Umhlahlandlela Womyalo Wezithombe ze-AI
I-Image Model ye-Shannon ifunda umyalo njengemodeli yolimi. Leli qiniso elilodwa lishintsha yonke into ngendlela okufanele uwubhale ngayo — kuqala ngokulahla amathegi ahlukaniswe ngokhefana.
Ngamafuphi
Bhala umbhalo ogelezayo ngemisho ephelele, hhayi amathegi amagama. Hlosa amagama angu-30–80. Wuhlele ngendlela imodeli ewufunda ngayo: isihloko → isenzo → isitayela → indawo nokukhanya → ubuchwepheshe, okubaluleke kakhulu kuqala. Yiba nesihloko esicacile. Uma ufuna isithombe esithwetshuliwe, sibhale njengesithombe esithwetshuliwe — bala ikhamera, ilensi, nefilimu noma ukukhanya. Namathisela imibala eqondile ezintweni ezibalulwe ngegama ngokuthi color #RRGGBB. Khetha esisodwa sezilinganiso zohlangothi ezinhlanu. Bese ulahla imikhuba yesikhathi se-SDXL: akukho "masterpiece, best quality, 8k", akukho zisindo ze-(word:1.3), futhi akukho myalo ophikisayo ekusukeni embhalweni uya esithombeni — chaza kuphela lokho okufunayo. Ukuhlela nokuhlanganisa kona kuyawamukela umyalo ophikisayo.
Abantu abaningi bafika benemikhuba yemiyalo abayifunde ku-SDXL: uchungechunge lwamabizo ahlukaniswe ngokhefana, isandulela samathokheni ekhwalithi, izisindo ezimbalwa ezisemabakithini, nomyalo ophikisayo omude njengophikayo. Leyo mikhuba yayilungile kulawo mamodeli futhi ilimaza lapha. Lena yinguqulo emfushane yalokho okusebenza ngempela — futhi ngoba i-Shannon ikubhalela umyalo wokwakha bese ikubonisa wona ngaphambi kokuba kwakhiwe okuthile, ngokulinganayo ingumhlahlandlela wokufunda nokulungisa lokho i-Shannon ekuhlongozayo.
01Kungani umbhalo unqoba amathegi: imodeli ifunda njengemodeli yolimi
Amamodeli amadala e-diffusion iseluleko esiningi semiyalo esasibhalelwe wona ayesebenzisa isibhali sombhalo esasiziphatha, ekusebenzeni, njengesikhwama semiqondo: uhlelo lolimi lwalungabalulekile neze, futhi isu elinqobayo kwakuwukunqwabelanisa amabizo afanele bese uwachema ngezisindo. Isobho lamathegi lalinesimo esifanele somyalo kuleso sibhali.
I-Image Model ye-Shannon ayisebenzi ngaleyo ndlela. Ifunda umyalo ngendlela imodeli yolimi efunda ngayo umbhalo, ngakho uhlelo lolimi luthwala incazelo. "Holding an umbrella against her shoulder" kuwubudlelwano phakathi komuntu, into nengxenye yomzimba; "umbrella, shoulder" kuyizinto ezimbili ezenzeka zikhona. Izandulelabizo zibeka izinto. Izenzo zenza ukuma. Imisho inamathisela imininingwane ebizweni elifanele kunokuyivumela ukuba intante kuso sonke isithombe.
Nasi isithombe esifanayo esicelwe ngezindlela zombili.
woman, 30s, tokyo, rain, night, crosswalk, neon signs, umbrella,
wet asphalt, reflections, masterpiece, best quality, 8k, ultra
detailed, (bokeh:1.3), cinematic lighting, photorealistic
A woman in her early thirties waits at a rain-soaked Tokyo
crosswalk at night, holding a clear vinyl umbrella against her
shoulder and glancing down the street. Neon storefront signs
bleed red and cyan into the wet asphalt at her feet. Shot on
Kodak Portra 400, 85mm f/1.4, shallow depth of field, the
traffic signal behind her thrown out of focus.
Bheka lokho inguqulo yethegi engakwazi ukukusho. Ayilokothi ithi isambulela siphethwe, ngakho singagcina silele phansi. Ayilokothi ithi i-neon ibonakala ku-asphalt, isho nje ukuthi kokubili kukhona, ngakho uthola izimpawu nomgwaqo omanzi kungekho budlelwano bokukhanya phakathi kwakho. Ayilokothi ithi owesifazane ubheka kuphi, ngakho ukuma kwakhe kuyinhlanhla. "Cinematic lighting" iyilebula lomoya, hhayi umthombo wokukhanya. Futhi amathokheni amahlanu okugcina angumsindo, afundwa njengamagama esiNgisi ngokuthwebula izithombe kunokuba yizibambo zokulawula.
