Dala futhi Hlela Izithombe nge-GPT Image Skill

Okwenziwa yi-GPT Image Skill

I-GPT Image Skill ivumela i-Codex, i-Claude Code, nama-agent endawo ahambisanayo ukuthi akhe noma ahlele izithombe esebenzisa ukubhalisa kwe-ChatGPT komsebenzisi. Isebenzisa i-$imagegen eyakhelwe ngaphakathi ku-Codex, noma ithuluzi le-image_gen elingelomethuleli uma litholakala. Ayishayi i-OpenAI Images API futhi ayidali isicelo se-Images API esikhokhelwa ngokwehlukana.

I-skill yamukela imiyalelo ebhalwe ngolimi lwemvelo kanye nezinkomba ezifundekayo zamafayela endawo e-PNG, JPEG, noma WebP. Igcina amafayela e-PNG akhiqiziwe ngaphansi kohla generated-images/ lwephrojekthi esebenzayo futhi ibuyise indlela yefayela kanye nolwazi lokubuka kuqala lwe-Markdown olushumekiwe.

Ukwakhiwa kwezithombe kudla ukusetshenziswa okufakiwe kwe-ChatGPT noma i-Codex futhi kuhlala kungaphansi kwemikhawulo yohlelo nendawo yokusebenza. Ukwakhiwa kwezithombe kungase kusebenzise imikhawulo efakiwe ngokushesha kunokujika okuvamile okungenazo izithombe, ngakho gcina amaqoqo asebenza ngokuhambisana emancane futhi ehleliwe.

Izici Eziyinhloko

• Ikhiqiza izithombe nge-Codex egunyazwe yi-ChatGPT, hhayi ngokhiye we-API.
• Ifaka i-skill efanayo ye-gpt-image ku-Codex naku-Claude Code.
• Igcina umyalo oqondile wesithombe ngaphandle kokwengeza imininingwane yokudala engacelwanga.
• Yakha imiyalo ehlukene lapho umsebenzisi enikeza ngokusobala imiqondo eminingi noma iziqondiso zokuklama.
• Isekela ireferensi eyodwa noma eziningi ngokulandelana okunqunyiwe kokunamathiselwe.
• Isekela ukuhlela, izibuyekezo eziqhubekayo, ukuhlukahluka, ukuhlanganisa izithombe, ukubonakala obala, umbhalo oqondile, kanye nemidwebo yezakhiwo eziminyene.
• Iqhuba imisebenzi yezithombe ezimele ngokuhambisana okuzenzakalelayo okungu-2 kanye nobukhulu obungu-4.
• Igcina amafayela kuphela ngaphakathi kwendawo yokusebenza esebenzayo futhi ngokuvamile igwema ukubhala phezu kwezithombe ezikhona.
• Isekela i-macOS, i-Linux, i-Windows yomdabu, ne-WSL2.
• Ivimba ukungena ngokhiye we-API futhi isusa okuguquguqukayo kwemvelo kwe-API okuhlobene ezinqubweni ezingaphansi ze-Codex.

Izidingo

Ngaphambi kokufaka i-skill, qinisekisa ukuthi imvelo inalezi zingxenye ezilandelayo:

• Node.js 22 noma entsha: Kunconywa ukukhishwa kwamanje okusekelwayo kwe-LTS.
• Git: Iyadingeka lapho kwenziwa i-clone yephrojekthi kusuka ku-GitHub.
• Codex CLI: Kufanele ibe ingene nge-ChatGPT ngaphandle kokuthi umethuleli obizayo anikeze ithuluzi lomdabu le-image_gen.
• Uhlelo olufanelekile: Uhlelo lwe-ChatGPT noma i-Codex kanye nendawo yokusebenza kufanele kuvumele ukwakhiwa kwezithombe.

I-WSL1 ayisekelwe. Uma usebenzisa i-WSL2, gcina i-Node.js, i-Codex, inqolobane, kanye nephrojekthi esebenzayo ohlangothini lwe-Linux. Gcina i-clone ngaphansi kohla lwasekhaya lwe-Linux esikhundleni se-/mnt/c.

