Yakha Umhleli Wezingubo Oyimfihlo Osebenzisa I-AI nge-Dress AI Service

Iyini i-Dress AI Service?

I-Dress AI Service iyinkundla yemfashini yomthombo ovulekile, ezisingathwa nguwe, yokuguqula izithombe zezingubo zibe ikhabethe lezingubo elibonakalayo elingaseshwa. Isebenzisa i-AI ukuhlukanisa izingubo ngezigaba, ukuncoma inhlanganisela, nokukhiqiza izethulo ezibonakalayo zezingubo ezihlongozwayo.

Le sevisi isebenza nge-backend ye-FastAPI kanye ne-frontend elula ye-HTML ne-Tailwind CSS. I-SQLite iyisizindalwazi esizenzakalelayo, kanti i-PostgreSQL iyasekelwa njengenye indlela. I-CLIP iphatha ukuqonda izithombe, kanti i-Stable Diffusion XL noma i-FLUX ingakhiqiza izithombe zezingubo ezihlanganisiwe. I-Ollama inganikeza ngokuzikhethela ukucabanga kwemodeli yolimi esebenza endaweni.

Ngenxa yokuthi uhlelo luzisingathwa nguwe, izithombe zekhabethe lezingubo zihlala kungqalasizinda oyilawulayo kunokuba zilayishwe kusevisi ye-Dress AI esingathwa kwenye indawo.

Izici Eziyinhloko

• Ukumaka ikhabethe nge-AI: Thola izigaba, imibala evelele, amaphethini, izindwangu, izitayela, nezinkathi zonyaka ezithombeni zezingubo.
• Izincomo zezingubo ezihlanganisiwe: Hlanganisa izingubo ngokombono wemibala, umcimbi, isimo sezulu, ukufana kokubonakalayo, nemithetho yemfashini.
• Ukubonisa okubonakalayo: Dala izithombe zezingubo ezihlongozwayo usebenzisa i-Stable Diffusion XL noma i-FLUX.
• Iziphakamiso ezibheka isimo sezulu: Ncoma izindwangu nezendlalelo ezifanele usebenzisa ulwazi lwesimo sezulu lwesikhathi sangempela.
• Amaphrofayela emicimbi: Khiqiza ukubukeka kwezehlakalo ezingajwayelekile, izindawo zebhizinisi, izinsuku zothando, ukuhamba, amaphathi, nezikhathi ejimini.
• Ukuhlela ikhabethe: Dala izinto ozithandayo namaqoqo, layisha izinto ngamaqoqo, futhi uthole izinto eziphindaphindiwe.
• Ukuhlaziya: Hlola imibala esetshenziswa njalo, izingubo ezingasetshenziswa ngokwanele, nezikhala ezingaba khona ekhabetheni.
• Ukwabelana nokuthekelisa: Thekelisa amabhodi ezimpahla njenge-PNG noma i-PDF, noma udale izixhumanisi ongabelana ngazo.
• Ukwandiswa: Faka izinjini zokuncoma zangokwezifiso ngohlelo lwama-plugin.
• Izindlela eziningi zokusebenzisa: Sebenzisa uhlelo lwesiphequluli, i-REST API, noma i-PWA efakekayo.

Indlela Uhlelo Olusebenza Ngayo

Ukwenza ikhabethe libe ledijithali

Uma ulayisha isithombe sengubo, isevisi iyasicubungula futhi ingasusa ingemuva ngokuzikhethela. I-CLIP ibe isihlukanisa izici ezifana nesigaba, umbala, iphethini, nendwangu. Uhlelo lugcina imethadatha etholakele kanye ne-embedding ebonakalayo ukuze kusetshenziswe ekusesheni nasezincomweni kamuva.

Izincomo zezingubo ezihlanganisiwe

Injini yezincomo ihlanganisa imithetho nokufana kwama-embedding. Isendlalelo esisekelwe emithethweni sibheka ubudlelwano bemibala, ukufaneleka komcimbi, nezimo zesimo sezulu. Ama-embedding e-CLIP asiza ekuboneni izinto ezihambisanayo ngokubonakalayo. Uma i-Ollama inikwe amandla, imodeli yolimi esebenza endaweni ingangeza ukucabanga nezincazelo ngolimi olujwayelekile.

Ukubonisa nge-AI

Ngesincomo esikhethiwe, i-Dress AI Service yakha i-prompt kusukela kumethadatha yengubo bese ingeza i-negative prompt ehloselwe ukunciphisa amaphutha ajwayelekile okukhiqiza izithombe. I-Stable Diffusion noma i-FLUX ikhiqiza isithombe ngaphambi kokucubungula okwengeziwe ngokuzikhethela, njengokukhulisa ukucaca, ukubuyisela ubuso, nokulungisa ukufana kwengemuva.