Inguqulo yombhalo ayindanga ngoba ihlobisiwe. Indanga ngoba iphendula imibuzo inguqulo yethegi eyishiye ivulekile. Yilo lonke iqhinga.
02Kufanele ube mude kangakanani umyalo wesithombe?
Hlosa amagama angu-30 kuya kwangu-80. Ungadluli kakhulu ku-100. Yilesi sikhwepha sekhwalithi esilula kunazo zonke onaso, futhi zombili izinhlangothi zebanga zibalulekile.
Ngaphansi kwamagama acishe abe ngu-30 ukwakheka, ukukhanya nohlaka akucacisiwe, ngakho imodeli ikhetha noma yini evamile ngokwezibalo. "A woman in a coffee shop" ngumyalo osemthethweni okhiqiza isithombe esikhohlakalayo, ngoba unikeze zonke izinqumo ezithakazelisayo kwenye into.
Ngaphezu kwamagama acishe abe ngu-100 kuvela okunye ukwehluleka: imisho yakamuva incintisana neyokuqala ngezindawo ezifanayo zesithombe, futhi esikhundleni sokunqwabelanisa imininingwane uthola ukuhlanjululwa. Incazelo yesithathu yokukhanya ichitha eyokuqala, futhi isihloko osimise emagameni ayisishiyagalombili okuqala sithola ukunakwa okuncane njalo ngomusho owengezayo. Imiyalo emide ifundeka njengephelele. Ayakhi njengephelele.
Isiyalo: bhala umyalo owufunayo, bese uwunciphisa ube yimisho ebizoshintsha isithombe uma isuswa.
03Isiyalo sokulandelana: into ebaluleke kakhulu kuqala
Ngoba imodeli ifunda ngokulandelana, indawo yigcizelelo: okuza kuqala kubopha isithombe, okuza ekugcineni kuyasilungisa. Beka izinto ngokulandelana imodeli ezidinga ngakho:
- Isihloko — ukuthi lesi sithombe singesikabani noma singesani, ngokucacile.
- Isenzo noma ukuma — ukuthi benzani, baphethwe kanjani.
- Isitayela noma umoya — isithombe esithwetshuliwe, umdwebo wamafutha, umdwebo wobuchwepheshe, nezinga elikuwo.
- Indawo nokukhanya — ukuthi lokhu kwenzekaphi nokuthi kukhanya kuphi.
- Ubuchwepheshe — uhlaka, ilensi, ukujula kwenkundla, imininingwane yombala.
Inxeba elivame kakhulu elizenzelwayo ukuqala ngomusho wobuchwepheshe. Ukuvula ngokuthi "Cinematic wide shot, anamorphic lens flare, teal and orange grade, of a man reading a newspaper" kutshela imodeli ukuthi ukubukeka yisona sihloko futhi indoda iwuchashazi. Uthola uhlaka oluhle, olungenalutho olunomuntu omncane ongacaci kulo.
Cinematic wide shot, anamorphic lens flare, teal and orange
color grade, shallow depth of field, golden hour, of a man
reading a newspaper on a park bench.
A man in his sixties in a gray wool overcoat sits on a park
bench reading a folded broadsheet newspaper, one ankle crossed
over his knee. Late golden-hour sun rakes across him from the
left through bare plane trees. Photographed wide on a 35mm
anamorphic lens, shallow depth of field, warm highlights and
cool shadows.
Zonke izinto ezisenguqulweni yokuqala ziyasinda kweyesibili; akukho okulahliwe. Imodeli manje isiyazi nje ukuthi isithombe singesani ngaphambi kokuba itshelwe ukuthi isithwebule kanjani — futhi lokho kuhlelwa kabusha kodwa kususa isihloko ekubeni ngesingabalulekile kuye ekubeni ngesiphakathi.