Faka Usebenzisa I-Agent Yasendaweni

Indlela elula yokusetha iwukucela i-Codex, i-Claude Code, noma enye i-agent ehambisanayo ukuthi ifake inqolobane. Yalela i-agent ukuthi ifunde i-AGENT_INSTALL.md, yenze i-clone eqhubekayo noma isibuyekezo esiphephile se-fast-forward, idale izixhumanisi ze-Codex ne-Claude Code, iqalise ukugunyazwa kwedivayisi ye-ChatGPT, bese isebenzisa:

bootstrap --target all --yes --json

Umbiko wokulungela kufanele uphele ngo-ok: true, status: ready, kanye no-best_practice_pass: true. Umsebenzisi uqedela yena mathupha ukugunyazwa kwesiphequluli noma kwedivayisi. I-skill ayiceli iphasiwedi, ithokheni, ukhiye we-API, noma okuqukethwe kwe-~/.codex/auth.json.

Inqubo yokufaka akufanele ikhiqize isithombe esibukhoma. Kufanele iphele ngomhlahlandlela omfushane wokuqala oqukethe izilinganiso zobude nobubanzi ezivamile, imishwana yekhwalithi, kanye nezibonelo zezicelo zokudala nezereferensi.

Ukufaka Mathupha

macOS, Linux, ne-WSL2

Sebenzisa i-clone eqhubekayo ngoba izixhumanisi ze-skill efakiwe zikhomba enqolobaneni:

REPOSITORY_URL="https://github.com/GENEXIS-AI/gpt-image-skill"
INSTALL_DIR="${XDG_DATA_HOME:-$HOME/.local/share}/gpt-image-skill"

git clone "$REPOSITORY_URL" "$INSTALL_DIR"
cd "$INSTALL_DIR"
node ./gpt-image/scripts/validate_skill.mjs
node ./gpt-image/scripts/gpt_image.mjs bootstrap --target all --yes --json

I-Native Windows PowerShell

$RepositoryUrl = "https://github.com/GENEXIS-AI/gpt-image-skill"
$InstallDir = Join-Path $env:LOCALAPPDATA "gpt-image-skill"

git clone $RepositoryUrl $InstallDir
Set-Location $InstallDir
node .\gpt-image\scripts\validate_skill.mjs
node .\gpt-image\scripts\gpt_image.mjs bootstrap --target all --yes --json

Ku-macOS, Linux, naku-WSL2, ukufaka kudala ama-symlink. I-Windows yomdabu isebenzisa ama-directory junction. Izindlela ezikhona ezingahlobene azishintshwa.

• Ukufakwa kwe-Codex: ~/.agents/skills/gpt-image
• Ukufakwa kwe-Claude Code: ~/.claude/skills/gpt-image
• Ukufakwa kwe-Windows Codex: $env:USERPROFILE\.agents\skills\gpt-image
• Ukufakwa kwe-Windows Claude Code: $env:USERPROFILE\.claude\skills\gpt-image

Yakha Isithombe Sakho Sokuqala

Ngemva kokufaka, biza i-skill ngokuqondile kumethuleli. Ku-Codex, sebenzisa:

$gpt-image A cobalt-blue glass robot on a warm off-white background.

Ku-Claude Code, sebenzisa:

/gpt-image A cobalt-blue glass robot on a warm off-white background.

Ungachaza ngokwemvelo indlela yokubiyela nekhwalithi:

$gpt-image Create a cozy reading room at sunset, 16:9, high quality.

Izicelo ezivamile zifaka 1:1, 16:9, 9:16, 4:3, kanye ne-3:4. Imishwana yekhwalithi ewusizo ihlanganisa draft, high quality, kanye ne-high detail / final quality. Lezi yizicelo zolimi lwemvelo esikhundleni sohlu olungaguquki losayizi be-API, ngakho izilinganiso eziqondile zamaphikseli zingahluka.

Sebenzisa I-Direct Runner

Kusuka enqolobaneni, ungabiza i-Node.js runner ngokuqondile:

node ./gpt-image/scripts/gpt_image.mjs generate \
  --prompt "Irobhothile yengilazi enombala oluhlaza okwesibhakabhaka we-cobalt ngemuva elimhlophe elifudumele." \
  --out "generated-images/glass-robot.png"

Ukukhiqiza okujwayelekile kwenza ukuhlola okusheshayo kokuqinisekisa ubuqiniso ku-ChatGPT, kwakhe isithombe esisodwa, kwenze ukuhlola okuncane kwe-PNG, bese kubuyisa inani elithi PATH=... kanye nenani eliphelele le-MARKDOWN=....

