Kyauta · yana aiki a kan GPU na Hugging Face, babu buƙatar shiga
Mai Ƙirƙirar Hotuna da AI na Qwen-Image-2.1-Turbo
Qwen-Image-2.1-Turbo ita ce sigar Qwen Image 2.1 ta Alibaba ta hukuma da aka matse zuwa matakai 8. Tana riƙe rubutu mai karantuwa da cikakken bayani na cikakken samfurin, tana ƙirƙirar hoton 1K cikin daƙiƙoƙi, kuma tana iya gyara hotunan misali har 3 da umarni ɗaya kacal.
Menene Qwen-Image-2.1-Turbo?
Qwen-Image-2.1-Turbo wani checkpoint ne mai sauri na Qwen Image 2.1, wanda ƙungiyar Qwen ta Alibaba ta fitar. Yana amfani da tsarin ƙirƙirar hoto na 7B ɗaya da cikakken samfurin, amma an matse shi don ya samar da cikakken hoto cikin matakai 8 kacal na cire hayaniya.
Ƙananan matakai na nufin sakamako mai sauri sosai: hoton 1K yana ɗaukar daƙiƙoƙi kaɗan na GPU, fosta na 2K kuma ƙasa da rabin minti. Tunda tsarin bai canza ba, Turbo yana riƙe abin da ya bambanta Qwen Image 2.1 — rubutu mai kyau da aka rubuta daidai, mutane da kayayyaki masu kama da gaske, da bayyanawa ta asali.
Turbo yana kuma gyara hotuna. Ƙara hotunan misali har uku sannan ka bayyana canjin: canza salon hoto, musanya bango, ko haɗa abubuwa daga hotuna da yawa cikin wuri ɗaya.
Me yasa za a yi amfani da Qwen-Image-2.1-Turbo
Matakai 8, daƙiƙoƙi a kowane hoto
Jadawalin samfur da aka matse yana zuwa tare da checkpoint ɗin kuma yana aiki a CFG 1, don haka kowane hoto yana da sauri ba tare da daidaita sampler ko adadin matakai ba.
Rubutu da ake karantawa daidai
Fostoci, marufi, bayanan hoto da takardun karatu suna fitowa da haruffa masu tsafta da aka rubuta daidai, cikin Turanci, Sinanci da sauransu.
Gyaran hotuna da yawa
Gyara hoto ɗaya ko haɗa har uku. Kira su da <image1>, <image2> da <image3> a cikin umarninka.
Ingancin hoto uku
1K don zane-zane masu sauri, 1.5K a matsayin tsaka-tsaki, da 2K — tsohuwar zaɓi na katin samfurin — don mafi cikakken bayani.
Bango mai bayyanawa
Mayar da bango zuwa Mai bayyanawa don sitika, tambari da hotunan da aka yanke masu ainihin tashar alpha.
Inganta umarni
Kunna ingantawa sai mai sake rubuta umarni na hukuma na Qwen ya mayar da gajeren ra'ayi zuwa umarni mai cikakken bayani kafin ƙirƙira.
Yadda ake gyara hotuna da Qwen-Image-2.1-Turbo
Ɗora hotunanka, sannan ka faɗi abin da ya kamata ya canza — samfurin zai riƙe sauran komai.
- Ba hotunan misalinka lamba a cikin umarnin: "Saka wa matar da ke <image1> jaket ɗin da ke <image2>".
- Bar girman hoto a Kai tsaye don riƙe siffar hotonka na farko, ko zaɓi wani girma don sake tsara shi.
- Da hoton misali ɗaya, ana amfani da ingancin da ka zaɓa; da biyu ko uku, ana rage shi kai tsaye don gyaran ya dace da ƙwaƙwalwar GPU.
Abin da za ka iya yi da Qwen-Image-2.1-Turbo
Fostoci da hotunan kafofin sada zumunta
Zane-zane inda dole taken ya zama daidai a karon farko.
Bayanan hoto da takardun karatu
Zane-zane masu yawan bayani da alamomi, da shafukan rubutun hannu.
Gyaran hotuna
Canza tufafi, haske, bango ko salo yayin da ake riƙe abin.
Haɗa hotuna
Kawo mutane, kayayyaki da wurare daga hotuna daban-daban cikin hoto ɗaya.
Sitika da hotunan da aka yanke
PNG masu bayyanawa a shirye don manhajojin hira, kayayyaki da fayilolin zane.
Gwaji cikin sauri
Gwada ra'ayoyi goma a 1K a lokacin da samfuri mai jinkiri ke yin ɗaya, sannan ka sake yin mafi kyau a 2K da seed ɗinsa.
Yadda ake amfani da Qwen-Image-2.1-Turbo
- Rubuta umarninka — ko zaɓi ɗaya daga cikin ra'ayoyin da ke ƙarƙashin akwatin umarni.
- Ƙara hotunan misali idan kana son gyara ko haɗa hotuna; tsallake wannan don ƙirƙira daga rubutu.
- Zaɓi tsari: girman hoto, inganci, da bango mai kauri ko mai bayyanawa.
- Danna Ƙirƙira sannan ka sauke hoton da aka gama daga tarihinka.
Tambayoyin da Ake Yawan Yi
Duk abin da kake buƙatar sani game da ƙirƙirar hotuna da Qwen-Image-2.1-Turbo.
- Shi ne checkpoint na hukuma na Qwen Image 2.1 da aka matse zuwa matakai 8, daga ƙungiyar Qwen ta Alibaba. Yana amfani da tsari ɗaya da cikakken samfurin kuma yana yin duka rubutu zuwa hoto da gyaran hotuna.
- Turbo yana ƙirƙira cikin matakai 8, don haka ya fi sauri sosai, kyauta ne a nan, kuma yana iya gyara hotunan misali. Cikakken samfurin yana ba da girma 21:9 da 9:21, fitarwa a WebP da JPEG, kuma shi ne za a yi amfani da shi don aikin kasuwanci: gwada shi a shafinmu na Qwen Image 2.1.
- Eh. Ƙara hotunan misali har uku (JPEG, PNG ko WebP, kowanne har 10 MB) sannan ka bayyana canjin, kana kiran kowanne da alamar da ke nunawa a kansa a cikin faifan.
- 1K, 1.5K da 2K, a 1:1, 4:3, 3:4, 3:2, 2:3, 16:9 ko 9:16. Kai tsaye yana bin hoton misalinka na farko, ko ya ƙirƙiri hoto mai murabba'i idan babu.
- Eh — babu kiredit kuma babu shiga. Ana ƙirƙirar hotuna a kan GPU na jama'a na Hugging Face kai tsaye daga burauzarka, wanda ke ba kowane baƙo lokacin GPU na kyauta a kowace rana. Gyara, bango mai bayyanawa da inganta umarni ma kyauta ne.
- Turbo yana aiki a kan GPU na Hugging Face na gama-gari. A lokutan cunkoso za ka iya jira a layi ko lokaci ya ƙare — kawai sake gwadawa, ko zaɓi 1K wanda ya fi sauri. Idan ka gama lokacin GPU ɗinka na kyauta, yana cika da kansa bayan ɗan lokaci, don haka sake gwadawa daga baya. Ana kuma ƙin gyare-gyaren da matatar tsaro ta yi wa umarninsu alama.
- An fitar da Qwen-Image-2.1-Turbo ƙarƙashin Qwen Research License, wanda ke ba da izinin bincike da kimantawa kawai. Don ayyukan kasuwanci, yi amfani da Qwen Image 2.1 maimakon haka.