04Yiba nesihloko esicacile
"A person" akusona isihloko, kuyindawo evulekile — futhi imodeli igcwalisa izindawo ezivulekile ngesilinganiso sakho konke ekubonile, okuyikho ncamashi lapho ukubukeka okucwebezelayo, okungenaminyaka, okungenabuntu kuvela khona. Ukucaca kubiza amagama amahlanu bese kukuthengela umuntu othile:
| Okungacacile | Okucacile |
|---|---|
| umuntu | indoda esemashumini amane anentshebe emnyama nemhlophe nemigqa ejulile yokumamatheka |
| owesifazane | owesifazane osemashumini amabili nesishiyagalolunye onezinwele ezigundwe ngaphansi nezinyo langaphambili elifileyo |
| isakhiwo esidala | indlu yokugcina impahla yezitezi ezine yezitini enophawu olupendiwe ngaphezulu namafasitela angenhla avaliwe ngamapulangwe |
| ukudla | indishi ye-ramen enaqanda elibilise kancane elisikwe kabili ngaphezulu, nomusi okhuphukayo |
Qaphela iphethini: iminyaka noma isikhathi, isici esisodwa sesakhiwo, ukungapheleli okukodwa okuhlukanisayo. Into yesithathu yiyona eyenza umphumela uzwakale uthwetshuliwe kunokwakhiwa. Izinyo elifileyo, umphetho ogugile, uphawu olupendiwe ngaphezulu — ukungalingani okuncane kuluphawu oluqine kunazo zonke lokulwa nokubukeka kwepulasitiki, futhi cishe akubizi lutho esabelweni sakho samagama.
05Ukubukeka njengesithombe sangempela: bhala isicelo njengesithombe esithwetshuliwe
Ukubukeka njengesithombe sangempela yizinga elizenzakalelayo ngaphandle uma ucela ubuciko, umdwebo noma i-anime. Kodwa igama elithi "photorealistic" alenzi lutho oluningi. Okukhiqiza isithombe esithwetshuliwe ukuchaza isenzo sokuthwebula: ikhamera yangempela, ilensi yangempela, nefilimu yangempela noma umthombo wokukhanya.
Isizathu sicacile. Sonke isithombe esasinamagama athi "Kodak Portra 400" senziwa ngenqubo ethile enesakhiwo esithile sokuqinile, ukuphela kokukhanya nokubukeka kwesikhumba, futhi ukubalula ifilimu kubiza konke lokho. "8k ultra detailed" sasivela kuphela kudoti wamasayithi ezithombe, ngakho sibiza udoti wamasayithi ezithombe. Amaresiphi ambalwa athembekile, ngalinye likhiqiza isithombe esihluke ngokusobala sesihloko esifanayo:
| Iresiphi | Ikunikani |
|---|---|
| Kodak Portra 400, 85mm f/1.4, soft window light | Imibala yesikhumba efudumele, ethethelelayo; ukwehlukaniswa okuthambile; isithombe somuntu sephephabhuku |
| Ilford HP5 pushed to 1600, 35mm, available light | Okumnyama nokumhlophe okunesihlabathi, okunomehluko omkhulu, namandla wobufakazi basemgwaqweni |
| Digital medium format, 80mm, single softbox at 45° | Ukubukeka okuhlanzekile kwesitudiyo sezentengiselwano, ukuphela kokukhanya okulawulwayo, izinga lomkhiqizo |
| 135mm f/2, backlit at golden hour, atmospheric haze | Ingemuva elicindezelwe, ukukhanya emaphethelweni, ukubukeka okuthandekayo kwelensi ende |
Fanisa isisekelo neresiphi — ukwakheka kwesikhumba ezingeni lezimbobo akusho lutho ku-24mm ebanzi lapho ubuso bungamaphikseli angamashumi ayisishiyagalombili ubude — bese uyakwengeza. Imininingwane esekelayo yi — ubufakazi bokwakheka bokuthi ilensi yakuqopha ngempela lokhu. Izimbobo zesikhumba ezibonakalayo nezinwele ezincane zobuso. Ukwelukwa kwehembe lelineni. Uthuli emsebeni wokukhanya. Ukuklwejeka esicathulweni sesikhumba. Izinwele ezindizayo ngokumelene nengemuva elikhanyayo. Ezimbili noma ezintathu zanele, futhi ziyona mehluko phakathi kokwakhiwa komuntu nesithombe esithwetshuliwe somuntu.
06Imibukiso enezingqimba: ingaphambili, iphakathi, ingemuva
Uma umbukiso unokujula, chaza izindiza ezintathu ngokwehlukana nangokulandelana. Leli yisu elisebenza kakhulu ezithombeni okufanele zizwakale njengezindawo kunokuba yizingemuva, futhi cishe akekho okwenzayo. Ngaphandle kwalo uthola ukubaphansi: konke okubalulwa umyalo kuwela ebangeni elicishe lifane, kuncintisana ngendawo efanayo ephakathi kohlaka. Ukubalula izindiza kunika imodeli isabelo sokujula esingasichitha.
