Yakha Amavidiyo Achazayo Anehluzo Nemifanekiso Ehambayo nge-anything2explainer

Yini i-anything2explainer?

anything2explainer iyikhono le-Claude Code ne-Codex eliguqula isihloko, i-athikili, noma idokhumenti ibe yividiyo echazayo enezithombe ezinyakazayo kanye nokulandisa. Ihlanganisa ucwaningo, ukubhala umbhalo, ukuguqula umbhalo ube yinkulumo, imibhalo engezansi eqondaniswe nohlaka, ukuhlela izigcawu, ukuthuthukiswa kwezithombe ngokufanayo, ukunikezwa kwevidiyo, nokulawulwa kwekhwalithi kube inqubo eyodwa eqondiswayo.

Le phrojekthi ayilona uhlelo oluzimele lomugqa wemiyalo. Kunalokho, inikeza i-ejenti yokubhala ikhodi ene-AI indlela ephelele yokukhiqiza: ithempulethi ye-Remotion 4 engahlanganiswa, izakhi ezibonakalayo ze-React ne-TypeScript, imibhalo yezwi nokunikezwa kwevidiyo, izincazelo zesitayela ezibhaliwe, imiyalelo yokwakha ngama-ejenti amaningi, amathuluzi e-QC, kanye nefilimu eyisibonelo.

Uhlaka ngalunye ludwetshwa ngekhodi kunokuba lukhiqizwe imodeli yevidiyo noma lukopishwe kumavidiyo akhona. Ukugqwayiza kuyimisebenzi emsulwa yenombolo yohlaka futhi kusebenzisa okungahleliwe okuqalwe ngenombolo ethile, okwenza ukunikezwa okuphindwayo kukhiqize imiphumela efanayo.

Okukhiqizwa Inqubo

Umphumela ozenzakalelayo yi-MP4 ye-H.264 engu-1280×720, 30fps. Iphrojekthi eqediwe ingafaka:

• Ukulandisa okuvumelaniswe ngesiShayina noma ngesiNgisi.
• Imibhalo engezansi eqondaniswe nemingcele yamagama.
• Amakhadi zezahluko kanye nebha yenqubekelaphambili yezahluko ezansi.
• I-HUD ephezulu emise okwekhepsuli kanye nomzila wenqubo ongakhethwa.
• Izingxenye ze-Remotion zesigcawu ngasinye.
• Idokhumenti yocwaningo enemithombo, ukulandisa, i-storyboard, imibiko ye-QC, namanothi okulethwa.

Uhlelo olubonakalayo lusebenzisa ikhanvasi emnyama, ubuciko bomugqa obumhlophe, iziphawu ezinsomi, izihloko ezigqamile, kanye nesizinda sensimu yezinkanyezi esinenkungu noma igagasi lensimu yamachashazi. Ithempulethi yamanje ihloselwe ividiyo evundlile kuphela; okukhiphayo okuqondile okungu-9:16 akusekelwa.

Ubude obusekelwayo

Amamuvi aphakathi kwemizuzu emibili neyisishiyagalombili ayasekelwa. Ividiyo yemizuzu emibili kuya kwemithathu ngokuvamile isebenzisa imigqa noma izigcawu ezingu-24–32 futhi ithatha cishe ihora. Ukukhiqizwa okusezingeni lesibonelo kwemizuzu emithathu kuya kwemihlanu kuvame ukusebenzisa izigcawu ezingu-40–50 futhi kuthathe cishe amahora amabili. Amamuvi amade emizuzu emihlanu kuya kweyisishiyagalombili angasebenzisa izigcawu ezingu-60–80 futhi athathe cishe amahora amabili kuya kwamathathu.

Lezi yizilinganiso zesikhathi sangempela, hhayi iziqinisekiso. Umsebenzi omningi wenziwa ama-ejenti okwakha asebenza ngokufana.

