
Isingeniso Sephrojekthi
lanshu-create-ai-presenter-video iyi-Skill ejwayelekile yokwenza amavidiyo anama-avatar edijithali, eyenzelwe i-Codex. Udinga kuphela ukunikeza isihloko noma umbhalo ophelele, kanye nesithombe sereferensi esigunyaziwe esinomuntu omdala obonakala kahle, bese ihlela izigaba ezibalulekile zokwenza ividiyo.
Inqubo ephelele ihlanganisa ukuhlela umbhalo, ukuqopha izwi, ukudala umuntu, ukulungisa ukuvumelanisa komlomo, imibhalo engezansi, imiphumela yokunyakaza kwamagama angukhiye, ukuhlela, ukunikezela kanye nokuhlola ikhwalithi. Iphrojekthi ikhetha amakhono ngokusekelwe kumathuluzi atholakalayo endaweni ekuyo, futhi ayibophi kumhlinzeki othile, igama lemodeli noma i-API eyimfihlo kwikhodi yayo.
Le Skill ayiyona imodeli ezimele yokudala amavidiyo. Kudingeka isebenze ngaphakathi kwe-Codex noma endaweni ye-Agent ehambisana nama-Skill asendaweni, futhi incike emandleni okukhiqiza amavidiyo, izwi kanye nokuvumelanisa komlomo atholakala kuleyo ndawo.
Izici Eziyinhloko
- Ayincikile kumhlinzeki othile: Inqubo ayibophekile kumodeli, inkundla noma i-API eyimfihlo ethile.
- Uchungechunge oluphelele lokukhiqiza: Kusukela ekuhloleni okokufaka, umbhalo nezwi, kuya ekuhleleni, emibhalweni engezansi, kukhava, ekuthunyelweni nasekuhlolweni, zonke izigaba zinenqubo ehambisanayo.
- Isisekelo sesikhathi esisodwa: Izwi eliphelele lisetshenziswa njengomugqa wesikhathi wayo yonke ividiyo ukuze kubekwe ividiyo yomuntu, imibhalo engezansi, izithombe, amagama angukhiye nezinguquko.
- Izinto eziyisethenjwa zilayishwa ngezigaba: I-Codex ifunda imibhalo efanele kuphela lapho kungena isigaba esihambisanayo, kuncishiswe ukusetshenziswa komongo.
- Ukulawula izindleko: Kuqale kwenziwe ividiyo yokuhlola yomuntu enezindleko eziphansi, kusethwe izincazelo zokukhiqiza okukhokhelwayo, umkhawulo wokuzama futhi nemithetho yokuqhubeka ngemva kokuphazamiseka.
- Ukuphepha nobumfihlo: Ngaphambi kokulayisha isithombe somuntu noma ukukopisha izwi, kuqinisekiswa imvume; indawo yokugcina ayigcini okhiye be-API, amathokheni okufinyelela noma izinto zomsebenzisi.
- Ukulethwa okujwayelekile: Inqubo ihlose ukukhipha inguqulo eyinhloko, inguqulo yokwabelana kanye nombiko we-QA.
Ukulungisa Okokufaka Nendawo Yokusebenza
Okokufaka Okudingekayo Kancane
- Isihloko sevidiyo noma umbhalo ophelele.
- Isithombe sereferensi esigunyaziwe esinomuntu omdala obonakala kahle.
Ungaphinde unikeze isampula yezwi, ukuqoshwa kwesikrini, izithombe, i-B-roll, izinto zomkhiqizo, inkundla okuqondiswe kuyo, ubude bevidiyo, isilinganiso sesikrini esivundlile noma esime mpo, isitayela sokubukeka, i-watermark kanye nomyalezo wokuphetha. Isampula yezwi ingasetshenziselwa ukukopisha izwi kuphela uma kutholwe imvume efanele.
Izidingo Zendawo
- Indawo ye-Codex noma i-Agent ehambisana nama-Skill asendaweni.
- Python
3.9+. FFmpegkanye ne-ffprobe.- Bash,
jq,awkkanye ne-sed. - Okungenani ikhono elilodwa lokukhiqiza amavidiyo, izwi kanye nokuvumelanisa komlomo elingasetshenziswa endaweni ekuyo.