In the foreground, a cracked concrete ledge with a half-empty
paper coffee cup and a scattering of cigarette ash, slightly out
of focus. In the midground, a bicycle courier in a yellow rain
jacket wheels her bike across a wet plaza, caught mid-stride. In
the background, a glass office tower dissolves into low fog.
Shot on a 50mm lens at f/2.8, flat overcast light.
Izilungiso ezimbili. Beka into engaphambili ngamabomu engagxilile — izithombe zangempela zemibukiso ejulile cishe njalo zinayo eyodwa, futhi ukungabikho kwayo kuluphawu olujwayelekile. Futhi sho ukuthi isihloko sikuyiphi indiza: uma umlethi wamabhayisikili eyiphuzu lesithombe ufanele abe ephakathi, uma kungenjalo umbhoshongo nenkomishi yekhofi kuzolwa naye ngohlaka.
07Imibala eqondile: amakhodi e-hex, anamathiselwe ezintweni
Emsebenzini webhrendi, ekuqhubekeni kwezingubo, noma noma ikuphi lapho u-"red" enganele, bhala ikhodi ye-hex ngokuthi color #RRGGBB bese uyinamathisela entweni ebalulwe ngegama:
...wearing a heavy wool peacoat, color #8B0000, over a cream
turtleneck, against a studio backdrop, color #1E3A5F.
...#8B0000, #1E3A5F, dark red and navy color palette, moody...
Imithetho emithathu esebenzayo. Sebenzisa i-hex emibaleni emibili noma emithathu ebalulekile ngempela bese usebenzisa amagama ajwayelekile kokunye — umyalo onamakhodi ayisithupha e-hex uchitha ingxenye enkulu yesabelo sawo emachashazini opende. Bopha ikhodi entweni enamaphethelo, ungalokothi uyibophe "kukukhanya" noma "kumoya". Futhi khumbula ukuthi icacisa umbala wento, hhayi inani lokugcina lephikseli: ijazi elingu-#8B0000 ekukhanyeni okuvela ngemuva kwehora legolide lakhiwa lifudumele kunechashazi, ncamashi njengoba kungenzeka empilweni.
08Ukukhetha isilinganiso sohlangothi
Kunezilinganiso ezinhlanu ezitholakalayo futhi esifanele imvamisa sishiwo yisihloko, kodwa isicelo esicacile esivela kuwe sinqoba njalo — cela isithombe sangemuva sefoni bese uthola u-tall kungakhathaliseki ukuthi okuqukethwe kusho ini.
| Isilinganiso | Sifanele | Kungani |
|---|---|---|
| square | Okuzenzakalelayo. Izinto ezizodwa eziphakathi, izithonjana, ukwakheka okulinganayo, okuthunyelwe kwezokuxhumana | Awukho umgqa obusayo, ngakho akukho okusikwa ukuchema kohlaka |
| portrait | Abantu, umzimba ogcwele noma uhhafu; izinto ezinde njengemibhoshongo, izihlahla, amabhodlela | Kunika umuntu omile indawo kusuka ekhanda kuya ezinyaweni ngaphandle kokumnciphisa |
| landscape | Izwe, ingaphakathi, amaqembu abantu, izithombe zomkhiqizo emongweni | Indawo evundlile yokuba indawo ibe yindawo ngempela |
| wide | Izinhlaka zebhayisikobho, izithombe zefilimu, amabhanela, izithombe zesihloko | Inketho ebanzi kunazo zonke; isikhala esingenalutho esiqinile, ukuzwakala kwe-anamorphic |
| tall | Amaphosta, izembozo zezincwadi, izithombe zangemuva zefoni, amafomethi aqondile ezokuxhumana | Ide kune-portrait; ishiya isikhala esenziwe ngamabomu sombhalo noma isibhakabhaka |
Ukwenza lokhu kabi kulimaza kakhulu kunokubukeka kwakho. Umuntu omile ogcwele ohlakeni lwe-wide akakuniki umuntu omncane ohlakeni olukhulu — imvamisa ukunika osikiwe, ngoba imodeli igcwalisa ububanzi bese ikhanda noma izinyawo kuphuma ohlakeni. Fanisa uhlaka nomgqa obusayo wesihloko bese izikhalazo eziningi zokwakheka ziyanyamalala.
Izindlela ezimbili zokuziphatha ezifanele zikhunjulwe: ukuhlela kugcina isilinganiso sohlangothi sesithombe somthombo ngokuzenzakalelayo, ngakho awukwazi ukushintsha uhlaka ngokuhlela, kanti ukuhlanganisa kukuvumela ukukhetha ikhanvasi entsha — ngokuvamile yiyona ndlela ehlanzeke kakhulu yokushintsha isimo sento osuvele unayo.
