Isifundo sika-Qiaomu Codex sokwenza izithombe: Vumela u-Claude, u-Cursor nanoma iyiphi i-Agent bakhethe isitayela ngaphambi kokwenza isithombe

Isingeniso Sephrojekthi

qiaomu-codex-imagegen igoqa amandla okwakha izithombe akhelwe ngaphakathi ku-Codex abe yisevisi ye-MCP, ithuluzi lomugqa womyalo kanye ne-Agent Skill. Ngemva kokuyifaka, i-Claude Code, i-Codex, i-Cursor noma enye i-Agent ekwazi ukusebenzisa imiyalo ingakha amafayela ezithombe ngokwezicelo zolimi lwemvelo.

Ihlukile endleleni evamile yokuthi “ufake umyalo owodwa bese uthola isithombe”: ngokuzenzakalela iqala isebenzisa suggest_directions ukuncoma okungenani iziqondiso ezine zesitayela ezisebenzisa izindlela ezihlukene. Ngemva kokuba umsebenzisi ekhethe isiqondiso, compose_prompt yandisa umyalo ophelele, bese kwakhiwa futhi kuhlolwe umphumela.

Iphrojekthi inezifanekiso zezimo ezingu-24, izilungiselelo ezichazwe kusengaphambili ezingu-48, izitayela ezijwayelekile ezingu-66 kanye nezitayela zabaklami bamaphosta be-Mondo abangu-20. Iphinde inikeze imithetho yombhalo, amaqiniso, ukwakheka nokwamukelwa, ngakho ifanele ukwenza amakhava, amaphosta, izithombe zokuxhasa ezinkundleni zokuxhumana, ukubukeka kwemikhiqizo kanye nokuhlela izithombe eziyizinkomba.

Ukuhamba Komsebenzi Okuyinhloko

1. Ukuncoma iziqondiso: Ngokusekelwe esihlokweni nasendaweni yokusetshenziswa, kubuyiswa okungenani iziqondiso ezine ezinezindlela zokubuka ezihlukene, njengombhalo ogqamile, ukwakheka okuyisicaba, abantu abathwetshuliwe, ukusondela komkhiqizo noma indawo yezinto.
2. Ukukhetha isiqondiso: Umsebenzisi angakhetha isiqondiso esisodwa noma eziningi, noma avumele i-Agent izinqumele. Uma enganelisekile, angakhipha iziqondiso ezikhona bese ecela izincomo futhi.
3. Ukwandisa umyalo: Amanani eziguquguqukayo zesifanekiso afakwa emyalweni ophelele, kuhlinzekwe kanye nezinto okufanele zigwenywe, imibandela yokwamukelwa, okucatshangwayo kanye nolwazi olushodayo.
4. Ukwakha isithombe: Isiqondiso ngasinye esikhethiwe sakhela isithombe esisodwa, bese kugcinwa eceleni kwesithombe ulwazi lwe-JSON olusetshenziswe ngempela, ukuze kube lula ukuphinda umphumela nokwenza okuqukethwe kochungechunge.
5. Ukwamukela nokwenza kabusha: I-Agent ihlola isithombe isebenzisa imibandela yokwamukelwa. Uma singamukeleki, iguqula ingxenye ehambisanayo kuphela bese isakha futhi.

Uma umsebenzisi esevele ecacise isitayela, njengokuthi “sebenzisa isitayela se-Mondo” noma enikeze ngokuqondile T03, uhlelo lungakweqa ukukhethwa kwesiqondiso luye ngqo ekuhleleni umyalo noma esigabeni sokwakha.

Okulungiselelwa Ngaphambi Kokufaka

Ngaphambi kokusebenzisa, lungiselela lezi zindawo ezilandelayo:

• Node.js 18 noma ngaphezulu: Ungahlola lokhu nge-node --version.
• Codex CLI: Kudingeka ifakwe futhi ungene kuyo, bese usebenzisa codex --version ukuqinisekisa ukuthi umyalo uyatholakala.
• Imvume yokwakha izithombe: I-akhawunti yamanje ye-Codex kufanele ikwazi ukusebenzisa ukwakha izithombe. Kunconywa ukuthi uqale wakhe isithombe ngesandla ku-Codex ukuze uqinisekise.