Izidingo Zokuqala

Ngaphambi kokufaka, lungiselela okulandelayo:

• Claude Code noma Codex.
• Node.js 18 noma kamuva.
• Python 3.
• FFmpeg yokukhipha amafreyimu nokuguqula amafomethi.
• Okungenani u-5 GB wendawo ekhululekile lapho usebenzisa ama-ejenti okwakha amaningi ngesikhathi esisodwa.

Imibhalo ye-shell efakiwe isebenzisa i-zsh futhi yathuthukiswa yaphinde yaqinisekiswa ku-macOS. I-Linux kufanele isebenze, kodwa i-Windows ayikahlolwa. I-GPU ayidingeki ngoba i-Remotion inikeza ividiyo nge-Chromium esebenza ngaphandle kwesibonisi ku-CPU.

Faka Ikhono

1. Khipha ikhophi yenqolobane

git clone https://github.com/Vincentwei1021/anything2explainer.git

2. Lixhumanise ne-ejenti yakho yokubhala ikhodi

Kwi-Claude Code, dala isixhumanisi esingokomfanekiso kufolda yamakhono yayo:

ln -s "$PWD/anything2explainer" ~/.claude/skills/anything2explainer

Kwi-Codex, sebenzisa ifolda ehambisanayo yamakhono ye-Codex:

ln -s "$PWD/anything2explainer" ~/.codex/skills/anything2explainer

3. Faka izidingo zesistimu neze-Python

Kwi-macOS, faka i-FFmpeg nge-Homebrew:

brew install ffmpeg

Dala indawo ezinikele ye-Python bese ufaka izidingo ezabiwe:

python3 -m venv ~/.venvs/a2e
source ~/.venvs/a2e/bin/activate
pip install 'edge-tts==7.2.8' numpy pillow scipy

I-README ibeka edge-tts enguqulweni engu-7.2.8 ngoba incike ku-endpoint ye-Microsoft futhi ingase iphuke lapho kuthuthukiswa. Iphakheji ye-scipy idingeka kuphela yiskripthi se-QC esilinganisa ngobuningi, i-frame_metrics.py.

4. Engeza izidingo zokulandisa ngesiNgisi

Ukulandisa kwesiNgisi ngokuzenzakalelayo kusebenzisa imodeli ye-kokoro-82m esebenza endaweni kanye nezwi le-am_liam. Faka izidingo ezengeziwe uma uhlose ukukhiqiza amavidiyo esiNgisi:

pip install kokoro soundfile
brew install espeak-ng

Ukulandisa kwesiShayina ngokuzenzakalelayo kusebenzisa izwi le-edge-tts elisekelwe efwini, elingu-zh-CN-YunxiNeural.

Dala Ividiyo nge-Claude Code noma i-Codex

Indlela elula ukuchaza ifilimu oyifunayo ngolimi olujwayelekile. Ikhono kufanele lizisebenzise lona lapho litholakala ku-ejenti.

Ngakhele ividiyo echazayo ngama-vector database.

Ungenza nesicelo ngesiShayina:

讲一下向量数据库，做成一条讲解视频

Okufakayo kungaba isihloko esibanzi noma i-athikili noma idokhumenti ekhona. Uzophinde ukhethe ubude nolimi oluqondiwe.