09Amaphethini amabi: yini okufanele uyiyeke
Amathokheni ekhwalithi: "masterpiece, best quality, 8k, ultra detailed"
Lezi kwakuyizibambo ezatholwa umphakathi zokusabalalisa okuthile kokuqeqeshwa okudala, lapho imishwana ethile yamagama yayihambisana nemithombo ethile yezithombe. Azidluli. Lapha zifundwa njengamagama esiNgisi avamile, azinikeli lutho, futhi zidla ingxenye yesabelo samagama osuvele utsheliwe ukuthi silinganiselwe. Okuthatha indawo yazo ukusho ukuthi yiziphi imininingwane oyifunayo: hhayi "ultra detailed" kodwa "visible weave in the coat fabric and fine stubble along the jaw" — inhloso efanayo, ivezwe njengento imodeli engayakha.
Uhlelo lwesisindo: (word:1.3), [word], {{word}}
Aluhlukaniswa. Amabakithi, amabakithi anamakhona, amakholoni nezinombolo kufika njengombhalo ngqo futhi, uma kukhona, kwenza umusho ozungezile ube nzima ukufundeka. Uma ufuna ukugcizelela okwengeziwe entweni ethile, sebenzisa izikhwepha ezimbili ezisebenzayo: yisuse uyibeke kuqala emyalweni, futhi uyinike amagama amaningi. "A crimson silk scarf, its fringed ends lifting in the wind" isinda kuno-"(red scarf:1.4)" ngayo yonke indlela ebalulekile.
Imiyalo ephikisanayo
Ukuxuba iziyalezo ezingahambisani akukuhlanganisi; kukhiqiza isilinganiso esingazinzile esivamise ukungagculisi nakunye. Abavame ukona:
- Uhlobo — "photorealistic" kanye "nokuthi watercolor"; "oil painting" kanye "no-shot on 85mm". Khetha okukodwa uzibophezele.
- Uhlaka — "wide establishing shot" kanye "ne-extreme close-up on her eyes"; "full body" kanye "ne-detailed facial pores".
- Ukukhanya — "harsh midday sun" kanye "ne-soft golden hour glow"; "low key, deep shadows" kanye "ne-bright and airy".
- Isikhathi — "1970s film aesthetic" kanye "ne-modern minimalist" kanye "ne-cyberpunk".
- Ukunqwabelanisa izitayela — abadwebi abane noma izinhlobo ezine kanyekanye. Izinkomba ezimbili ezihlobene zingasebenza; ezine ziyachitheka zibe yinto evamile.
- Ukusabalala — "rustic, weathered, ancient, timeworn wooden door" ngumqondo owodwa obhalwe kane, futhi ukuphindaphinda ngakunye kuhlambulula okunye komyalo.
10Awukho umyalo ophikisayo ekusukeni embhalweni uya esithombeni
Leli yiphethini elibi elibamba kanzima kakhulu abasebenzisi abanolwazi, ngoba ezitakini ezindala umyalo ophikisayo wawuyingxenye eyodwa yobuciko.
| Umsebenzi | Umyalo ophikisayo | Yini okumele uyenze |
|---|---|---|
| Umbhalo uye esithombeni (ukwakha) | Awusekelwa | Chaza kuphela lokho okufunayo. Ungalokothi ubhale "no X". |
| Ukuhlela isithombe esisodwa | Uyasekelwa | Beka izinto ezingafunwa lapho, ugcine isiyalezo siphozitha. |
| Ukuhlanganisa ezimbili noma ngaphezulu | Uyasekelwa | Kuyafana — izinto ezingafunwa kokuphikisayo, izindima kokuphozitha. |
Ekwakheni umthetho uqinile futhi ushaya ngokuphambene nomqondo: ukubhala "no text, no watermark, no extra fingers" emyalweni kwenza lezo zinto zibe nakho kakhulu ukwenzeka, hhayi kancane. Imodeli ifunda "no text" njengomusho oqukethe umqondo othi text; ukuphika kuluphawu olubuthaka kanti ibizo lolo oluqinile. Empeleni ucele umbhalo. Njalo kusho ngendlela ephozitha esikhundleni salokho:
| Esikhundleni sika- | Bhala |
|---|---|
| no background clutter | a clean seamless gray backdrop |
| not blurry | tack sharp on the eyes, f/8 |
| no other people | alone in an empty street at dawn |
| no modern objects | a 1930s kitchen with enamelware and a cast-iron stove |
| no cartoon style | shot on Kodak Portra 400, 85mm, natural skin texture |
Ekuhleleni nasekuhlanganiseni umyalo ophikisayo ukhona futhi kufanele uwusebenzise — kodwa ngezinto ezingafunwa, hhayi ngokuqukethwe. Okufanele: uphawu lwamanzi, umbhalo ongaphezulu, imilenze ephindaphindiwe, izandla ezisontekile, imisebe yokubukhali ngokweqile, ukuvela kwamabhloko e-JPEG, imigqa yomthungo emphethweni wokuhlanganisa. Okungafanele: noma yini emayelana nokuthi isithombe singesani, okufanele kube sesiyalezweni.