Iphrojekthi ngokwayo iyi-node.js project engenazo iz依依 dependencia, kodwa ukwakha izithombe kwangempela kusadinga i-Codex CLI etholakala emshinini kanye nesabelo se-akhawunti.

Faka i-Agent Skill

Uma i-Agent oyisebenzisayo isekela ama-Agent Skills, ungasebenzisa ngokuqondile:

npx skills add joeseesun/qiaomu-codex-imagegen

I-Skill isiza i-Agent iqonde ukukhethwa kwesimo, ukukhethwa kwesitayela, ukubhala incazelo kanye nendlela yokwamukela, futhi yakhe izithombe nge-MCP noma ithuluzi lomugqa womyalo eliza nephrojekthi.

Bhalisa Isevisi Ye-MCP

Bhalisa Ku-Claude Code

Faka esikhundleni indlela engezansi ufake ifolda yangempela iphrojekthi ekuyo:

claude mcp add --scope user qiaomu-codex-imagegen -- node /绝对路径/qiaomu-codex-imagegen/scripts/mcp-server.mjs

Lungiselela Amanye Amaklayenti E-MCP

I-Cursor, i-Codex noma amanye amaklayenti e-MCP angafaka lokhu efayeleni lokumisa:

{
  "mcpServers": {
    "qiaomu-codex-imagegen": {
      "command": "node",
      "args": ["/绝对路径/qiaomu-codex-imagegen/scripts/mcp-server.mjs"]
    }
  }
}

Isevisi ye-MCP inikeza amathuluzi ayisikhombisa:

• suggest_directions: Incoma ngokushesha futhi mahhala okungenani iziqondiso ezine zesitayela ezinezindlela ezihlukene.
• compose_prompt: Yandisa izifanekiso nezilungiselelo ezichazwe kusengaphambili zibe umyalo ophelele, bese ibuyisa izinto okufanele zigwenywe, imibandela yokwamukelwa kanye nolwazi olushodayo.
• generate_image: Yakha isithombe esisha noma yenza ukuhlela isithombe esiyinkomba, bese ibuyisa indlela yefayela nobukhulu ngamaphikseli.
• search_prompts kanye get_prompt: Isesha futhi ifunde iqoqo lemiyalo eyizinkomba eligcinwe endaweni.
• build_prompt: Ihlola umyalo okhiqizwe umzila olula, kodwa ayakhi isithombe ngempela.
• list_catalog: Ibonisa izilungiselelo zezimo, izifanekiso, okhiye bezitayela kanye nesimo seqoqo lendawo.

Sebenzisa Umugqa Womyalo Kuphela

Uma ungayidingi i-MCP, ungakwazi futhi ukukopisha indawo yokugcina bese usebenzisa i-CLI ngokuqondile:

git clone https://github.com/joeseesun/qiaomu-codex-imagegen
cd qiaomu-codex-imagegen
node scripts/cli.mjs "窗边翻开的书，书页上升起咖啡的热气" --preset xiaohongshu

Izithombe ezakhiwe ngokuzenzakalela zigcinwa ku-~/Pictures/qiaomu-codex-imagegen/<日期>. Ungasebenzisa futhi --out, --name kanye --json ukulawula ifolda yokuphumayo, igama lefayela kanye nomphumela ofundeka ngomshini.

Ukusetshenziswa Okuyisisekelo: Ncoma Iziqondiso Kuqala

Emisebenzini yamaphosta, amakhava kanye nezithombe zokuxhasa ezinkundleni zokuxhumana, kunconywa ukuthi kusetshenziswe kuqala indlela ethi “ncoma kuqala, wakhe kamuva”. Umyalo olandelayo uzoncoma iziqondiso eziningi zomsebenzi wokufunda esitolo sekhofi, kodwa ngeke usebenzise isabelo sokwakha izithombe:

node scripts/cli.mjs suggest "一个月的咖啡店阅读活动" --for 视频封面

Umphumela ngokuvamile uchaza indlela yokubuka yesiqondiso ngasinye, ukukhiya kwesitayela, ulwazi oludingeka lufakwe kanye nezibonelo eziyizinkomba. Ungakhetha isiqondiso esisodwa noma ucele i-Agent ukuthi yakhe iziqondiso eziningi ngesikhathi esisodwa ukuze uziqhathanise.