Ngaphambi kokuqala umsebenzi, qinisekisa ukuthi le miyalo namakhono okukhiqiza ayatholakala endaweni yamanje. Indawo yokugcina iphrojekthi ingabonwa ku-GitHub.
Ukufaka i-Skill
Qalisa lo myalo olandelayo ukuze ukopishe indawo yokugcina uyiyise kufolda yama-Skill e-Codex:
git clone https://github.com/cclank/lanshu-create-ai-presenter-video.git \
~/.codex/skills/lanshu-create-ai-presenter-video
Ngemva kokufaka, indlela kufanele ibe:
~/.codex/skills/lanshu-create-ai-presenter-video
Ukuqonda Isakhiwo Samafolda
lanshu-create-ai-presenter-video/
├── SKILL.md
├── README.md
├── agents/
│ └── openai.yaml
├── assets/
│ └── job.template.json
├── references/
│ ├── generation.md
│ ├── editing.md
│ └── qa-recovery.md
└── scripts/
├── init_job.py
├── preflight.py
└── finalize_delivery.sh
Imibhalo emithathu eyisethenjwa isebenza ezigabeni ezahlukene zokukhiqiza:
generation.md: Iphatha ukuhlolwa kokokufaka, umbhalo, izwi, ukukhetha amakhono, ukukhiqiza okukhokhelwayo, amagama ezeluleko zomuntu kanye nokungaguquguquki kwakhe.editing.md: Iphatha umugqa wesikhathi, isiqalo nesiphetho, izilungiselelo zemibhalo engezansi, imiphumela yokunyakaza kwamagama angukhiye eceleni komuntu, ikhava nokuthunyelwa.qa-recovery.md: Iphatha ukuhlolwa kobuchwepheshe, ukuhlolwa komuntu kanye nokulungisa amaphutha avamile.
Indlela Yokuqala: Ukwenza Ividiyo Ngokuqondile Engxoxweni
Indlela elula ukucela i-Skill ngokuqondile engxoxweni ye-Codex, unikeze umbhalo, isithombe somuntu kanye nezidingo zevidiyo. Isibonelo:
Sebenzisa i-$lanshu-create-ai-presenter-video ukwenza le mibhalo nesithombe somuntu kube yividiyo echaza nge-avatar yedijithali engu-16:9, enemizuzwana engu-30 nemibhalo engezansi yesikhathi sangempela.
Lapho uthumela isicelo, kunconywa ukuthi uchaze nendawo yezinto, inkundla okuqondiswe kuyo, ubude, isilinganiso sesikrini, uhlobo lwemibhalo engezansi, isitayela, i-watermark kanye nomyalezo wokuphetha. Uma ungenayo isampula yezwi egunyaziwe, ngokuzenzakalelayo kuzokhethwa izwi elifanele elikhona, kunokuba kukopishwe izwi ngaphandle kwemvume.
Indlela Yesibili: Ukuqalisa Ifolda Yomsebenzi Ejwayelekile
Kumaphrojekthi adinga ukuhlolwa kokucushwa, ukugcinwa kwamarekhodi omsebenzi noma ukwenziwa ngezigaba, qala usebenzise iskripthi sendawo yokugcina ukwakha ifolda yomsebenzi ejwayelekile.
Isinyathelo Sokuqala: Setha Indlela Ye-Skill
SKILL_DIR=~/.codex/skills/lanshu-create-ai-presenter-video
Isinyathelo Sesibili: Qalisa Umsebenzi
python3 "$SKILL_DIR/scripts/init_job.py" \
--job-dir ~/Videos/my-presenter-video \
--presenter-image ~/Pictures/presenter.png \
--topic "Isihloko sevidiyo" \
--duration 60 \
--aspect 9:16 \
--rights-confirmed \
--adult-presenter-confirmed
Lesi sibonelo sakha umsebenzi onobude bemizuzwana engu-60 nesilinganiso esingu-9:16. I---rights-confirmed isho ukuthi imvume yokusebenzisa isithombe somuntu iqinisekisiwe, kanti i---adult-presenter-confirmed isho ukuthi kuqinisekisiwe ukuthi umuntu osesithombeni sereferensi ungumuntu omdala.
Sebenzisa le mingcele yokuqinisekisa kuphela ngemva kokuqeda ngempela ukuhlola imvume nesimo somuntu. Ngaphambi kokulayisha kude, kusamele kuhlolwe imininingwane yemvume ekucushweni komsebenzi.