11Ukufunda umyalo i-Shannon ekubhalela wona
Nakhu okwenza lo mhlahlandlela usebenziseke ngokushesha kunokuba ube ngowethiyori. Akudingeki uwubhale wena umyalo wokwakha. Ubuza ngolimi olujwayelekile, nganoma yiluphi ulimi; i-Shannon ibona ukuthi kuceliwe isithombe, ibhale umyalo wokwakha wesiNgisi, bese ikubonisa wona ekhadini lokuqinisekisa ngaphambi kokuba kwakhiwe okuthile. Akukho okwakhiwayo futhi akukho okukhokhiswayo uze uqinisekise.
Ngakho ikhono elikhokha kakhulu akusikho ukubhala imiyalo kusukela ekuqaleni — ukufunda lowo osekhadini. Landela lolu hlu:
- Ingabe isihloko sicacile? Uma ikhadi lithi "a person" noma "a woman", uyithole intuthuko enkulu kunazo zonke etholakalayo. Sinike iminyaka, isakhiwo, isici.
- Ingabe isihloko sikuqala? Uma umyalo uvula ngesitayela noma ngekhamera, cela ukuba uhole ngesihloko esikhundleni salokho.
- Ingabe usebangeni? Ngaphansi kwamagama angu-30, cela ukucaciswa okwengeziwe kokukhanya nendawo. Ngaphezu kuka-100, cela ukuba unqunywe.
- Ingabe ukhona umthombo wokukhanya? Hhayi igama lomoya — ukukhanya kwangempela: ifasitela, isibani somgwaqo, u-flash, isibhakabhaka esiguqubele, i-softbox.
- Kukhona okuphikisanayo? Bheka izinhlobo ezimbili, izinhlaka ezimbili, noma izimo ezimbili zokukhanya emyalweni ofanayo.
- Kukhona ukuphika okushushunjiswe ngokuthula? Uma ikhadi liqukethe u-"without", u-"no", noma u-"avoid", lowo ngumyalo wombhalo-uye-esithombeni ocela leyo nto awubalulayo.
- Ingabe isilinganiso silungele isihloko? Umuntu omile ogcwele ku-
widekungukungahambisani okujwayelekile.
Izilungiso zenziwa ngengxoxo: "make her early thirties with a scar through one eyebrow, shoot it on Portra at 85mm, and use portrait" kuyisiyalezo esihle ngokuphelele. Uthola ikhadi elisha bese uqinisekisa lelo esikhundleni salelo. Ayikho inhlawulo yokuphindaphinda, ngoba akukho okusakhiwe.
Futhi i-Shannon ayakhi ingacelwanga. Ukubingelela, umbuzo, noma isithombe osinamathiselayo ukuze uxoxe ngaso akuqalisi ukwakha. Ukucela i-Shannon ukuba ibuke isithombe — isichaze, isifunde, isihlaziye — kuwumsebenzi ovamile wombono, hhayi ukwakha, futhi akubizi lutho olwengeziwe.
12Ubuciko bushintsha kanjani ekuhleleni nasekuhlanganiseni
Imithetho engenhla ibhalelwe ukwakha. Izindlela ezimbili ezisondelene zishintsha isimo somyalo.
Ukuhlela isithombe esisodwa
Umyalo wokuhlela uyisiyalezo, hhayi incazelo: sho ukuthi yini eshintshayo neyini ehlalayo. Ukubalula izingxenye ezingashintshi kuyisikhwepha esikhulu kunazo zonke sekhwalithi emkhiqizweni — "Change the background to a rainy night city street with neon reflections, keep the woman exactly as she is" kubambelela ndawonye lapho isicelo esifanayo ngaphandle kwalowo musho sidweba kabusha ubuso bakhe ngokuthula. Cacisa ngento oyibophayo: gcina ubuso, ukuma, izinwele nezingubo kufana ncamashi. Kusetshenziswa isithombe esisodwa somthombo ncamashi, futhi okukhishwayo kugcina isilinganiso sohlangothi saso.