Yandisa Umyalo Wesifanekiso

Ngemva kokunquma isiqondiso, ungabheka kuqala umyalo ophelele nemibandela yokwamukelwa:

node scripts/cli.mjs compose --template T03-1 --var topic=谷雨 --var subject=嫩芽 --var headline=谷雨

Isifanekiso sikhiqiza incazelo ehlelekile ngokusekelwe eziguquguqukweni. Ngaphezu kwesihloko nesitayela, singafaka izigaba zombhalo, indawo engenalutho, indawo ephephile, izimo zokwehluleka kanye namazinga okuhlola. Ukuhlola umyalo kuqala kusiza ukuthola iziguquguquko ezingekho noma umbhalo omude kakhulu ngaphambi kokwakha okusemthethweni.

Yakha Izithombe Usebenzisa Isifanekiso

Ngemva kokuqinisekisa ukuthi umyalo ulungile, ungakha ngokuqondile:

node scripts/cli.mjs generate --template T03-1 --var topic=谷雨 --var subject=嫩芽 --var headline=谷雨

Iphrojekthi inezigaba zezifanekiso ezingu-24, ezihlanganisa umbhalo omkhulu nendaba encane, izinto eziphuma emngceleni, isikhala sokukhanya esimaphakathi, ukuhlela ngoyinki waseMpumalanga, imifanekiso yephepha nezindawo zendwangu, imifanekiso yabantu, ukusondela kokudla, ukubukeka okuyinhloko kwemikhiqizo yobuchwepheshe, imidwebo yezakhiwo namakhadi olwazi. Izifanekiso ezihlukene azishintshi umbala noma isihlungi kuphela, kodwa zishintsha indlela yokwakheka, ubudlelwano bezinto, izigaba zombhalo kanye nobudlelwano phakathi kwezihloko.

Yakha Ngokushesha Usebenzisa Izilungiselelo Nezitayela

Uma ususazi isimo nesitayela, ungasebenzisa umzila olula ngaphandle kokuncoma iziqondiso. Isibonelo, ukwakha izinhlelo ezintathu zekhava yevidiyo:

node scripts/cli.mjs "一个巨大的红色播放键像日出一样升起" --preset video-cover --style saul-bass --count 3

Izilungiselelo zezimo ezivame ukusetshenziswa zihlanganisa:

• xiaohongshu kanye xiaohongshu-square: Amakhava e-Xiaohongshu anesilinganiso esingu-3:4 noma ayisikwele.
• video-cover: Amakhava avundlile angu-16:9 e-YouTube, B站 nezinye.
• video-vertical: Amakhava ame mpo angu-9:16 e-Douyin, WeChat Channels nama-Reels.
• wechat-cover: Isithombe esiyinhloko se-akhawunti esemthethweni ye-WeChat esingu-2.35:1.
• x-cover: Ikhava yekhasi lomuntu ku-X engu-5:2.
• moments: Iphosta ye-Moments engu-4:5.
• poster kanye movie-poster: Amaphosta ajwayelekile namaphosta amafilimu.
• book-cover, album-cover, article kanye paper: Izithombe zezincwadi, ama-albhamu, izihloko namachaza esayensi.