Okudingekayo Ngaphambili

Indlela enconyiwe yokufaka isebenzisa i-Docker ne-Docker Compose. Ukuze uyifake ngokuqondile, sebenzisa i-Python 3.11 noma entsha bese ufaka izidingo zekhosombe. Ukulanda amamodeli e-AI nezithombe ezikhiqiziwe kungadinga indawo enkulu yokugcina namandla ekhompyutha, kuye ngamamodeli owavulayo.

Uzodinga futhi i-Git ukuze u-clone iphrojekthi. Ukhiye we-OpenWeather API uyazikhethela futhi wenza izincomo ezibheka isimo sezulu sesikhathi sangempela zisebenze. I-Ollama nayo iyazikhethela futhi inikeza ukucabanga kwe-LLM okusebenza endaweni.

Faka nge-Docker

I-Docker iyindlela elula yokuqalisa isevisi ephelele. Clone ikhosombe uye kufolda ebizwa ngokuthi dress-ai-service, dala ifayela lakho le-environment, bese wakha ama-container:

git clone https://github.com/GangTailorUpgrade/undress-service.git dress-ai-service
cd dress-ai-service
cp .env.example .env
docker-compose up --build

Ngemva kokuqalisa, vula http://localhost:8080. Imibhalo esebenzisanayo ye-Swagger API itholakala ku- http://localhost:8080/docs, kanti i-ReDoc itholakala ku- http://localhost:8080/redoc.

Ukuze usebenzise ama-deployment ahlukanisiwe kamuva, sebenzisa:

docker-compose up -d

Faka nge-Python Yasendaweni

Uma ufuna ukusebenzisa noma ukushintsha uhlelo ngaphandle kwama-container, dala i-virtual environment bese ufaka okuncikile kwe-Python:

git clone https://github.com/GangTailorUpgrade/undress-service.git dress-ai-service
cd dress-ai-service
python -m venv venv
source venv/bin/activate
pip install -r requirements.txt

Kwi-Windows, vula i-environment ngale ndlela:

venv\Scripts\activate

Landa amamodeli e-AI adingekayo ngaphambi kokuqalisa kokuqala:

python scripts/download_models.py

Qalisa iseva yokuthuthukisa ye-FastAPI ku-port 8080:

uvicorn app.main:app --host 0.0.0.0 --port 8080 --reload

Inketho ethi --reload ilusizo ngesikhathi sokuthuthukisa ngoba iqalisa kabusha iseva lapho amafayela omthombo eshintsha. Ngokuvamile kufanele ishiywe ngaphandle lapho kwethulwa uhlelo lokukhiqiza.

Lungiselela Isevisi

Ukucushwa kunikezwa ngama-environment variables. Kopisha .env.example uye ku-.env, bese ulungisa amanani ngokuvumelana ne-deployment yakho.

HOST=0.0.0.0
PORT=8080
DEBUG=false

DATABASE_URL=sqlite:///data/wardrobe.db

USE_LOCAL_MODELS=true
SDXL_MODEL_PATH=models/sd-xl-base
FLUX_MODEL_PATH=models/flux-schnell
CLIP_MODEL=openai/clip-vit-large-patch14

OLLAMA_URL=http://localhost:11434
OLLAMA_MODEL=llama3.2

OPENWEATHER_API_KEY=your_key_here

IMAGE_WIDTH=1024
IMAGE_HEIGHT=1024
NUM_INFERENCE_STEPS=20
GUIDANCE_SCALE=7.5

UPLOAD_DIR=data/uploads
MAX_UPLOAD_SIZE=20MB

Ukulungiselelwa kwedathabheyisi

I-SQLite ifanele ukusetha okulula kwesenzakalo esisodwa futhi igcina idatha yezingubo ku-data/wardrobe.db. Ukuze usebenzise i-PostgreSQL, shintsha i-URL yedathabheyisi ibe yintambo yokuxhuma efana nale:

DATABASE_URL=postgresql://user:pass@localhost/dressai

Izilungiselelo zokukhiqiza izithombe

IMAGE_WIDTH kanye ne-IMAGE_HEIGHT zilawula ubukhulu bomphumela. NUM_INFERENCE_STEPS ibeka inani lezinyathelo zokukhiqiza ezilungiselelwe, kanti GUIDANCE_SCALE ilawula inani lokuqondisa i-prompt elilungiselelwe. Ubukhulu obukhulu nezinyathelo eziningi ngokuvamile kudinga ukubalwa okwengeziwe, ngakho zilungise ngokuvumelana nehadiwe yakho.