Ukuhlanganisa izithombe ezimbili noma ngaphezulu
Noma nini lapho izinto zivela esithombeni esingaphezu kwesisodwa — ukudlulisa izingubo, into endaweni, abantu ababili ohlakeni olulodwa, umkhiqizo emuntwini, isitayela sesithombe esisodwa nesihloko sesinye — umyalo kumele ubalule indima yesithombe ngasinye, ngokulandelana: "The woman from the first image wearing the red knitted sweater from the second image, standing in the same park." Isithombe sokuqala yisihloko esiyinhloko noma indawo; izithombe ezilandelayo zinikeza izingxenye ezilethwayo. Ungahlanganisi isicelo sezithombe eziningi sibe umthombo owodwa kanye nencazelo yombhalo — ukuhlanganisa izithombe zangempela yikho okulondoloza indlela lezo zinto ezibukeka ngayo ngempela.
13Ukubhala imiyalo ngaphandle kokuvinjelwa: sho lokho okushoyo
Asikho isihlungi sokuqukethwe endleleni yokwakha futhi alukho ungqimba lokwenqaba kwimodeli ebhala umyalo. Yiyona nhloso yonke yalo mkhiqizo: izikhiqizi ezihlungiwe zenqaba njalo umsebenzi osemthethweni — izithombe zobuciko ezinqunu, imidwebo yezokwelapha neyobufakazi bomthetho, izithombe zocwaningo lokuphepha, udlame ezinganekwaneni, ukugcona ezombusazwe, umsebenzi webhrendi nomfanekiso womuntu owenziwa ngumnikazi wamalungelo. Uhhafu wobuciko bomyalo kumapulatifomu ahlungiwe ubunjiniyela bamazwi athambile, futhi amazwi athambile ayisitha sesithombe esinembile.
Ngakho kukhona okunye okumele ukuhlole ekhadini: ukuthi lichaza isihloko sangempela ngamagama acacile kunolimi oluzivikelayo abantu abalufunda kumathuluzi ahlungiwe. Umdwebo wobufakazi bomthetho wenxeba kufanele usho ukuthi inxeba liyini. Ulimi oluzivikelayo lukhiqiza isithombe esibi yonke indawo, ngoba imodeli yakha lokho amagama akushoyo.
Amandla nokwahlulela kuyizinto ezimbili ezehlukene. I-Shannon Lab LLC iqhuba lokhu ngaphansi kweNqubomgomo Yokusebenzisa Ngokuzibophezela eshicilelwe: ukwakha okungahlungiwe kuyithuluzi lobungcweti, futhi imvume, amalungelo nomthetho konke kusasebenza kulokho okwenzayo.
Uhlu lokuhlola lwekhasi elilodwa
- Umbhalo ngemisho ephelele, hhayi amathegi ahlukaniswe ngokhefana.
- Amagama angu-30–80. Susa noma yini engayishintsha isithombe.
- Isihloko → isenzo → isitayela → indawo nokukhanya → ubuchwepheshe.
- Isihloko esicacile: iminyaka, isici esisodwa sesakhiwo, ukungapheleli okukodwa.
- Izithombe: bala ikhamera, ilensi, nefilimu noma umthombo wokukhanya.
- Imibukiso ejulile: ingaphambili, iphakathi, ingemuva, kubalulwe ngokwehlukana.
- Imibala eqondile njenge-
color #RRGGBBeboshelwe entweni ebalulwe ngegama. - Khetha isilinganiso ngomgqa obusayo wesihloko.
- Ukwakheka okusekelayo okubili noma okuthathu, kufaniswe nohlaka.
- Awekho amathokheni ekhwalithi, alukho uhlelo lwesisindo, ayikho imiyalo ephikisanayo.
- Akukho kuphika ekusukeni embhalweni uya esithombeni. Ukuphika kungokokuhlela nokuhlanganisa, futhi kungokwezinto ezingafunwa kuphela.
