.

Izici ezithuthukisiwe kanye nokusetshenziswa okuhlangene nomsindo nezindaba

Ngale kwesizukulwane sakudala "sombhalo-kuya-esithombeni", I-Microsoft izama ukusebenzisa i-MAI-Image-1 ngendlela ethuthukisiwe kakhulu kuxhunywe kwezinye izinhlobo zokuqukethwe. Enye yezindawo lapho kuthuthuka khona okuthakazelisayo ukuhlanganiswa komsindo nesithombe ngaphakathi kwe-Copilot namathuluzi ayo ahambisanayo.

Ngokuqondile, Nge-Copilot Audio Expressions, ukudalwa kwezithombe ezivela kokuqukethwe komsindo kuyahlolwa.Ukuhlola ukuhlaziywa kokuqhathanisa kwe I-AI YezwiLokhu kusho ukuthi uhlelo lungahlaziya ifayela lomsindo, luchaze okuqukethwe kwalo okulandisayo noma okungokomzwelo, bese lukhiqiza isithombe esifanelana nendaba elandiswayo noma ithoni yomlayezo. Kungumqondo othakazelisayo kakhulu kuma-podcast, izindaba zomsindo, izinto zemfundo, noma okuqukethwe kwe-multimedia okusebenzisanayo.

Ngaphakathi kwemodi ye-Story ebizwa ngokuthi i-Copilot Labs, I-MAI-Image-1 ingakhiqiza izithombe ezenziwe ngokwezifiso ezihambisana nendabaIsibonelo, uma ukuqoshwa komsindo kuchaza uhambo lwasezintabeni, imodeli ingadala umfanekiso ohambisana naleso simo. Umgomo we-Microsoft ngalezi zici ukuqinisa ukuhlanganiswa phakathi kwamafomethi ahlukene nokwenza i-AI ekhiqizayo ibe umthombo ohlanganisa umsindo, umbhalo, nezithombe.

Nakuba lezi zinketho zisesesigabeni sokuhlola, Zibonisa ukuzibophezela kweMicrosoft ekuthatheni i-MAI-Image-1 ngale kwesizukulwane esilula esihlukileUmqondo uwukuthi imodeli izoba yingxenye yemisebenzi yokudala ebanzi, lapho ingaphelelisa khona imisebenzi efana nokubhala iskripthi, izwi, ukuhlela ividiyo, noma ukuklama izinto ezisebenzisanayo.

Ngesikhathi esifanayo, iMicrosoft iyaqhubeka nokuthuthukisa ulwazi ezimweni zokusetshenziswa zendabuko, njengokudala imifanekiso yezindatshana, amabhanela omkhankaso, amaphrothokholi omkhiqizo, noma imibono emfushane ebonakalayo yokwethula. Kuzo zonke lezi zimo, ikhono lokukhiqiza iziphakamiso eziningi ngemizuzwana nokugcina isitayela esifanayo Kubaluleke kakhulu kumaqembu adinga ukuphindaphinda nokuhlola imibono eminingi ngesikhathi esifushane.

Ukutholakala komhlaba wonke kanye nokungafani kwe-European Union

Ngokuphathelene nokuthunyelwa kwendawo, I-MAI-Image-1 manje isiyatholakala kabanzi kubasebenzisi emhlabeni jikeleleLokhu kusebenza kokubili ku-Bing kanye ne-Bing Image Creator, kanye nokunye okuhlangenwe nakho okuhlobene ne-Copilot. Kodwa-ke, kukhona isixwayiso esibalulekile: i-European Union, okwamanje, ihlukile kakhulu kulo mkhuba.

U-Mustafa Suleyman wachaza lokho esidlangalaleni Isevisi ayikakavunyelwa e-EU Ukufika kwayo kuzofika kamuva, uma iMicrosoft isiqedile ukulungisa okudingekayo ukuze ihambisane nemithethonqubo kanye nezimfuneko zamanje. Azikho izinsuku ezithile ezinikeziwe, kodwa kugcizelelwe ukuthi ukwethulwa kweYurophu kuhlelwe "maduze."