Qonda Izigaba Eziyisishiyagalolunye Zokukhiqiza

1. Ukwakha isisekelo: I-ejenti ikopisha ithempulethi ye-Remotion iyifake kuphrojekthi entsha.
2. Ucwaningo: I-ejenti yocwaningo ilungisa idokhumenti enemithombo, efaka izinombolo eziwusizo nezifaniso ezinama-URL.
3. Ukulandisa nomugqa wesikhathi: I-ejenti ibhala umbhalo, ikhiqize noma ingenise ukulandisa, bese iguqula imingcele yamagama ibe umugqa wesikhathi oqondaniswe nohlaka kanye nethebula lemibhalo engezansi.
4. I-Storyboard: Isigcawu ngasinye sithola ububanzi bamafreyimu, isigqi sendaba, uhlelo olubonakalayo, indlela yokunyakaza, into eyinhloko, kanye nohlelo lokukhanyisa.
5. Izendlalelo nezakhi: Iphrojekthi ingeza isihloko sayo, amakhadi ezahluko, i-HUD, umzila wenqubo, nezithonjana ezimbili kuya kweziyisihlanu eziqondene nesihloko.
6. Isivivinyo: I-ejenti eyodwa yakha iqembu lokuqala lezigcawu bese inikeza ukubuka kuqala kwemizuzwana engu-30.
7. Ukwakha ngokufana: Amanye ama-ejenti akha amaqembu asele, ngokuvamile ephatha izigcawu ezinhlanu kuya kweziyisikhombisa ngalinye.
8. Ukunikezwa: Ifilimu ephelele iyanikezwa bese ihlaziywa ngamamethrikhi ohlaka alinganiswa ngobuningi.
9. I-QC nokulungisa: Ama-ejenti e-QC abuyekeza izahluko, ama-ejenti okulungisa abhekana nezinkinga, bese iphrojekthi iqinisekiswa futhi ngaphambi kokulethwa.

Sebenzisa Izindawo Zokuhlola Ezine Zokubuyekeza

Ukuhamba komsebenzi kuma ngamabomu izikhathi ezine ukuze wenze izinqumo ezibalulekile ngaphambi kokuba izinguquko zibize kakhulu.

Indawo yokuhlola 1: Ubude nolimi

Khetha ubude befilimu kanye nesiShayina noma isiNgisi ngaphambi kokuqala ukubhala umbhalo. Ubude buthinta ububanzi bokuqukethwe, usayizi wokulandisa, inani lezithombe, ukusetshenziswa kwediski, kanye nenani lama-ejenti asebenza ngokufana. Ulimi lushintsha ukubhalwa kwezinhlamvu, izabelo zemibhalo engezansi, ijubane, kanye nezwi elizenzakalelayo nge-lang ku-src/config.ts.

Indawo yokuhlola 2: Ukwamukelwa kokulandisa

Buyekeza ukulandisa okuphelele ngaphambi kokukhiqiza izwi. Uma i-voiceover isigcinwe, izinombolo zamafreyimu zifakwa ezithombeni ngazinye. Ukushintsha ngisho negama elilodwa kungadinga ukuthi yonke ifilimu ihlelwe kabusha ngesikhathi, ngakho lena yindawo eyonga kakhulu yokwenza izilungiso zokuhlela.

Indawo yokuhlola 3: Ukukhetha i-voiceover

Khetha izwi elizenzakalelayo, enye injini ye-TTS, noma ukurekhoda kwakho okuphelile. Uma ungacacisi enye indlela, i-pipeline isebenzisa i-edge-tts Yunxi yesiShayina kanye ne-kokoro-82m Liam yesiNgisi.

Indawo yokuhlola 4: Imizuzwana yokuqala engu-30

Buyekeza i-pilot render ngaphambi kokwakhiwa kwamaqembu ezithombe asele. Ukulungisa indlela yokubukeka ngalesi sikhathi kuthinta iqembu lokuqala kuphela; ukukulungisa ngemva kwe-render ephelele kungadinga izinguquko kuwo wonke amaqembu.