14Ingena kuphi i-API
Iphuzu elilodwa lokunemba kubafundi abafika bevela ekusesheni i-API: ukwakha, ukuhlela nokuhlanganisa izithombe kuhlala ohlelweni lwengxoxo lwe-Shannon, hhayi ku-API ye-/v1. I-API ikhonza umbhalo nombono — ingakwazi ukufunda, ukuchaza nokuhlaziya izithombe ozithumelayo kuzo zontathu izinhlobo (/v1/chat/completions, /v1/messages, /v1/responses, zonke zisakaza) — kodwa ayizakhi. Uma wakha uhlelo oludinga ukuqonda izithombe, idokhumentheshini ye-API yiyona ndawo yokuqala. Uma ufuna ukwenza izithombe, lokho kusohlelweni lwengxoxo, futhi konke okukulo mhlahlandlela kusebenza khona. Izithombe zakhiwa kuqoqo lwethu lwama-GPU.
15Imibuzo ebuzwa njalo
Kufanele ube mude kangakanani umyalo wesithombe se-AI?
Hlosa amagama angu-30 kuya kwangu-80 futhi ungadluli kakhulu ku-100. Ngaphansi kwamagama angu-30 ushiya ukwakheka, ukukhanya nohlaka kunhlanhla. Ngaphesheya kwamagama angu-100 imisho elandelayo iqala ukuncintisana neyokuqala futhi imininingwane iqala ukuphela kunokunqwabelana.
Kufanele ngibhale amathegi amagama noma imisho ephelele?
Imisho ephelele. I-Image Model ye-Shannon ifunda umyalo ngendlela imodeli yolimi efunda ngayo umbhalo, ngakho uhlelo lolimi luthwala ulwazi. "A woman in her 30s at a rain-soaked Tokyo crosswalk at night" ikhiqiza umbukiso ohlangene; "woman, 30s, Tokyo, rain, night" ikhiqiza inqwaba yezimpawu ezixekethile ezingenabo ubudlelwano phakathi kwazo.
Ngingawusebenzisa umyalo ophikisayo ekusukeni embhalweni ngiye esithombeni?
Cha. Ukusuka embhalweni uye esithombeni ku-Shannon akwamukeli myalo ophikisayo, ngakho chaza kuphela lokho okufunayo futhi ungalokothi ubhale "no X" emyalweni uqobo — ukubalula into kuvame ukuyibiza. Ukuhlela izithombe nokuhlanganisa izithombe kuyawusekela umyalo ophikisayo, futhi yileyo ndawo efanele yezinto ezingafunwa njengezimpawu zamanzi, umbhalo ongaphezulu noma imilenze ephindaphindiwe.
Ingabe amathokheni afana no-"masterpiece, best quality, 8k" ayasithuthukisa isithombe?
Cha. Lawo ayeyizibambo zesizukulwane esidala samamodeli futhi awanalo ulwazi lapha — afundwa nje njengamagama, futhi achitha ingxenye yesabelo sakho samagama angu-30–80 engasho lutho. Uhlelo lwesisindo olunjenge-(word:1.3) nalo aluhlukaniswa; lufika njengombhalo ngqo. Kunalokho chaza imininingwane yangempela oyifunayo.
Ngiwuthola kanjani umbala oqondile?
Bhala ikhodi ye-hex ngokuthi color #RRGGBB bese uyinamathisela entweni ethile ebalulwe ngegama, isibonelo "a wool jacket, color #8B0000". Ikhodi ye-hex entanta ingenanto eyibophayo ayinalutho engabopheka kulo. Yisebenzise emibaleni emibili noma emithathu ebalulekile ngempela bese uchaza okunye ngamagama.
Ingabe ngingakwakha izithombe nge-API ye-Shannon?
Cha. Ukwakha, ukuhlela nokuhlanganisa izithombe kwenzeka ohlelweni lwengxoxo lwe-Shannon. I-API ye-/v1 ikhonza umbhalo nombono — ingakwazi ukufunda nokuhlaziya izithombe ozithumelayo, kodwa ayizakhi. Konke okukulo mhlahlandlela kusebenza ohlelweni lwengxoxo.
Yizame emyalweni wangempela
Buza nganoma yiluphi ulimi. Funda ikhadi. Lilungise. Bese uqinisekisa.
Vula I-Shannon Chat Olunye UcwaningoAkukho okwakhiwayo, futhi akukho okukhokhiswayo, uze uqinisekise umyalo
Okuhlobene ongakufunda: Ukwakha izithombe nge-AI okungavinjelwe · Ukuhlela izithombe nge-AI okungavinjelwe · INqubomgomo Yokusebenzisa Ngokuzibophezela · Idokhumentheshini ye-API · Lonke ucwaningo lwe-Shannon. Iseluleko lapha sithathwe ekuhloleni kwethu i-Image Model ye-Shannon; iseluleko esabhalelwa izitaki ze-diffusion ezindala imvamisa asidluli.