Khiqiza Isithombe Ngereferensi

Amareferensi kufanele kube amafayela angempela afundekayo. Ku-Claude Code, khetha i-@path noma indlela ecacile yesistimu yamafayela:

/gpt-image Sebenzisa i-@references/robot.png njenger referensi yomlingiswa. Yidwebe igibele ibhayisikili.

Isithombe esinamathiselwe noma esihudulelwe esiboniswa ku-Claude asidluliselwa ngokuzenzakalelayo enqubweni ye-Codex efakwe ngaphakathi. Uma u-Claude eveza indlela enembile yokunamathiselwe kwesikhashana, i-agent ingakopisha lelo fayela ku-generated-images/inputs/. Kungenjalo, gcina isithombe ngaphakathi kwephrojekthi bese unikeza indlela yaso. I-skill ayiqageli ukuthi yiliphi ifayela okufanele ilisebenzise kusilondolozi sezithombe sika-Claude.

Umyalo oqondile olinganayo uthi:

node ./gpt-image/scripts/gpt_image.mjs generate \
  --mode generate \
  --prompt "Dweba lo mlingiswa egibele ibhayisikili." \
  --reference "/absolute/path/robot.png" \
  --out "generated-images/robot-bicycle.png"

Namathisela Amareferensi Amaningi

Phinda i---reference ngohlelo oludingekayo. Engeza izindima ezihambisanayo kuphela uma amareferensi enezindima ezicacile:

node ./gpt-image/scripts/gpt_image.mjs generate \
  --prompt "Sebenzisa i-Image 1 kumlingiswa ne-Image 2 ekwakhiweni kwebhayisikili." \
  --reference "/absolute/path/character.png" \
  --reference-role "character" \
  --reference "/absolute/path/bicycle.png" \
  --reference-role "bicycle design" \
  --out "generated-images/combined.png"

Ekukuhlelweni, okuqondiswe ukuhlelwa yi-Image 1. Amareferensi asekelayo alandela yona ngokohlelo lomugqa womyalo.

Hlela Futhi Buyekeza Isithombe

Sebenzisa i---mode edit ne---edit-target ukuze uguqule isithombe esikhona:

node ./gpt-image/scripts/gpt_image.mjs generate \
  --mode edit \
  --prompt "Faka esikhundleni sikabhasikidi webhayisikili ibhokisi elincane lokhuni." \
  --edit-target "generated-images/robot-bicycle.png" \
  --out "generated-images/robot-bicycle-crate.png"

Ekubuyekezeni ngakunye okulandelayo, hlela umphumela wakamuva esikhundleni sokubuyela emthonjeni wokuqala:

node ./gpt-image/scripts/gpt_image.mjs generate \
  --mode edit \
  --prompt "Yenza ibhokisi lokhuni libe luhlaza ngokumnyama." \
  --edit-target "generated-images/robot-bicycle-crate.png" \
  --out "generated-images/robot-bicycle-green-crate.png"

Phinda unamathisele wonke amareferensi asekelayo asadingeka. Ukucelwa ngakunye kwebhuloho kuyaphela, ngakho akugcini ngokuzenzakalelayo okunamathiselwe kwangaphambilini. Ukuhlela umthombo wokuqala esikhundleni somphumela wakamuva kungalahla isibuyekezo sangaphambilini.

Qalisa Izithombe Eziningi Ngokuhambisana

Sebenzisa i-generate endleleni yesithombe esisodwa. Emisebenzini emibili noma ngaphezulu elungile, i-skill ingasebenzisa iqoqo elinomkhawulo. Ukusebenza okuzenzakalelayo ngesikhathi esisodwa ngu-2, kanti okuphezulu ngu-4. Ukuqinisekisa ubuqiniso kuhlolwa kanye kuphela eqoqweni, futhi imisebenzi ehlulekile ayiphindwa ngokuzenzakalelayo.