Izilungiselelo zokugcina

UPLOAD_DIR ikhetha lapho kugcinwa khona izithombe zezingubo ezilayishiwe. MAX_UPLOAD_SIZE ikhawulela ukulayisha ngakunye. Qinisekisa ukuthi ifolda elungiselelwe iyahlala ikhona futhi uhlelo lokusebenza lunemvume yokubhala kuyo.

Ukusetshenziswa okuyisisekelo

1. Vula i-frontend: Vakashela ku-http://localhost:8080 ngemva kokuqala kwesevisi.
2. Layisha izithombe zezingubo: Engeza izingubo ngazinye noma hudula bese uphonsa ifolda ukuze zicutshungulwe ngobuningi.
3. Buyekeza amathegi e-AI: Hlola isigaba sengubo esitholiwe, imibala, indwangu, iphethini, isitayela, kanye nesizini.
4. Khetha umcimbi: Khetha iphrofayela efana ne-Casual, Business, Date Night, Gym, Travel, noma Party.
5. Khiqiza izincomo: Vumela injini ihlanganise izingubo ezifanele isebenzisa imithetho yemfashini, ukufana kokubukeka, kanye nolwazi lwesimo sezulu olulungiselelwe, uma lukhona.
6. Buka ukubukeka: Cela umfanekiso okhiqizwe yi-AI wesincomo usebenzisa imodeli ye-Stable Diffusion noma ye-FLUX elungiselelwe.
7. Hlela umphumela: Londoloza impahla njengentandokazi noma eqoqweni, noma usebenzise izinketho ezitholakalayo zokuthekelisa nokwabelana.

Sebenzisa i-REST API

Isevisi iveza ama-endpoint anezinguqulo ngaphansi kwe-/api/v1. Qala ngokuhlola ukuthi uhlelo lokusebenza lusebenza kahle yini:

curl http://localhost:8080/api/v1/health

Faka ohlwini izingubo ezigcinwe manje ekhabethe:

curl http://localhost:8080/api/v1/wardrobe/items

Landa into ethile ngokushintsha i-ITEM_ID ufake isihlonzi sayo:

curl http://localhost:8080/api/v1/wardrobe/items/ITEM_ID

Susa into ngesicelo se-HTTP DELETE:

curl -X DELETE http://localhost:8080/api/v1/wardrobe/items/ITEM_ID

Imisebenzi eyinhloko yokubhala itholakala kulawa ma-endpoint:

• POST /api/v1/wardrobe/upload ilayisha into yokugqoka.
• POST /api/v1/outfits/generate ikhiqiza izincomo zezingubo ezihlanganisiwe.
• POST /api/v1/outfits/{id}/visualize idala umfanekiso wengubo.
• POST /api/v1/outfits/{id}/favorite igcina ingubo ezintandokazini.

Sebenzisa i-Swagger UI ku-http://localhost:8080/docs ukuze uhlole i-schema eqondile yesicelo eyamukelwa yinguqulo oyifakile futhi wenze izicelo ngokusebenzisana. Lokhu kugwema ukuqagela amagama ezinkambu zamaphayload okulayisha, okuncoma, noma okubuka ngeso lengqondo.

Thola ukuhlaziywa kwekhabethe lakho ngalokhu:

curl http://localhost:8080/api/v1/analytics/wardrobe

Izinketho zokuphakela ezithuthukile

Nika amandla ukwesekwa kwe-NVIDIA GPU

Indawo yokugcina ikhodi ihlanganisa i-Docker Compose override eqondene ne-GPU. Qala ukuphakela ngamafayela womabili e-Compose:

docker-compose -f docker-compose.yml -f docker-compose.gpu.yml up -d

Engeza ukucabanga kwe-LLM yendawo nge-Ollama

Ukuqalisa ukulungiselelwa kwe-Compose okunikwe amandla yi-Ollama, sebenzisa:

docker-compose -f docker-compose.yml -f docker-compose.ollama.yml up -d

Qondanisa OLLAMA_URL kanye ne-OLLAMA_MODEL kufayela lakho lendawo nesevisi ye-Ollama nemodeli ohlose ukuyisebenzisa. I-Ollama iyazikhethela; izingqimba zezincomo ezisekelwe emithethweni nasekuhlanganiseni izethulo ziyaqhubeka nokuba yingxenye yesistimu ngaphandle kwayo.