Isinyathelo Sesithathu: Qedela i-job.json
Ngemva kokuqalisa, ngena kufolda yomsebenzi ubuke i-job.json ekhiqiziwe. Gcwalisa izinto zokuhlola ezenziwa umuntu kanye nemvume yokulayisha kude ngokwesimo sangempela somsebenzi. Ungathumeli isithombe somuntu, isampula yezwi noma ezinye izinto ezingagunyaziwe kumasevisi akude.
Isinyathelo Sesine: Qalisa Ukuhlolwa Kwangaphambili
python3 "$SKILL_DIR/scripts/preflight.py" ~/Videos/my-presenter-video/job.json
Ukuhlolwa kwangaphambili kusetshenziselwa ukuhlola ukucushwa komsebenzi ngaphambi kokukhiqiza ngokusemthethweni. Lungisa kuqala izinkinga ezitholakele, bese uqhubekela ezigabeni ezingase zibize imali zokukhiqiza ividiyo, izwi noma ukuvumelanisa komlomo.
Ukuqonda Inqubo Ephelele Yokukhiqiza
- Thola isihloko noma umbhalo kanye nesithombe somuntu: Qinisekisa ukuthi izinto ziyasebenziseka, futhi uhlole imvume yesithombe nesimo sokuba ngumuntu omdala.
- Khiya umbhalo nezwi eliphelele: Qala unqume okuqukethwe kokugcina okuzokhulunywa, bese ukhiqiza izwi layo yonke ividiyo.
- Yenza ividiyo yokuhlola yomuntu enezindleko eziphansi: Hlola umphumela womuntu nendlela yobuchwepheshe ngaphambi kokukhiqiza okuningi okukhokhelwayo.
- Khiqiza impahla eyinhloko ye-avatar yedijithali eqhubekayo: Khiqiza izithombe eziyinhloko ngokusekelwe ezwini nasezilungiselelweni zomuntu osekuqinisekisiwe.
- Hlela ngokomugqa wesikhathi womsindo ofanayo: Yenza izithombe zomuntu, imibhalo engezansi, izithombe, amagama angukhiye nezinguquko zabelane ngesisekelo sesikhathi esisodwa.
- Faka imibhalo engezansi, imiphumela yamagama angukhiye nekhava: Qedela ukuqinisa ulwazi nokupakisha okudingekayo ekushicileleni.
- Yenza ukuhlolwa kwekhwalithi: Hlola ukuvumelanisa komlomo, ukungaguquguquki komuntu, ikhwalithi yezwi neyesithombe.
- Qedela ukulethwa: Khipha inguqulo eyinhloko, inguqulo yokwabelana kanye nombiko we-QA.
Izwi eliphelele liyisisekelo esibalulekile senqubo. Ukukhiya umsindo kuqala bese kuhlelwa ividiyo nemibhalo engezansi ngokuwulandela kunciphisa ukuphambuka kokunyakaza komlomo nezinkinga zokuhlanganisa izingxenye ezahlukene.
Izilungiselelo Zokukhipha Ezizenzakalelayo
Uma zingekho izidingo ezicacile, inqubo isebenzisa lezi ziqondiso ezizenzakalelayo:
- Isikrini esime mpo
9:16, ukulungiswa1080×1920, izinga lozimele30fps. - Amavidiyo akhiqizwa ngesihloko ngokuvamile aba phakathi kwemizuzwana engu-45 no-75.
- Uma kungekho sampula yezwi egunyaziwe, kusetshenziswa izwi elifanele elikhona.
- Isakhiwo ngokuvamile sihlanganisa isiqalo esicacile, izigqi zokuqukethwe ezingu-2 kuya kwezi-4 kanye nesiphetho esifushane.
- Umculo kanye nomyalezo wezohwebo kufakwa ngokwezidingo.
- Umgomo wevolumu yokushicilela uba cishe ku-
-16 LUFS.
Uma umgomo kuyividiyo evundlile, cacisa u-16:9 njengoba kwenziwa esibonelweni sengxoxo; uma ufuna okuqukethwe kwezinkundla zamavidiyo amafushane, ungagcina u-9:16 ozenzakalelayo noma uwusete ngokusobala nge---aspect 9:16 lapho uqala umsebenzi.
