Icyitegererezo cyo Gutandukanya
Moderi yatinze itanga amashusho mugukoresha uburyo bwo gukwirakwiza ahantu hihishe inyuma aho kuba pigiseli mbisi, kugabanya ibiciro bya compte.
Incamake
They are the engine behind Stable Diffusion and most modern open-source image generators.
Kwibira cyane
Moderi isanzwe yo gukwirakwiza yiga guhindura inzira yo gusakuza: itangirira ku rusaku rwiza kandi igenda ihinduka ishusho. Gukora ibi kuri pigiseli birahenze kuko ishusho ya 512x512 ifite agaciro k'ibihumbi magana. Gukwirakwiza bidatinze, byatangijwe na Rombach na bagenzi be mu 2022, ubanza gukoresha autoencoder (VAE) byateganijwe mbere yo guhuza ishusho mumashanyarazi mato mato (akenshi 64x64x4, hafi 48x ntoya). Ikwirakwizwa U-Net noneho ryiga kwerekana imbere muri uwo mwanya uhishe, uyobowe ninyandiko ukoresheje guhuza ibitekerezo. Kurangiza, decodeur ya VAE yubaka pigiseli yuzuye. Uku kwikuramo gushishoza kugumisha amakuru asobanutse mugihe utaye ibisobanuro bidasobanutse, bigatuma ibisekuru byujuje ubuziranenge bishoboka kubakoresha GPUs.
Ubushishozi
Amayeri y'ingenzi ni ugutandukanya kwikuramo gushishoza no kwerekana imiterere. VAE ikora ibisobanuro byinshi bya pigiseli irambuye inshuro imwe, kandi U-Net yerekana gusa urugero rwo hasi rwihishwa rwo gukwirakwiza. Imyandikire itondekanya yatewe binyuze murwego rwo guhuza ibitekerezo, aho U-Net igaragara ahantu hatabira ibimenyetso byerekana kuva kodegisi yinyandiko nka CLIP. Kuberako ibyihishe bigera hafi kuri 48 kurenza pigiseli, buri ntambwe yerekana ko ihendutse cyane mububiko ndetse na FLOPs.
Ingaruka z'Ingamba
Umuvuduko n'igipimo
AI igaragara irashobora gukora igenzura, gutahura, no gutondekanya imirimo kurwego.
Build choices
Amakipe arema arashobora prototype ibitekerezo byihuse hamwe nintoki nkeya.
Itsinda hamwe nakazi
Ibikorwa birashobora gukoresha amashusho nibimenyetso bya videwo byari bigoye gutunganya.
Kazoza ka Moderi Yatinze
Gukwirakwiza bidatinze bigenda byiyongera kurenza amashusho muri videwo (Video itajegajega itajegajega), umutungo wa 3D, hamwe na ecran ya majwi, byose bikoresha compress imwe hanyuma noneho-denoise. Ubushakashatsi burimo gutera intambwe nkeya yo gutoranya hakoreshejwe uburyo bwo gutandukanya no kwerekana imiterere, VAEs nziza ibika inyandiko nziza nisura nziza, hamwe nibisobanuro bikosorwa nkibiri muri Stable Diffusion 3 igorora inzira yibisekuru kubisubizo byihuse, bikaze.
Gushyira mu bikorwa Isi
Diffusion ihamye itanga ibihangano hamwe nigishushanyo mbonera uhereye kumyandiko isaba umuguzi umwe GPU
Adobe na Canva bihindura inyandiko-ku-shusho hamwe na generator-yuzuza ibintu byubatswe ku mugongo wihishe inyuma
Sitidiyo yimikino itanga amakarita yimiterere, sprite, nibidukikije byubuhanzi kugirango byihute mbere yumusaruro
Amatsinda-shusho hamwe namakipe yo kwamamaza arema ibicuruzwa biranga ibicuruzwa no kwerekana amashusho adafite ifoto
Ingaruka & Kurinda
Uburenganzira bwishusho hamwe no kwemererwa birashobora guhinduka ibyago byemewe n'amategeko niba ibimenyetso bidasobanutse.
Imikorere yicyitegererezo irashobora gutandukana kumurika, demografiya, nibidukikije.
Ibyiza byibinyoma birashobora kutamenyekana keretse niba ibyiringiro byateganijwe bikurikiranwa.
Igishushanyo mbonera
Sobanura ibipimo byo kwemererwa kugiciro, kwibutsa, nibiciro byamakosa.
Gerageza hamwe namakuru ajyanye nuburyo nyabwo bwo gukora.
Ongeraho isubiramo ryabantu kubwizere buke cyangwa guhanura cyane.
Kurikirana icyitegererezo cya drift hanyuma uhindurwe nyuma ya kamera cyangwa dataset ihinduka.
Komeza Ubushakashatsi
Free newsletter
Get the daily AI briefing
Three verified AI stories every weekday morning, written in plain English. Free forever, no ads.
One email each weekday. Unsubscribe in one click. We never sell or share your address.
Test yourself
Take the Latent Diffusion Models quiz
Instant feedback on every answer, and a shareable certificate with a verifiable ID once you pass a course.
Support free AI education. AI Understanding is a 501(c)(3) nonprofit — no ads, no paywall, ever. Make a donation
Ubuyobozi bukurikira
Amashusho Yerekana Amashusho
Ibibazo bikunze kubazwa
What is Latent Diffusion Models?
Moderi yatinze itanga amashusho mugukoresha uburyo bwo gukwirakwiza ahantu hihishe inyuma aho kuba pigiseli mbisi, kugabanya ibiciro bya compte. Nibimoteri inyuma ya Stable Diffusion hamwe nibindi bigezweho bifungura-amasoko yerekana amashusho.
Nibihe bishya byingenzi byerekana ikwirakwizwa ryihishwa ugereranije na pigiseli-umwanya wo gukwirakwiza?
Gukwirakwiza bidatinze bigabanya amashusho mumwanya muto wihishe ukoresheje VAE, bityo denoising ihenze ibaho kubiciro bike ugereranije na pigiseli yuzuye.
Niki kintu kigabanya ishusho mumwanya wihishe hanyuma ikongera ikubaka nyuma?
VAE yateguwe mbere yerekana ishusho mumashanyarazi yihishe kandi decoder yayo ikubaka pigiseli yuzuye-impera.
Nigute inyandiko yinjizwa muri denoising U-Net mugukwirakwiza rwihishwa?
Gushira kumurongo bivuye kumyandikire ya kodegisi bigaburirwa murwego rwo guhuza ibitekerezo, kureka ibiranga umwanya bikitabira ikibazo.
Kuki gukorera mumwanya wihishe bigabanya ibiciro byo kubara?
Tegereza - impamvu yukuri nuko gride yihishe ari nto cyane (akenshi ~ 48x) kuruta ishusho ya pigiseli, bityo buri ntambwe yo kwerekana ikenera FLOPs na memoire nkeya.
Ni ubuhe buryo bukoreshwa cyane bwugurura-isoko yamamaye ikwirakwizwa ryihishwa?
Stable Diffusion, yasohotse mu 2022, yazanye ikwirakwizwa ryibikoresho byabaguzi no kwakirwa na benshi.