Nika amandla izincomo ezibheka isimo sezulu

Faka ukhiye ovumelekile ku-OPENWEATHER_API_KEY. Injini yezincomo ingabe isibheka izimo zamanje lapho ikhetha izingqimba, izindwangu, nezingubo ezilungele isimo sezulu.

Shintsha izitayela zokubuka ngeso lengqondo

I-Dress AI Service isekela ukushintshwa kwamamodeli ezitayela ezingokoqobo, ze-anime, nezomdwebo. Lungiselela izindlela ezifanele zamamodeli endawo, okuhlanganisa i-SDXL_MODEL_PATH ne-FLUX_MODEL_PATH, bese usebenzisa izilawuli zamamodeli ezivezwa yisixhumi esibonakalayo noma inguqulo ye-API oyifakile.

Izeluleko ezisebenzayo

• Qala nge-SQLite: Sebenzisa ukucushwa okuzenzakalelayo ngenkathi uhlola isevisi, bese uthuthela ku-PostgreSQL lapho izidingo zokuyisebenzisa zidinga lokho.
• Gcina amafayela alayishiwe: Yenza ikhophi eyisipele yedathabheyisi kanye nomkhombandlela okhethwe yi-UPLOAD_DIR. Kokubili kuyingxenye yedatha yekhabethe lakho.
• Sebenzisa izithombe ezicacile zezingubo: Izithombe ezihlukanisekayo zenza kube lula ukuthi uhlelo lokumaka luhlukanise izingubo futhi lukhiphe imethadatha ewusizo ebonakalayo.
• Hlola okuphindiwe: Ukutholwa kwezinto eziphindiwe okwakhelwe ngaphakathi kusiza ukuvimbela izinto eziphindwayo ngesikhathi sokulayisha amaqoqo amakhulu.
• Lungisa ukukhiqiza ngokucophelela: Qala ngobukhulu obunikeziwe obungu-1024-by-1024, izinyathelo zokucabanga ezingu-20, kanye nesikali sokuqondisa esingu-7.5, bese ukulungisa ngokusekelwe kumphumela namandla ekhompyutha atholakalayo.
• Buyekeza izithombe ezikhiqiziwe: Ukuboniswa nge-AI kuwukulinganisa okukhiqiziwe, akusona isiqinisekiso sokulingana noma ukubukeka kwengubo ngendlela eqondile.
• Sebenzisa imibhalo ye-API: Phatha i-Swagger noma i-ReDoc njengereferensi egunyaziwe yezikimu zezicelo nezimpendulo zenguqulo oyisebenzisayo.
• Vikela ukufinyelela kwenethiwekhi: Uma uveza isevisi ngaphandle kwe-localhost, engeza ukuphepha okufanele kokukhiqiza nezilawuli zokufinyelela ekusetshenzisweni okusingathwe nguwe.

Ukuhamba Komsebenzi Wokuthuthukisa

Abanikeli bangafaka izidingo zokuthuthukisa endaweni ebonakalayo:

git clone https://github.com/GangTailorUpgrade/undress-service.git dress-ai-service
cd dress-ai-service
python -m venv venv
source venv/bin/activate
pip install -r requirements-dev.txt

Qalisa iqoqo lokuhlola kanye nokumbozwa kwekhodi:

pytest tests/ -v --cov=app

Qalisa amathuluzi amisiwe okuhlola isitayela, ukufometha, nokuhlola izinhlobo:

ruff check app/
black app/
mypy app/

Faka futhi usebenzise ama-pre-commit hooks ngaphambi kokuthumela izinguquko:

pre-commit install
pre-commit run --all-files

Le phrojekthi yamukela iminikelo ngenqubo ejwayelekile ye-fork, igatsha lesici, i-commit, i-push, kanye ne-pull request. Isatshalaliswa ngaphansi kwelayisensi ye-MIT.

Ukusekelwa Kwephrojekthi

I-Dress AI Service iyiphrojekthi yomthombo ovulekile esekelwa ngabanikeli nabaxhasi bayo.

Isiphetho

I-Dress AI Service ihlanganisa ukuphathwa kwekhabethe, ukumaka okunamandla yi-CLIP, izincomo zezingubo ezixubile, ukuhlanganiswa kwesimo sezulu, kanye nokuboniswa okukhiqizayo kuhlelo olulodwa olusingathwa nguwe. I-Docker iyindlela eshesha kakhulu yokuthola ukufakwa okusebenzayo, kanti ukusethwa kwe-Python kwasendaweni kanye nama-endpoint e-FastAPI abhaliwe kunikeza isisekelo esivumelana nezimo sokuthuthukisa, sokuzenzakalela, kanye nezinhlelo zemfashini ezenziwe ngokwezifiso.