GPT Images 2.6 Uyasebenza

I-GPT Images 2.6 iyimodeli yesizukulwane esisha yokukhiqiza nokuhlela izithombe eyethulwe yi-OpenAI, futhi isiyatholakala ngokuphelele ku-ChatGPT, Codex naku-API. I-GPT Images 2.6 ikhiqiza ngokushesha ngo-50%, futhi ingakhiqiza isithombe ngemizuzwana engu-6; ohlangothini lwe-API ihlukaniswe yaba amamodeli amabili, i-Flare ne-Sunburst. Amapharamitha e-complexity ne-quality alawula ngokwahlukana ukuhlelwa kwesigcawu nekhwalithi yemininingwane, okuhlanganisa isivinini nokungaguquguquki. I-GPT Images 2.6 isekela ngokwemvelo ukukhiqizwa kwama-GIF agqwayizayo, ukuthunyelwa kwama-vector e-SVG, futhi ivula umsebenzi wokuphinda kudwetshwe ingxenye ekhethiwe ku-API.

Ukukhiqiza okusheshisiwe: Isikhathi sonke sokukhiqiza sincishiswe ngo-50%; imodeli ye-Flare ingakhiqiza isithombe ngemizuzwana engu-6 esigabeni sekhwalithi ephansi, futhi ilungiselelwe ukuhlelwa kwezithombe zezentengiso njengamaphosta nezithombe zemikhiqizo ye-e-commerce.
Ukuhlukaniswa kwamamodeli amabili: Uhlangothi lwe-API luhlukaniswe lwaba amamodeli amabili, i-Flare ne-Sunburst.
I-GIF yemvelo: Ingakhiqiza ngokuqondile ama-GIF ajikelezayo anengemuva elihlanzekile nokunyakaza okuqhubekayo, alungele ama-mascot emikhiqizo, ama-meme namavidiyo amafushane ezimpawu zobuso zabasizi ababonakalayo.
Ukuthunyelwa kwe-SVG vector: Ithumela ngokwemvelo ngefomethi ye-SVG; izithonjana nemifanekiso zingakhuliswa ngokungenamkhawulo, kususwe isinyathelo sokuguqula i-bitmap ibe yi-vector ngamathuluzi ezinkampani zangaphandle, futhi zingabhalwa ngokuqondile kulabhulali yekhodi ye-frontend.
Ukuphinda kudwetshwe ingxenye nge-API: Umsebenzi wokukhetha nokuphinda udwebe, obutholakala kuphela kuwebhu ye-ChatGPT, manje usuvulelwe i-API. Usekela ukucacisa indawo yokulungiswa ngezixhumanisi noma imaski, futhi izithombe zingalungiswa ngobuningi ngenqubo ezenzakalelayo.

Ukuqamba nokunikezela kuhlukanisiwe: I-API ingeza ipharamitha ye-complexity, elawula ngokukhethekile amandla ekhompuyutha asetshenziswa imodeli ekuhleleni isigcawu ngaphambi kokudweba; ipharamitha ekhona ye-quality ibhekene kuphela nemininingwane yokubukeka nokwethembeka. Lapho kwenziwa ukulungiswa okuncane kwansuku zonke, ukwehlisa i-complexity nokugcina i-quality iphezulu kwenza imodeli ingadingi ukuphinda ihlele ngokuphelele ukwakheka kwesigcawu; yenza kuphela ukudweba kabusha okunembe kakhulu. Lokhu kufana nokulungisa ukujula kokucabanga kumamodeli amakhulu ombhalo, futhi kunciphisa kakhulu izindleko nesikhathi sokulungiswa okuningi.
Uhlelo lwamafreyimu alandelanayo lwe-GIF: Imodeli isebenzisa umqondo we-Sprite Sheet embonini yemidlalo, idwebe amafreyimu angu-6 opopayi abamisiwe emigqeni emi-3 × emi-2 esithombeni esisodwa. Amafreyimu kufanele abe nosayizi ofanayo, ikhamera ihlale endaweni eyodwa, abalingiswa nezinto zesigcawu kunganyakazi, kushintshe kuphela izimpawu zobuso nokunyakaza kancane kancane. Uhlaka lokugcina lushintshela kahle kolokuqala, bese amafreyimu esikwa ahlanganiswe abe yi-GIF.
Indlela ye-Sunburst yokugcina ukungaguquguquki: I-Sunburst inikeza ukunemba okuphezulu kwemininingwane ezingeni lekhwalithi eliphakathi neliphezulu, futhi ixazulula ikakhulukazi inkinga yokuthi abantu nemikhiqizo “ishintshe ukubukeka njengoba kulungiswa” ezingxoxweni eziningi. Ngokuhambisana nekhono lokumaka izithombe ku-API, ukulungiswa ngobuningi kungaqhubeka ngokuzinza kusetshenziswa isisekelo esifanayo sokubukeka.