Lo mehluko ekutholakaleni ubonisa ukwanda kobunzima bokulawula obuzungeze ubuhlakani bokwenziwa, ikakhulukazi maqondana ukuvikelwa kwedatha, ukucaca, ilungelo lobunikazi kanye nokusetshenziswa kabi okungenzeka yamamodeli akhiqizayo. I-Microsoft ikhetha ukuthatha isikhathi esengeziwe ukuze ivumelanise le nsizakalo nalesi simo ngaphambi kokuyivula ngokuphelele emazweni angamalungu.

Nokho, kwezinye izifunda, I-MAI-Image-1 manje ingazanywa ngaphandle kwezindleko eziqondile kusukela kumapulatifomu enkampani, okubonisa ithuba labasebenzisi ngabanye, amabhizinisi amancane kanye nezinhlangano ezinkulu ezifuna ukuzama ukukhiqiza izithombe ngaphandle kokutshala imali ezixazululweni ezikhokhelwayo kusukela ekuqaleni.

Okwamanje, eYurophu, kulindeleke ukuthi, uma izidingo zomthetho sezihlangatsheziwe, Ithuluzi lizofika namakhono afanayo nalawo asevele ebonakala kwamanye amamakethe., okuhlanganisa ukuhlanganiswa ne-Bing, uhlelo lokusebenza lweselula, kanye nezici ezixhunywe ku-Copilot kanye ne-Copilot Labs.

I-MAI-Image-1 ngokumelene ne-DALL·E, Uhambo Oluphakathi kanye Nokusabalalisa Okuzinzile

Ngokungafani namamodeli agxile kakhulu esitayeleni sobuciko noma ekuhlolweni, i-MAI-Image-1 ivelele ngekhono layo lokukhiqiza izithombe ezicacile neziqondakalayo ezinokwethembeka okuphezulu kulokho okukhulunywa ngakhoLokhu kwenza kube ithuluzi eliguquguqukayo kokubili kubasebenzisi abajwayelekile kanye nabadali abangochwepheshe.

• Uma kuqhathaniswa DALL·EI-MAI-Image-1 ivame ukunikeza ukufana okukhulu kwemininingwane kanye nokuthambekela okuncane ekuphambukeniikakhulukazi ezintweni eziyinkimbinkimbi njengezandla, ukwakheka komzimba womuntu, noma umbhalo ofakiwe.
• Ukubhekana Uhambo oluphakathiUmehluko ubonakala kakhulu. I-Midjourney yaziwa ngobuhle bayo bobuciko, ukuthungwa okunemininingwane eminingi, kanye nekhono lokukhiqiza izithombe ezikhangayo, yize ivame ukwethula izakhi zesitayela ezingacelwanga. I-MAI-Image-1, ngakolunye uhlangothi, ibeka phambili ukucaca, ukuba ngokwemvelo kanye nokugcwaliseka okunembile kwesicelo.
• Uma kuqhathaniswa Ukusabalala OkuzinzileI-MAI-Image-1 inikeza ulwazi olulawulwa kakhulu futhi ayincikile kakhulu ekucushweni kobuchwepheshe. I-Stable Diffusion ivelele ngesimo sayo esivulekile kanye nekhono elikhulu lokwenza ngokwezifiso ngamamodeli, ama-LoRA, noma izindawo zokuhlola ezikhethekile, kodwa idinga ulwazi olujulile ukuze kutholakale imiphumela emihle kakhulu. I-MAI-Image-1 iletha Imiphumela eqinile ngaphandle kokulungiswa okuyinkimbinkimbiisebenza njengesixazululo "esilungele ukusetshenziswa".

Sekukonke, i-MAI-Image-1 izibeka njengesibonelo okulinganiselayo, okunembile nokufinyeleleka kalulaIlungele labo abafuna ikhwalithi yobungcweti ngaphandle kokudela ukulawula ukulandisa kwesicelo. Ngenkathi i-DALL·E ikhanya ngomcabango, i-Midjourney ngobuhle, kanye ne-Stable Diffusion ngokuguquguquka, i-MAI-Image-1 ivelele ngekhwalithi yayo. ukwethembeka nokungaguquguquki, izici ezimbili ezibalulekile ekusetshenzisweni okungokoqobo nokobungcweti.