Isibonelo, ukwenza iphosta yefilimu engenawo umbhalo kusetshenziswa isitayela se-Mondo:

node scripts/cli.mjs "沙漠中巨大的沙虫轮廓，渺小的人影" --preset movie-poster --style olly-moss

Lawula Umbhalo Ongaphakathi Esithombeni

Amaphosta namakhava ngokuvamile adinga isihloko, isihlokwana, umugqa wolwazi noma uphawu oluncane, hhayi isihloko esisodwa kuphela. Ungasebenzisa --text ukucacisa umbhalo:

node scripts/cli.mjs "一座灯塔" --preset wechat-cover --text "夜航船"

Uma kufakwe isiShayina noma umbhalo omude, kunconywa ukuthi unikeze imisho emifushane uhlamvu ngohlamvu futhi uyihlole ngokuyikhulisa ngemva kokwakha. Incazelo yephrojekthi iveza ukuthi imodeli isengenza amaphutha amancane lapho ikhiqiza umbhalo wesiShayina noma imigqa yolwazi emincane. Emifanekisweni yabantu noma ekusondeleni komkhiqizo okungadingi ungqimba lolwazi, ungasebenzisa isixazululo esingenawo umbhalo.

Hlela Isithombe Usebenzisa Isithombe Esiyinkomba

Ungasebenzisa --ref ukunikeza isithombe esisodwa noma eziningi eziyizinkomba. Indlela yesithombe esiyinkomba kufanele ibe ngokuphelele:

node scripts/cli.mjs "保持同一支铅笔，背景换成暖米色纸张" --ref /abs/pencil.png --ratio 1:1

Le ndlela ifanele ukushintsha ingemuva, ukugcina isihloko singashintshi noma ukwenza ukulungiswa okubonakalayo kwendawo ethile. Iphrojekthi ayisinciphisi isithombe buthule. Uma isilinganiso sokugcina singahambisani nesidingo, umehluko wobukhulu uzoboniswa emiphumeleni.

Hlola Umyalo Futhi Useshe Izibonelo Zasendaweni

Uma ufuna ukuhlola umyalo kuphela ngaphandle kokwakha isithombe, sebenzisa:

node scripts/cli.mjs "咖啡和书" --preset xiaohongshu --show-prompt

Uma usuwakhile iphakheji yeqoqo lendawo ongayikhetha, ungasesha nezibonelo ezifanayo:

node scripts/cli.mjs search "节气 茶" --limit 5

Iqoqo liqukethe cishe imiyalo eyizinkomba engu-689 nezithombe ezincane, kodwa le datha evela kwabanye ayishicilelwa kanye nendawo yokugcina. Ngaphandle kweqoqo, izifanekiso, izitayela kanye nenqubo yokwakha izithombe kusengasetshenziswa ngokujwayelekile.

Ukumisa Okuthuthukile Nezeluleko Zokusebenza

Gcina Isiqondiso Sifana Ezithombeni Zochungechunge

Lapho wenza iqoqo lezithombe ze-Xiaohongshu noma amakhava amaningi, kunconywa ukuthi uqale ukhethe isifanekiso nesitayela sesithombe sokuqala, bese uvumela izithombe ezilandelayo zisebenzise ukukhiya kwesitayela, isilinganiso kanye nesakhiwo solwazi okufanayo. Lokhu kwenza kube lula ukugcina ukuvumelana kunokufaka “isitayela esifanayo” esithombeni ngasinye ngokwehlukana.

Lawula Isabelo Lapho Usebenzisa Izinguqulo Ezifanayo

--count ingakhiqiza ngokuhambisana okuhlukile okungu-1 kuya ku-4, kodwa isithombe ngasinye sisebenzisa isabelo se-akhawunti ye-Codex noma ye-OpenAI. Isithombe esisodwa ngokuvamile sithatha cishe imizuzwana engu-30 kuya ku-60, futhi njengoba inani lezithombe ezikhiqizwa ngokuhambisana landa, nokusetshenziswa kuyanda ngokufanayo.

Ukusingatha inkinga yengemuva elisobala

Iphrojekthi ihlola izindawo ezisobala emiphumeleni ekhiqiziwe. Ngokuzenzakalela, uma kutholwa izindawo ezisobala, zibekwa phezu kwengemuva elimhlophe; uma kudingeka ngempela ukugcina ingemuva elisobala, ungasebenzisa transparent_background: true. Lokhu kusetshenziselwa ukuvimbela ezinye izinhlelo zokubuka izithombe ukuthi zibonise izindawo ezisobala njengemingcele emnyama.