Imodeli yokuqoqwa isekela izakhiwo ezintathu ezivamile:

• Ukwahluka okunehange elabiwe: Wonke umsebenzi ufunda umklamo ofanayo okhona bese usebenzisa isitayela esihlukile.
• Imiqondo enikezwe i-agent: I-agent idala umyalo ophelele nowehluke ngokunenjongo wesithombe ngakunye esiceliwe, igcine imikhawulo efanayo.
• Ukunikezwa okuphindaphindiwe: Uma kungekho mehluko oceliwe, umsebenzi ngamunye uphinda usebenzise umyalo ofanayo ncamashi.

Dala i-manifest engaphakathi kwendawo yokusebenza efana ne-image-jobs.json:

{
  "version": 1,
  "jobs": [
    {
      "id": "watercolor",
      "mode": "variation",
      "prompt": "Gcina umklamo ofanayo bese uwunikeza ngesitayela se-watercolor.",
      "edit_target": "references/base-design.png",
      "references": ["references/watercolor-style.png"],
      "reference_roles": ["style reference for this output"],
      "out": "generated-images/design-watercolor.png"
    },
    {
      "id": "clay",
      "mode": "variation",
      "prompt": "Gcina umklamo ofanayo bese uwunikeza ngesitayela sobumba.",
      "edit_target": "references/base-design.png",
      "references": ["references/clay-style.png"],
      "reference_roles": ["style reference for this output"],
      "out": "generated-images/design-clay.png"
    }
  ]
}

Qalisa i-manifest ngalo myalo:

node ./gpt-image/scripts/gpt_image.mjs batch \
  --manifest "image-jobs.json" \
  --concurrency 2

Imisebenzi eminingana ingafunda ngokuphepha ihange elifanayo. Namathisela kuphela ireferensi yesitayela efanele kumphumela ngamunye. Emiqondweni engahlobene, susa i-edit_target evamile bese unikeza umsebenzi ngamunye umyalo ozimele namareferensi awo.

Umphumela odalwe umsebenzi owodwa weqoqo awunakuba okokufaka komunye umsebenzi kulelo qoqo elifanayo. Qalisa ukuhamba komsebenzi okunokuncika kokuphuma kuya kokokufaka ngezigaba. Uma kungehange elivamile, khiqiza umphumela wokuqala oceliwe bese usebenzisa indlela yawo ebuyisiwe ezinguqulweni zakamuva; i-skill ayidali ihange elifihlekile elengeziwe.

Amathiphu Okusetshenziswa Okuthuthukile

Gcina Inhloso Yomyalo

Esicelweni esisodwa esiqondile, amagama omsebenzisi yiwona agunyazayo futhi adluliselwa engashintshiwe. Uma umsebenzisi ecela imiklamo, imiqondo, izindlela, izinketho noma ezinye izindlela eziningana ezihlukene, inhloso yokudala enikezwe i-agent nayo iyagunyaza. Kuleso simo, i-agent yenza umyalo ohlukile olungele isithombe somphumela ngamunye, igcine isihloko esifanayo, umkhiqizo, umbhalo, amareferensi, isilinganiso sobude nobubanzi, neminye imikhawulo.

Izinombolo zemisebenzi zifakwa kuma-ID ama-manifest nasemagameni wamafayela, hhayi emiyalweni yezithombe. Gwema ukwengeza imishwana efana nokuthi lena inketho yesibili.

Khetha Imodi Efanele

• Sebenzisa --mode generate emsebenzini wokuguqula umbhalo ube isithombe.
• Engeza --reference PATH ukuze udale isithombe esisha esiqondiswa ifayela elikhona.
• Sebenzisa --mode edit --edit-target PATH ukuze uguqule isithombe esikhona.
• Sebenzisa --mode variation --edit-target PATH ukuze udale inguqulo ehlukile yomklamo okhona.
• Sebenzisa --background transparent kuphela uma kuceliwe ukuba kube nokusobala.
• Sebenzisa i-edit target efanayo emisebenzini eminingi yokwakha izinguqulo, ukuze uthole ukwakheka okukodwa okuhunyushwe ngezitayela ezahlukene.
• Sebenzisa isithombe esifanayo njengereferensi emisebenzini eminingi yokukhiqiza lapho ubuwena kufanele buvele ezigcawini noma ezakhiweni ezahlukene.