Qondisa I-Template Ngokuzenzela

Ungasebenzisa imibhalo efakiwe ngokuqondile uma ufuna ukulawula okwengeziwe enqubweni yokukhiqiza. Dala iphrojekthi usebenzisa i-template bese ungene kufolda yayo:

template/scripts/new_project.sh ~/work/my-video myslug
cd ~/work/my-video

Bese ulandela uchungechunge lwezinyathelo zokuzenzela:

1. Bhala ifayela locwaningo ku-research/调研.md.
2. Dala i-script/narration.txt bese ukhiqiza umugqa wesikhathi we-voiceover.
3. Bhala i-script/storyboard_src.md bese ukhiqiza i-storyboard.
4. Hlela ulimi, ingemuva, kanye nezilungiselelo ezihlobene ku-src/config.ts.
5. Qalisa amaqembu ezithombe ngaphansi kwe-src/shots/G1 kuya ku-src/shots/Gn.
6. Khupha i-preview yokuvula.
7. Khupha ividiyo yokugcina enenguqulo bese usebenzisa amamethrikhi amafreyimu.
8. Buyekeza, lungisa, bese ukhipha izinguqulo ezilandelayo ezifana no-v2 noma u-v3.

python3 scripts/tts_build.py
python3 scripts/render_storyboard.py
scripts/preview.sh 30
VER=v1 scripts/render.sh
python3 scripts/frame_metrics.py

Yenza Ngokwezifiso Ulimi Nengemuva

Izilungiselelo eziyinhloko zephrojekthi ziku-src/config.ts. Setha i-lang ibe ngu-'zh' noma 'en'. IsiShayina sisebenzisa isabelo semibhalo engezansi esiyizinhlamvu ezingu-16 ngebhulokhi, kanti isiNgisi sisebenzisa izinhlamvu ezingu-48 ngebhulokhi.

Iswishi yengemuva esekelwayo yile:

bg: 'stars' | 'dots'

Inketho yezinkanyezi ihlanganisa insimu yezinkanyezi ne-gradient yenkungu. Inketho yamachashazi isebenzisa igagasi lenkambu yamachashazi. Iphrojekthi inikeza ngamabomu ulimi olulodwa lokubukeka olunalezi zinketho ezimbili zengemuva.

Ukuze wenze izinguquko ezibanzi zokubukeka, hlela i-reference/style-guide.md kanye ne-src/ui.tsx. Izingxenye zezithombe zithembele kuma-primitives abiwe, ngakho ukushintsha lezo zincazelo ezimaphakathi kungcono kunokwenza isitayela sesithombe ngasinye ngokwehlukana.

Sebenzisa I-Voiceover Yakho

Awukhawulelwanga ezilungiselelweni ezizenzakalelayo ze-TTS ezihlanganisiwe. Faka umsindo oqediwe ku:

public/assets/<slug>/audio.wav

Bese ugcwalisa i-src/common/timeline.ts kanye ne-subs.ts ngokuzenzela. Ifomethi edingekayo ichazwe ekuqaleni kwe-tts_build.py. Uma idatha yomugqa wesikhathi nemibhalo engezansi isikhona, izigaba ze-storyboard, ukwakhiwa kwezithombe, i-rendering, kanye ne-QC zihlala zingashintshile.

Amacebiso Athuthukile Okukhiqiza

Vumela amagama ngaphambi kokupholisha izithombe

Phatha ukwamukelwa kokulandisa njengokuvala okuqinile. Ngenxa yokuthi izithombe zisebenzisa izigaba zamafreyimu ezimisiwe, izinguquko zakamuva zombhalo ziphazamisa kakhulu kunezilungiso ezijwayelekile zokubukeka.

Sebenzisa i-pilot njengokuhlola isitayela

Hlola ukubhalwa kwezinhlamvu, ukwakheka, ijubane lokunyakaza, ukukhanya, ukufundeka kwemibhalo engezansi, kanye nengemuva ngesikhathi sokubuyekezwa kokuqala kwemizuzwana engu-30. Ungayivumeli i-pilot ngenxa nje yokuthi ikhishwe ngempumelelo.

Khulisa umsebenzi ofanayo ngokucophelela

Amafilimu amade angase asebenzise ama-ejenti okwakha angu-10–14. Ama-ejenti amaningana angahlanganisa i-Remotion ngesikhathi esisodwa, okwandisa ukusetshenziswa kwesitoreji nomthwalo wesistimu. Gcina okungenani u-5 GB wesikhala samahhala futhi uthumele ama-ejenti ngamagagasi lapho usondela emkhawulweni wamaphaneli e-tmux ochazwe yiphrojekthi.

