Ụdị mgbasa ozi nzuzo
Ụdị mgbasa ozi ezoro ezo na-emepụta ihe oyiyi site na iji usoro mgbasa ozi n'ime oghere oghere abịakọrọ kama ịbụ pikselụ raw, na-ebelata ọnụ ahịa ọnụ.
Nchịkọta
They are the engine behind Stable Diffusion and most modern open-source image generators.
Ime miri emi
Ụdị mgbasa ozi ọkọlọtọ na-amụta ịgbanwe usoro mkpọtụ: ọ na-amalite site na ụda dị ọcha ma jiri nwayọọ nwayọọ na-atụgharị n'ime ihe oyiyi. Ime nke a ozugbo na pikselụ dị oke ọnụ n'ihi na onyonyo 512x512 nwere ọtụtụ narị puku ụkpụrụ. Mgbasa ozi nzuzo, nke Rombach na ndị ọrụ ibe webatara na 2022, na-ebu ụzọ were variational autoencoder (VAE) a zụrụ azụ iji pịnye onyonyo n'ime obere grid latent (na-abụkarị 64x64x4, ihe dị ka 48x pere mpe). U-Net mgbasa ozi ahụ na-amụta ịkatọ n'ime oghere nzuzo ahụ, nke ederede na-eduzi site na nlebara anya. N'ikpeazụ VAE decoder na-ewughachi pikselụ nwere mkpebi zuru oke. Mkpakọ nghọta a na-edobe ozi ahụ bara uru n'ụzọ zuru oke ma na-atụfu nkọwa zuru oke, na-eme ka ọgbọ dị elu nwee ike na GPU ndị ahịa.
Nghọta nka nka
Isi aghụghọ bụ ikewapụta mkpakọ nghọta na imepụta ụdị. VAE na-ejikwa nkọwapụta pikselụ dị elu otu ugboro, na U-Net na-egosipụta nkesa latent dị ala. A na-agbanye ntụnye ederede site na nleba anya nlebara anya, ebe njirimara oghere U-Net na-aga na ntinye akara sitere na ntinye ederede dị ka CLIP. N'ihi na latent dị ihe dị ka ugboro 48 pere mpe karịa pikselụ, usoro nkwuputa ọ bụla dị ọnụ ala karịa na ebe nchekwa yana FLOP.
Mmetụta atụmatụ
Ọsọ na ọnụ ọgụgụ
Visual AI nwere ike megharịa nyocha, nchọpụta na mkpado ọrụ n'ọtụtụ.
Mee nhọrọ
Otu ndị na-emepụta ihe nwere ike imepụta echiche ngwa ngwa site na ngbanwe akwụkwọ ntuziaka ole na ole.
Team na usoro ọrụ
Ọrụ nwere ike iji onyonyo na akara vidiyo siri ike ịhazi.
Ọdịnihu nke ụdị mgbasa ozi nzuzo
Mgbasa mgbasa ozi na-agbasa karịa onyonyo n'ime vidiyo (Stable Video Diffusion), akụ 3D, na spectrogram ọdịyo, ha niile na-eji otu nchịkọta mkpọ-mgbe ahụ-denoise uzommeputa. Nnyocha na-aga n'ihu na usoro nlele dị ole na ole site na ntugharị na ụdị nkwụsi ike, VAE ka mma nke na-echekwa ederede na ihu dị mma, yana usoro nhazigharị dị ka nke dị na Stable Diffusion 3 nke na-agbatị usoro ọgbọ maka ngwa ngwa na nsonaazụ dị nkọ.
Mmejuputa n'ezie n'ụwa
Stable Diffusion na-emepụta nka na atụmatụ echiche sitere na ederede kpaliri n'otu GPU ndị ahịa
Adobe na Canva na-eme ka ederede gaa na onyonyo yana njiri njupụta nke arụpụtara n'ọkpụkpụ azụ azụ azụ.
Ụlọ ihe nkiri egwuregwu na-emepụta mapụ udidi, sprites, na nka echiche gburugburu ebe obibi iji mee ka ọ dịwanye elu tupu mmepụta
Otu ndị na-ere ahịa na-ere ahịa na-emepụta ihe njakịrị ngwaahịa na ihe ngosi mgbasa ozi na-enweghị foto
Ihe ize ndụ & okporo ụzọ nche
Ikike onyonyo na nkwenye nwere ike bụrụ ihe egwu dị n'iwu ma ọ bụrụ na edoghị anya.
Ọrụ nlereanya nwere ike ịdịgasị iche n'ofe ọkụ, igwe mmadụ, na gburugburu.
Enwere ike ghara ịhụ ihe dị mma ma ọ bụrụ na enyochaghị oke ntụkwasị obi.
Map mmejuputa
Kọwaa ụkpụrụ nnabata maka nkenke, icheta, na ụgwọ njehie.
Nwalee na data dabara na ọnọdụ mmepụta n'ezie.
Tinye nyocha mmadụ maka obere obi ike ma ọ bụ amụma mmetụta dị elu.
Sochie ihe nlere anya wee megharịa ka emechara mgbanwe igwefoto ma ọ bụ dataset.
Nọgide na-eme nchọpụta
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
Ntuziaka na-esote
Ụdị mgbasa ozi vidiyo
Ajụjụ a na-ajụkarị
What is Latent Diffusion Models?
Ụdị mgbasa ozi ezoro ezo na-emepụta ihe oyiyi site na iji usoro mgbasa ozi n'ime oghere oghere abịakọrọ kama ịbụ pikselụ raw, na-ebelata ọnụ ahịa ọnụ. Ha bụ injin dị n'azụ Stable Diffusion na ọtụtụ ndị na-emepụta ihe onyonyo mepere emepe nke ọgbara ọhụrụ.
Kedu ihe bụ isi ihe ọhụrụ nke ụdị mgbasa ozi zoro ezo ma e jiri ya tụnyere pixel-space mgbasa?
Mgbasa mgbasa ozi na-ejikọta onyonyo n'ime oghere dị nta na-eji VAE, yabụ nkwubi okwu dị oke ọnụ na-eme na ụkpụrụ dị ntakịrị karịa pikselụ zuru oke.
Kedu akụrụngwa na-akpakọ onyonyo n'ime oghere zoro ezo wee wughachi ya ma emechaa?
VAE a zụrụ azụ na-etinye ihe onyonyo a ka ọ bụrụ grid kọmpat yana ihe ndozi ya na-ewughachi pikselụ nwere mkpebi zuru oke na njedebe.
Kedu ka esi agbaba ederede n'ime U-Net na-ekwupụta na mgbasa ozi nzuzo?
A na-etinye ntinye akara sitere na koodu ntinye ederede n'ime ọkwa nlebara anya, na-ekwe ka njirimara ohere banye ozugbo.
Kedu ihe kpatara ịrụ ọrụ na oghere nzuzo ji ebelata ọnụ ahịa mgbakọ?
Chere - ihe kpatara ya bụ na grid latent dị ntakịrị (mgbe ọ bụla ~ 48x) karịa onyonyo pixel, yabụ nzọụkwụ ọ bụla na-agbagha chọrọ FLOP na ebe nchekwa dị ole na ole.
Kedu ihe nlere mepere emepe nke a na-ejikarị eme ihe na-agbasa mgbasa ozi nzuzo?
Stable Diffusion, ewepụtara na 2022, wetara mgbasa ozi latent na ngwaike ndị ahịa yana nkuchi oke.