Ukucacisa indlela ye-Codex nesikhathi sokuvala

Uma uhlelo lungayitholi i-Codex, ungacacisa indlela yefayela elisebenzisekayo usebenzisa okuguquguqukayo kwemvelo QIAOMU_CODEX_BIN. Uma ukukhiqiza kuphelelwa isikhathi, ungakhuphula inani le---timeout, noma wenze incazelo ibe lula kuqala:

node scripts/cli.mjs "Ikhava yomcimbi wokufunda esitolo sekhofi" --preset video-cover --timeout 120000

Ukwakha iphakheji yedatha yendawo

Uma unohla lwakho lwamafayela agciniwe oluqukethe archive.json, prompts/ kanye images/, ungasebenzisa:

node scripts/build-corpus.mjs /path/to/提示词归档

Uhla oluzenzakalelayo lwedatha yendawo ngu-~/.local/share/qiaomu-codex-imagegen/corpus, futhi ungasebenzisa QIAOMU_CORPUS_DIR ukuze ukhombe kwenye indawo. Iphakheji yedatha yendawo kufanele isetshenziswe kuphela emshinini wakho; ungathumeli noma udlulisele okuqukethwe kwabanye okungagunyaziwe.

Ukuqinisekisa ukuthi ukufakwa kuphumelele

Iphrojekthi inikeza imiyalo yokuhlola ungaxhunyiwe ku-inthanethi neyokuhlola izinkomba:

npm test
node scripts/cli.mjs --list
node scripts/cli.mjs "Umbuthano obomvu" --show-prompt

npm test ayisebenzisi isabelo sokukhiqiza izithombe, futhi ingasetshenziselwa ukuqinisekisa ukuhlanganiswa kwezifanekiso, imodi yombhalo, ukushintshwa kwesakhiwo, ukuhlukahluka kokuqondiswa, ukusesha idatha yendawo, ukuphathwa kwengemuva elisobala kanye nephrothokholi.

Imibuzo evame ukubuzwa

• cannot start codex: Qinisekisa ukuthi i-Codex CLI ifakiwe, noma usethe QIAOMU_CODEX_BIN.
• codex finished without producing an image: Hlola isimo sokungena ku-Codex nokuthi i-akhawunti inemvume yokukhiqiza izithombe yini.
• timed out: Khulisa isikhathi sokuvala, noma unciphise ubunkimbinkimbi bencazelo.
• Isilinganiso asihambisani nalokho obekulindelekile: Gcizelela futhi isilinganiso encazelweni, noma usebenzise --ratio; ithuluzi ngeke lisike isithombe ngokuzenzakalelayo ngaphandle kokwazisa.
• Isithombe esiyireferensi asitholakali: Shintsha inani le---ref libe indlela ephelele.

Isiphetho

Okubalulekile ku-qiaomu-codex-imagegen akukhona nje ukudlulisela umusho owodwa kumodeli yezithombe, kodwa ukuhlela indawo, isitayela, ubudlelwano bokwakheka, amazinga ombhalo kanye nemigomo yokwamukelwa kube ukugeleza komsebenzi okungasetshenziswa kabusha. Ungaqala nge-Agent Skill noma i-MCP, uvumele i-Agent ukuthi iphakame izincomo futhi ihlele umsebenzi; noma usebenzise i-CLI ukuze ulawule ngokunembile okusethiwe, isitayela, isilinganiso, umbhalo kanye nesithombe esiyireferensi.

Emisebenzini edinga ukukhiqizwa okuzinzile kwamakhava, amaphosta kanye nochungechunge lwezithombe ezihambisanayo, kunconywa ukusebenzisa ukugeleza okuthi “qala ngokuncoma isiqondiso, bese wandisa ukwaziswa kombhalo, ngemva kokukhiqiza wamukele ngokwemibandela”, futhi ubuyekeze mathupha umbhalo ongaphakathi kwesithombe nosayizi bezithombe ezizosetshenziswa ngokusemthethweni.