Ukukhetha indawo yokungena: I-GPT Images 2.6 isiyatholakala ngokuphelele ku-ChatGPT, Codex naku-API, futhi abasebenzisi bamahhala nabakhokhayo bangayisebenzisa ngokuqondile kudeskithophu, kumakhalekhukhwini noma kuwebhu.
Umbhalo uye esithombeni: Ku-ChatGPT noma ku-Codex, chaza izidingo ngolimi lwemvelo ukuze ukhiqize isithombe. Ukuze kusheshiswe ukukhiqiza, khetha imodeli ye-Flare egxile esivinini.
Ukulawula amapharamitha: Lapho ubiza i-API, sebenzisa ipharamitha ye-complexity ukulawula amandla okuhlela isigcawu, bese usebenzisa i-quality ukulawula imininingwane nokwethembeka kwesithombe.
Ukukhiqiza i-GIF: Emyalweni cela i-GIF ejikelezayo engenamthungo, bese uchaza ukulandelana kokunyakaza nendawo yekhamera engashintshi. Imodeli izokhiqiza ngokuzenzakalelayo isithombe sopopayi esingasikwa sihlanganiswe kusetshenziswa amafreyimu alandelanayo.
Ukuthumela i-SVG: Uma udinga i-vector, cela ngokuqondile ifomethi ye-SVG. Izithonjana, izingxenye nemifanekiso zingakhuliswa ngokungenamkhawulo futhi zibhalwe ngokuqondile kulabhulali yekhodi ye-frontend.
Ukulungiswa kwengxenye: Kuwebhu ye-ChatGPT noma ku-API, sebenzisa ukukhetha kanye nezixhumanisi noma imaski ukucacisa indawo yokulungiswa. Lapho wenza ukulungiswa okuningi, kunconywa imodeli ye-Sunburst enokungaguquguquki okuphezulu.

Isivinini siphindwe kabili: Isikhathi sonke sokukhiqiza sincishiswe ngo-50%; imodeli ye-Flare ingakhiqiza isithombe ngemizuzwana engu-6 esigabeni sekhwalithi ephansi, okwenza ukuphindaphinda kushibhile.
Ukulawuleka okuphezulu: Amapharamitha e-complexity ne-quality ahlukanisa ukuhlelwa kwesigcawu nokunikezelwa kwekhwalithi, ukuze onjiniyela bakwazi ukulungisa ngokunembile ngokwezidingo.
Ukungaguquguquki okuphezulu: Imodeli ye-Sunburst igcina abantu, imikhiqizo nemininingwane yomkhiqizo kuzinzile ekulungisweni kwezingxoxo eziningi, futhi ayishintshi kalula ukubukeka “njengoba kulungiswa”.
Opopayi bomdabu: Ngaphandle kokusebenzisa imodeli yevidiyo, ikhiqiza ama-GIF agelezayo najikelezayo ngokuqondile ngokusebenzisa amafreyimu alandelanayo, ngokubambezeleka okuphansi nezinto ezingaguquki phakathi kwamafreyimu.
Ilungele ubunjiniyela: Ukuthunyelwa kwama-vector e-SVG komdabu kanye nokuphinda kudwetshwe kwengxenye ekhethiwe nge-API kususa isinyathelo sokuguqula ifomethi, futhi kungafakwa ngokuqondile ku-frontend nasezinqubweni zokulungisa izithombe ngobuningi.

Iwebhusayithi yephrojekthi: https://openai.com/index/introducing-images-2-6/

I-e-commerce nokumaketha kwemikhiqizo: Sebenzisa imodeli ye-Flare ukukhiqiza ngokushesha amaphosta nezithombe zemikhiqizo. Esigabeni sekhwalithi ephansi, isithombe sikhiqizwa ngemizuzwana engu-6, okulungele ukubuyekezwa kaningi kwezithombe zezentengiso.
Okuqukethwe kwezokuxhumana: Khiqiza ngokwemvelo ama-GIF ajikelezayo ukuze wenze opopayi bama-mascot emikhiqizo, ama-meme nezitika zokusabela, okuhambisana nesimo sokusabalalisa ezinkundleni zokuxhumana.
Ukulungiselela ukukhiqizwa kwamavidiyo: Sebenzisa ibhodi lezindaba eliqagqwayo ukuhlola ukunyakaza kwabalingiswa nesigqi sekhamera ngaphambi kokuqinisekisa umqondo wokukhiqiza ividiyo.
Ukuthuthukiswa kwe-frontend ne-UI: Ku-Codex, khiqiza ngokuqondile izithonjana ze-SVG, izingxenye nemifanekiso, bese ukubhala kulabhulali yekhodi, ususe isinyathelo esiphakathi sokuguqula i-bitmap ibe yi-vector.
Ukusebenzisana okushumekwe ezinhlelweni ze-AI: Sebenzisa i-API ukukhiqiza opopayi bezimpawu zobuso zesikhathi sangempela, njengokuthayipha nempendulo, emikhiqizweni efana nabasizi ababonakalayo; noma usebenzise ikhono lokumaka izithombe ukulungisa izithombe ngobuningi nokwakha inqubo yokuhlela yezinga lebhizinisi.