SDXL na mgbasa ozi Cascaded
SDXL bụ ihe ngosi dị elu nke ederede gaa na onyonyo Stability AI nke na-ejikọ onye nrụpụta ntọala siri ike yana onye nnụcha, ebe agbụ mgbasa ozi na-ejikọta ọtụtụ ụdị iji wuo onyonyo site na obere ruo na mkpebi dị elu.
Nchịkọta
Together they explain how modern open-source image generators hit photorealistic quality.
Ime miri emi
SDXL (Stable Diffusion XL) bụ ihe nlere mgbasa ozi dị ihe dị ka ijeri 3.5 nke na-emepụta onyonyo 1024 × 1024, na-amali elu karịa 512 × 512 izizi Stable Diffusion. Ọ na-eji koodu ntinye ederede abụọ (OpenCLIP ViT-bigG na CLIP ViT-L) maka nghọta ngwa ngwa karịa, gbakwunyere nha na nhazi ihe ọkụkụ ka ihe nlereanya ahụ mara mkpebi na nhazi nke ebumnuche. Ụgbọ mmiri SDXL dị ka pipeline nwere ọkwa abụọ: ụdị ntọala na-emepụta onyonyo a na-adịghị ahụ anya, mgbe ahụ ihe nchacha nhọrọ na-agbakwunye nkọwa dị mma na usoro mmechi ikpeazụ. Mgbasa mgbasa ozi bụ echiche sara mbara n'azụ nke a: kama otu ihe nlereanya na-eme ihe niile, ị na-ekenye obere ihe nlereanya nke na-emepụta onyonyo dị ala nke nwere ụdị mgbasa ozi nwere nnukwu mkpebi na-ebuli ya elu, onye ọ bụla zụrụ azụ maka ọkwa ya. Imagen Google mere ka usoro nlegharị anya gbasaa.
Nghọta nka nka
Ha abụọ na-arụ ọrụ n'usoro nkwuputa: malite site na mkpọtụ enweghị usoro wee buru amụma ma wepụ ya, nke ederede na-eduzi. SDXL na-arụ ọrụ n'ime oghere oghere abịakọrọ site na VAE, yabụ nkwuwa okwu dị ọnụ ala karịa ịrụ ọrụ na pikselụ raw. Onye nnụcha bụ ụdị ọkachamara dị iche iche nke na-ejikwa naanị usoro ikpeazụ, obere mkpọtụ. N'ime ezigbo cascade, ihe nlereanya ntọala na-ewepụta obere onyonyo, mgbe ahụ ụdị mgbasa ozi nwere oke ọnọdụ na-ebuli ya elu, nke ọ bụla nwere ọnọdụ na mmepụta mkpebi dị ala, na-ejikarị nkwalite mkpọtụ mkpọtụ ka ọ kwụsie ike.
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 SDXL na mgbasa ozi Cascaded
Ihe a na-eme na-aga n'ihu nrịgo ole na ole, ngwa ngwa yana ụlọ ejikọtara ọnụ. Uzo distillation dị ka SDXL Turbo na Model Consistency Model ebipụla ọgbọ ruo otu nzọụkwụ anọ. Ndị na-eme mgbanwe mgbanwe (dị ka ọ dị na Stable Diffusion 3 na FLUX) na-edochi anya azụ azụ U-Net, na njedebe njedebe na njedebe dị elu na-ebelata ịdabere na cascades doro anya. Na-atụ anya nchikota nke nchacha nke ọma, ntụgharị ederede ka mma, yana ezigbo oge onyonyo n'ime ngwaọrụ ka arụmọrụ na-aga n'ihu na-akawanye mma.
Mmejuputa n'ezie n'ụwa
Ịmepụta 1024x1024 ire ahịa na nka echiche ozugbo site na mkpali ederede na-enweghị ihe elu dị iche iche.
Iji SDXL base-plus-refiner pipeline iji gbakwunye nkọwa kpụ ọkụ n'ọnụ na ihu na textures na njakịrị ngwaahịa.
Na-agba ọsọ SDXL Turbo maka nlebanya onyonyo dị nso na ngwa imewe mmekọrịta
Ịmepụta cascade nke nwere nnukwu mkpebi iji tụgharịa sketị ndị dị ala ka ọ bụrụ ihe atụ dị elu.
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 SDXL and Cascaded Diffusion 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
Nlegharị anya Mgbasa Multi-Concept Tuning
Ajụjụ a na-ajụkarị
What is SDXL and Cascaded Diffusion?
SDXL bụ ihe ngosi dị elu nke ederede gaa na onyonyo Stability AI nke na-ejikọ onye nrụpụta ntọala siri ike yana onye nnụcha, ebe agbụ mgbasa ozi na-ejikọta ọtụtụ ụdị iji wuo onyonyo site na obere ruo na mkpebi dị elu. Ha jikọtara ọnụ na-akọwa ka ndị na-emepụta ihe onyonyo mepere emepe nke ọgbara ọhụrụ si metụta àgwà foto.
Kedu mkpebi obodo SDXL na-ewepụta onyonyo na, ma e jiri ya tụnyere nke izizi Stable Diffusion?
SDXL na-ewepụta onyonyo 1024 × 1024, mpaghara okpukpu anọ karịa nke mbụ Stable Diffusion's 512 × 512.
Kedu ọrụ nke ụdị nnụcha SDXL?
Onye na-anụcha ihe bụ ụdị ọkachamara dị iche nke etinyere na njedebe nke pipeline iji mee ka nkọwa zuru ezu mgbe ihe nrịbama ahụ mechara ihe oyiyi nzuzo.
Kedu ihe ntinye ederede SDXL na-eji?
SDXL na-eji ntinye ederede abụọ, OpenCLIP ViT-bigG na CLIP ViT-L, jikọtara maka nghọta ngwa ngwa karịa.
Gịnị bụ isi echiche nke cascaded mgbasa?
Chainịnụ mgbasa ozi agbasasịla jikọtara obere ihe nrụpụta ntọala nwere otu ụdị mgbasa ozi ma ọ bụ karịa nke ukwuu, nke ọ bụla dabara na nsonaazụ mkpebi dị ala gara aga.
Kedu ihe kpatara SDXL ji ekwu okwu n'ime oghere zoro ezo karịa ozugbo na pikselụ?
VAE na-echikota onyonyo n'ime oghere dị obere, yabụ usoro mgbasa ozi dị ọnụ ala karịa ịrụ ọrụ na pikselụ raw zuru oke.