Gcina ukulandeleka komthombo

Yonke inombolo, unyaka, inhlangano, kanye negama lesiNgisi eliboniswa esikrinini kufanele kulandeleleke ku-URL ekudokhumentini yocwaningo. Izimangalo ezingaqinisekisiwe kufanele zihlale ngaphandle kokubili kokulandisa nokubukwayo.

Sebenzisa ama-render anenguqulo

Setha VER=v1, buyekeza amamethrikhi nemiphumela ye-QC yezinga lesahluko, bese udala izinguqulo ezilungisiwe ezifana no-v2 noma u-v3. Lokhu kugcina umlando ocacile wokubuyekeza esikhundleni sokubhala phezu komphumela wangaphambilini.

Funda izinto eziyizikhombo

Ifolda ye-examples/rag/ iqukethe umlando wokukhiqizwa kwefilimu eyisikhombo, okuhlanganisa ucwaningo, ukulandisa, i-storyboard, umthombo wezithombe, imibiko, namanothi okulethwa. Ifolda ye-examples/contrast/ iqukethe amapheya ayisithupha amafreyimu amabi naqondile okuhlola ukwakheka nokukhanya.

Ifolda ye-reference/ iphinde ibhale ngezindawo eziphephile, amaphalethi, amafonti, amafomula okunyakaza, imithetho yokukhanya, izivumelwano ze-storyboard, izinqubo zama-ejenti, ama-template emiyalelo, kanye nezifundo ezitholwe kumafilimu adlule.

Imikhawulo Ebalulekile Namalayisense

• Izilimi ezisekelwayo isiShayina nesiNgisi.
• Le themplethi ithatha ukuthi ividiyo yesimo esivundlile ingu-1280×720 futhi okwamanje ayikusekeli okukhiphayo okungu-9:16.
• Aklanyelwe izethulo ezibonisa umuntu okhuluma ngqo nekhamera, imikhiqizo egxile kakhulu ekuthathweni kwezithombe zangempela, noma ukukhiqiza kabusha ividiyo ekhona.
• Iziqeshana zangempela ezengeziwe ze-B-roll kufanele zitholakale emithonjeni engenazo izimali zamalungelo okusetshenziswa futhi zirekhodwe ku-manifest kanye ne-URL yomthombo, ilayisensi, i-hashi ye-SHA-256, nokusetshenziswa kwazo.
• I-toolkit isebenzisa ilayisensi ye-PolyForm Noncommercial 1.0.0. Ukusetshenziswa okungekona okokuhweba kumahhala, kanti ukusetshenziswa kokuhweba kudinga imvume yangaphambili yombhali.
• Amavidiyo adalwe nge-toolkit angawabakhi bawo.
• Amafonti afakiwe ne-Remotion anemigomo yawo yamalayisensi.

Buyekeza ikhasi lephrojekthi lenqolobane kanye namafayela amalayisensi ngaphambi kokukhiqiza noma ukusabalalisa, ikakhulukazi emsebenzini wokuhweba.

Isiphetho

anything2explainer inikeza okungaphezu kwethemplethi yevidiyo: inika uClaude Code noma uCodex uhlelo lokukhiqiza oluhlelekile lokuguqula izinto ezicwaningiwe zibe opopayi be-Remotion abanqunywe ngendlela engaguquki. Qala ngesihloko esigxilile, wenze izinqumo ngokucophelela ezindaweni zokuhlola ezine, gunyaza ukulandisa ngaphambi kwe-TTS, uqinisekise umshayeli wemizuzwana engu-30, bese usebenzisa ukunikezwa okunenguqulo kanye ne-QC yobuningi ukuze uthuthukise ifilimu yokugcina.