Modèlu diffusion yu nëbbu
Modèlu diffusion yu nëbbu yi dañuy defar ay nataal ci def diffusion bi ci barab bu nëbbu buñ komprime ci barabu pixel yu ñor yi, di wàññi njëgu ordinatër bi.
Résumé
They are the engine behind Stable Diffusion and most modern open-source image generators.
Plongeur bu xóot
Benn xeetu diffusion buñ miin dafay jàng ni ñuy delloosi ab bruit: dafay tàmbalee ci bruit bu seer ba noppi di denoise ndànk-ndànk dem ba nekk nataal. Def lii ci pixel yi lu seer la ndax nataalu 512x512 amna téemeeri junni valeur. Diffusion latente, bi Rombach ak ay naataango dugal ci 2022, dafa njëkka jëfandikoo autoencodeur variationnel (VAE) buñ tàggat bu njëkk ngir kompresse ab nataal ci benn griy latente bu ndaw (dafay faral di nekk 64x64x4, lu tollu ci 48x lu gëna ndaw). Diffusion U-Net dafay jàng dindi bruit ci biir barab bu nëbbu bu kompact, di tegtal ci mbind jaaraleko ci cross-attention. Fi muy mujjee mooy decodeur VAE dafay defaraat pixel yu mat sëkk. Bii compression perceptuel dafay tëye leeral yi am solo ci wàllu semantik, ci noonu lañuy sànnee ay detay yuñu dul gis, suko defee ñu mëna defar ay kalite yu baax ci GPU yiy jëfandikoo.
Gis-gis xarala
Li gëna am solo mooy tàqale kompresioŋ perseptuel ak modelaasioŋ generatif. VAE dafay jëfandikoo pixel yu bari yi benn yoon, te U-Net dafay modele distribution latente bu gëna ndaw. Kondisioneeru mbind dañu koy dugal ci ay diisaay cross-attention, foofu la man-mani spatial yu U-Net di toppatoo embeddings token yu bawoo ci encodeur mbind bu melni CLIP. Ndax latents yi ñoo gëna ndaw pixel yi lu tollu ci 48 yoon, jéego bu nekk ci denoising moo gëna xéewale ci memory ak FLOPs yépp.
njeextalu pexe
Gaawaay ak yaatuwaay
Visual IA mën na otomatise saytu, gis ak etiketu liggéey ci eskaal.
Tabax tànneef
Ekipu kreatif yi mën nañu defar konsept yu gëna gaaw te duñu def lu bari ci loxo.
Ekip ak def liggéey
Liggéeyukaay yi mën nañu jëfandikoo siñaal nataal wala wideo yu jafewoon lool ci liggéey.
Ëlëgu xeetu diffusion yu nëbbu
Diffusion bu nëbbu mingi dem ba weesu nataal yi dem ci wideo (Diffusion wideo bu dëgër), ay mbiri 3D, ak ay spectrogram audio, ñoom ñépp di jëfandikoo benn rëset buy kompresse-ba noppi-denoise. Gëstu dafay puus ci jéego yu néew ci jël misaal jaaraleko ci distilaasioŋ ak xeetu lëkkaloo, VAE yu gëna baax yuy baña yàq mbind yu rafet ak kanam yi, ak formulaasioŋ yu baax yu melni ñooñu ci Diffusion Stable 3 yuy jubbanti yoon wi ñuy jaar ngir gëna gaaw, gëna ñaw.
Doxal ci àdduna dëgg
Diffusion bu dëgër buy defar ay artwork ak ay konsept ci ay laaj ci benn GPU konsomatër
Adobe ak Canva dañuy dundal mbind-ci-nataal ak man-mani fees yu defar ci kaw yaxu ndigg lu nëbbu
Istijoo yiy defar ay kàrtu texture, ay sprite, ak konsept art environmaa bi ngir gaawlu defar bi njëkk
Nataalu stock ak ekipu fësal njaay yi defar ay maquette produit ak nataali yëgle te duñu def benn sesioŋ foto
Risk yi ak balustrade yi
Yelleefi nataal ak nangu mën na nekk risku yoon sudee fi ñu bawoo leerul.
Performance model bi mën na wuute ci leeraay bi, demographie bi ak environmaa bi.
Njuumteg positive yi mën nañu dem te kenn duko seetlu fileek xool wuñu buntu wóolu sa bopp.
Roadmap ngir samp gi
Mandargal kritërium nangug njub, woowaat ak njëgu njuumte.
Saytu ak done yu méngoo ak anam yi ñuy liggéeyee dëgg.
Yokk jàngat nit ngir xam fu wóorul dara wala am njeexital yu rëy.
Toppal model drift bi nga baaxal ko ginaaw bi kamera bi wala done yi soppeekoo.
Weyal di banneexu
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
Gis bi ci topp
Modèlu diffusion wideo
Laaj yi ñuy faral di laaj
What is Latent Diffusion Models?
Modèlu diffusion yu nëbbu yi dañuy defar ay nataal ci def diffusion bi ci barab bu nëbbu buñ komprime ci barabu pixel yu ñor yi, di wàññi njëgu ordinatër bi. Ñoom ñooy motër bi ci ginaaw Stable Diffusion ak generatëri nataal yu bees yi.
Lan mooy xeetu diffusion bu nëbbu bi gëna am solo buñu ko méngale ak diffusion pixel-espace?
Diffusion Latent dafay kompresse nataal yi ci barab bu gëna ndaw bu nëbbu jëfandikoo VAE, kon denoising bu seer bi dafay am ci valeur yu néew lool ci pixel yu mat.
Ban cër mooy kompresse ab nataal ci espace bu nëbbu bi ba noppi defaraat ko ci ganaw?
VAE buñu njëkka tàggat dafay enkode nataal bi ci benn griy bu nëbbu bu dëgër, dekodeer bi dafay defaraat pixel yu mat sëkk ci njeexte li.
naka lañuy dugal mbind ci biir U-Net biy dindi bruit ci diffusion bu nëbbu?
Token yiñ dugal ci encodeur text dañu leen di dugal ci ay couche cross-attention, suko defee man-mani spatial yi mëna topp laaj bi.
Lan moo waral liggéey ci espace latent wàññi njëgu ordinatër?
Xaaral - sabab bu dëggu bi mooy griy bu nëbbu bi dafa gëna ndaw (dafay faral di nekk ~ 48x) ci nataalu pixel, kon jéego bu nekk bu denoising soxla FLOPs ak memory yu néew.
Ban xeetu open-source lañu gëna jëfandikoo ngir siiwal diffusion latente?
Stable Diffusion, biñu génne ci 2022, dafa indi diffusion bu nëbbu ci jumtukaayi konsomatër yi ak jëfandikoo gi ci mbooloo mi.