Gwema Ukubhala Ngaphezulu Kwamafayela Ngengozi

Isiqalisi asibhali ngaphezulu kwesithombe esikhona ngaphandle kokuthi --overwrite sinikezwe ngokusobala. Nikeza umphumela ngamunye indlela echazayo ngaphansi kwe-generated-images/.

Ukuhlola Izinkinga Nezinyathelo Zokusetha

Ukuhlela nokuxilonga okuningiliziwe kuyizinketho. Akudingeki ngaphambi kokukhiqiza okujwayelekile. Kusebenzise lapho ukusetha, ukuqinisekiswa, amareferensi, noma amafayela emiphumela kudinga uphenyo:

node ./gpt-image/scripts/gpt_image.mjs doctor --json
node ./gpt-image/scripts/gpt_image.mjs guide
node ./gpt-image/scripts/gpt_image.mjs capabilities --json
node ./gpt-image/scripts/gpt_image.mjs inspect \
  --input "generated-images/combined.png" --json

Ukuhlola ukungena ngemvume nezindlela zamafayela ngaphandle kokudala isithombe:

node ./gpt-image/scripts/gpt_image.mjs generate \
  --prompt "test" \
  --out "generated-images/test.png" \
  --dry-run --json

Ukuqinisekisa i-batch manifest nokuhlelwa kwayo ngaphandle kokungena ngemvume noma ukukhiqiza izithombe:

node ./gpt-image/scripts/gpt_image.mjs batch \
  --manifest "image-jobs.json" \
  --check-only --json

Izithombe ezikhiqiziwe zithola ukuhlolwa okulula kwefayela nesiginesha esikhundleni samarisidi e-SHA-256. I-SHA-256 isatholakala lapho inengqondo khona, njengokuqinisekisa izifaki:

node ./gpt-image/scripts/gpt_image.mjs verify-installers --json

Izinyathelo Zokuvikela Ukuphepha Nobumfihlo

• Ikhono lisusa OPENAI_API_KEY, OPENAI_BASE_URL, OPENAI_ORG_ID, OPENAI_PROJECT_ID, kanye ne-CODEX_ACCESS_TOKEN ezinqubweni ezingaphansi ze-Codex.
• Ukukhiqiza kuyavinjelwa ngaphandle kokuthi ukuxilonga okususwe imininingwane ebucayi kuqinisekise ukuqinisekiswa kwe-ChatGPT.
• Ayikho i-endpoint ye-OpenAI Images API noma isicelo se-/v1/images esifakiwe.
• Ikhono alifundi amafayela okuqinisekisa.
• Amafayela akhiqiziwe ahlala ngaphakathi kwendawo yokusebenza esebenzayo.
• Lapho kufinyelelwa emikhawulweni yokubhalisa, isiqalisi asibuyeli ku-Images API.

Buyekeza Ikhono

Kuma-macOS, Linux, noma WSL2, hambisa phambili ikhophi ephikelelayo bese uphinda uqalise i-bootstrap:

INSTALL_DIR="${XDG_DATA_HOME:-$HOME/.local/share}/gpt-image-skill"
git -C "$INSTALL_DIR" pull --ff-only
node "$INSTALL_DIR/gpt-image/scripts/gpt_image.mjs" \
  bootstrap --target all --yes --json

Kwi-Windows PowerShell yomdabu:

$InstallDir = Join-Path $env:LOCALAPPDATA "gpt-image-skill"
git -C $InstallDir pull --ff-only
node "$InstallDir\gpt-image\scripts\gpt_image.mjs" bootstrap --target all --yes --json

Isiphetho

GPT Image Skill inikeza indlela egxile endaweni yokusebenza yokukhiqiza nokuhlela izithombe kusuka ku-Codex, Claude Code, noma kwenye i-agent ehambisanayo esebenzisa ukuqinisekiswa kwe-ChatGPT. Qala ngesicelo esiqondile ngolimi lwemvelo, nikeza izindlela zasendaweni ezizinzile zamareferensi, sebenzisa umphumela wakamuva ekubuyekezeni ngakunye, futhi ugcine ama-batch anomkhawulo emiqondweni ezimele noma ezinguqulweni ezabelana nge-anchor.

Ukuze uthole inkontileka yokufaka kanye nemininingwane ejulile yokusebenza, vakashela indawo yokugcina ye-GPT Image Skill.