Amaqhinga Okusebenzisa Athuthukile
Khiya Izwi Kuqala, Bese Uhlela Zonke Izinto Ezibukwayo
Ungawakhi imigqa yesikhathi engaxhumene yemibhalo engezansi, izithombe nezingxenye zomuntu. Qala ukhiqize futhi uqinisekise izwi eliphelele, bese uqondanisa imibhalo engezansi, izithombe zomuntu, imiphumela yamagama angukhiye nezinguquko kumsindo ofanayo, ukuze kuthuthukiswe ukuvumelanisa okuphelele.
Yenza Ividiyo Yokuhlola Ngaphambi Kokukhiqiza Okukhokhelwayo
Ngaphambi kokuqala ukukhiqiza okukhokhelwayo, cacisa okuqukethwe okulayishwayo, ubude bokukhiqiza, isisekelo sentengo, uhlelo lwevidiyo yokuhlola kanye nomkhawulo wokuzama futhi. Ividiyo yokuhlola enezindleko eziphansi ingaveza kusenesikhathi izinkinga zokubukeka komuntu, ukuvumelanisa komlomo noma ukuhambisana kwamathuluzi, igweme ukumosha ngokukhiqiza ividiyo ephelele ngokushesha.
Ngemva Kokuphazamiseka, Qala Ngokubuyisela Umsebenzi Okhona
Uma umsebenzi okude uphazamiseka, ungawuthumeli futhi ngokushesha. Qala ubuze i-ID yomsebenzi okhona, uqinisekise ukuthi usaqhubeka noma usuvele ukhiqize umphumela, ukuze ugweme ukukhokhiswa kabili. Ngemva kokwehluleka kwabazokhethwa abathathu abakhokhelwayo ngokulandelana, yima ekukhiqizeni okuqhubekayo bese ufingqa inkinga.
Hlukanisa Ukuhlolwa Kobuchwepheshe Nokomuntu
Ukuhlolwa kobuchwepheshe kungabheka amaphutha asezingeni lamafayela nemidiya, kanti ukuhlolwa komuntu kufanele kugxile ekutheni ukunyakaza komlomo kungokwemvelo yini, umuntu uyahambisana yini, izwi lifanele yini, imibhalo engezansi ifundeka kalula yini, nokuthi isithombe siyahlangabezana yini nezidingo zokushicilela. Imithetho ehambisanayo yokuhlola nokubuyisa ukusebenza itholakala ku-references/qa-recovery.md.
Vikela Amarekhodi Omsebenzi Nobumfihlo Bezinto
- Ungabhali okhiye be-API, amathokheni okufinyelela noma amakheli okudawuniloda asayiniwe endaweni yokugcina.
- Ngaphambi kokuthumela amarekhodi ezicelo ezisezingeni lomsebenzi, susa imininingwane eyimfihlo nama-URL esikhashana.
- Emibikweni yokuhlolwa kwangaphambili neyokulethwa, gcina amagama amafayela kuphela; ungabhali izindlela eziphelele zemishini yonjiniyela.
- Qinisekisa imvume yokusebenzisa isithombe somuntu ngaphambi kokulayisha kude.
- Qinisekisa imvume yezwi ngaphambi kokusebenzisa ikhono lokukopisha izwi.
Isiphetho
lanshu-create-ai-presenter-video ihlukanisa ukwenziwa kwamavidiyo anama-avatar edijithali kube yizigaba ezijwayelekile ezingahlolwa futhi zibuyiselwe, futhi yehlisa izinkinga zokuvumelanisa umlomo nokuhlanganisa ukuhlela ngokusebenzisa umugqa wesikhathi wezwi ofanayo. Ngemva kokufaka i-Skill, ungaqala umsebenzi ngokuqondile engxoxweni ye-Codex, noma uqale wakhe i-job.json bese wenza ukuhlolwa kwangaphambili ukuze ulawule ngokuqinile imvume, izindleko, inqubo yokukhiqiza kanye nekhwalithi yokugcina.
Le phrojekthi isebenzisa i-MIT License, evumela ukusetshenziswa, ukuguqulwa nokusabalaliswa ngokukhululeka. Impendulo yokusetshenziswa ingathunyelwa nge-GitHub Issue, kanti ukuthuthukiswa kwenqubo, ukuhambisana kanye nokuhlolwa kwekhwalithi kunganikwa nge-Pull Request.