Umongo webhizinisi kanye nokutshalwa kwezimali okukhulu kwingqalasizinda ye-AI

Ngenkathi iqinisa ikhathalogi yayo yamamodeli, I-Microsoft iphinde yabona inani layo lemakethe yamasheya likhuphuka kakhulu, liqhutshwa ukutshalwa kwezimali kwayo kubuhlakani bokwenziwa. kanye nokukhula kwe-Azure, ipulatifomu yayo yamafu. Le nkampani idlule amaRandi ayizigidigidi ezingu-4 ekutshalweni kwezimali emakethe okokuqala, isekelwa ukwanda okungu-18% kwemali engenayo kanye nezinhlelo zokutshalwa kwezimali ezinkulu zengqalasizinda.

Ngalo mqondo, Le nkampani ihlela ukwabela ingqalasizinda imali engaphezu kwezigidigidi eziyi-120.000 zamaRandi. okuhlobene ne-cloud computing kanye ne-AI eminyakeni ezayo. Lokhu kufakwa kwenzelwe ukusekela amamodeli e-OpenAI ahlala ehlanganiswe nezinsizakalo zawo kanye nezinhlelo ezintsha ezizimele, okuhlanganisa umndeni wakwaMaia kanye namamodeli akhethekile njenge-MAI-Image-1.

Ngokwabo, I-OpenAI iqinisa nokuzimela kwayoLe nkampani iqalise izinhlelo ezifana neProject Stargate, ezibandakanya abadlali abakhulu njengeSoftBank kanye ne-Oracle, okuhloswe ngazo ukuthuthukisa nokuphatha ingqalasizinda yayo yamafu. Ngaphezu kwalokho, ivale izivumelwano zezigidi zamaRandi nezinkampani ezifana neCoreWeave, iSamsung, i-Oracle, kanye neNvidia ukuqinisekisa ukunikezwa kwamandla ekhompyutha adingwa amamodeli ayo.

Lo mongo uchaza ukuthi kungani Ukuncintisana phakathi kweMicrosoft ne-OpenAI sekuqine kakhulu ngisho noma beqhubeka nokusebenzisana eduze. Iqembu ngalinye lifuna ukuvikela ikusasa lalo lobuchwepheshe nelezezimali ngokushintshashintsha amamodeli alo, abaphakeli, kanye nengqalasizinda.

Phakathi kwakho konke lokhu, I-MAI-Image-1 imelela isinyathelo esibonakalayo kakhulu esu le-MicrosoftKubonisa ukuthi inkampani ingakha amamodeli asezingeni eliphezulu ngokwayo ezindaweni lapho ibithembele khona kubuchwepheshe bezinkampani zangaphandle, futhi ikwenza lokho emkhakheni onemithombo yezindaba emihle kanye nomthelela wokudala njengokukhiqiza izithombe.

Nge-MAI-Image-1, iMicrosoft ihlanganisa imodeli esheshayo nekhululekile yokukhiqiza izithombe ezingokoqobo Njengoba inesu elibanzi lokuqinisa isikhundla sayo kubuhlakani bokwenziwa, ukunciphisa ukuthembela kwayo kubalingani bangaphandle, nokunikeza amathuluzi asebenzayo kubadali, amabhizinisi, kanye nabasebenzisi bokugcina, ukuhlanganiswa kwayo ne-Bing, i-Copilot, kanye nokuhlangenwe nakho kwe-multimedia kwesikhathi esizayo, kuhlanganiswe nokubuyekezwa kwayo okuhle kumapulatifomu omphakathi, kubeka lo modeli njengomunye wabancintisani abakhulu benkampani bokuncintisana enkathini entsha ye-AI yokukhiqiza.

Isihloko esihlobene:

I-Mistral 3: igagasi elisha lamamodeli avulekile we-AI esabalalisiwe

UDaniel Terrasa

Umhleli okhethekile kwezobuchwepheshe kanye nezindaba ze-inthanethi onolwazi olungaphezu kweminyaka eyishumi kumidiya ehlukene yedijithali. Ngisebenze njengomhleli kanye nomdali wokuqukethwe kwe-e-commerce, ukuxhumana, ukumaketha ku-inthanethi kanye nezinkampani zokukhangisa. Ngike ngabhala kumawebhusayithi ezomnotho, ezezimali neminye imikhakha. Umsebenzi wami nawo uwuthando lwami. Manje, ngokusebenzisa izihloko zami ku Tecnobits, ngizama ukuhlola zonke izindaba namathuba amasha izwe lobuchwepheshe elisinikeza lona nsuku zonke ukuze sithuthukise izimpilo zethu.