Ntuziaka Visual AI

Stable Diffusion

Stable Diffusion bụ ihe ngosi nke mepere emepe nke ederede gaa na onyonyo, nke Stability AI wepụtara na 2022, nke na-ewepụta foto site na iji nke nta nke nta wepụ mkpọtụ site na mmalite mmalite.

2 nkeji na-agụEmelitere ikpeazụ

Nchịkọta

N'ịbụ ndị mepere emepe ma nwee ike ịgba ọsọ na GPU ndị na-azụ ahịa, ọ kpalitere nnukwu ngwa ọrụ, ụda ọma na ngwa.

Ime miri emi

Ụdị mgbasa ozi na-amụta ịgbanwe usoro mkpọtụ. N'oge ọzụzụ, ezigbo onyonyo nwere mkpọtụ na-enweghị usoro agbakwunyere site na nzọụkwụ ruo mgbe ha guzoro ọtọ; ihe nlereanya na-amụta ịkọ na ibelata mkpọtụ ahụ. Iji mepụta, ọ na-amalite site na mkpọtụ dị ọcha wee na-ekwupụta ugboro ugboro ruo mgbe onyonyo nwere ọnụ ahịa pụtara, nke ederede gị na-eduzi. Stable Diffusion's isi arụmọrụ aghụghọ bụ akụkụ 'latent': kama ịrụ ọrụ na pikselụ nwere mkpebi zuru oke, ọ na-atụgharị onyonyo n'ime oghere dị ntakịrị site na iji variational autoencoder, na-eji nwayọ denoising ebe ahụ, wee dekọọ azụ na pikselụ. Nke a bụ ya mere ọ nwere ike ịgba ọsọ na GPU egwuregwu a na-ahụkarị karịa ebe data. Ihe ngbanwe ederede (CLIP na nsụgharị mbụ) na-atụgharị ngwa ngwa gị ka ọ bụrụ nduzi, U-Net na-ekwupụtakwa ya. Arọ ya mepere emepe nyeere ControlNet aka, ụda LoRA dị mma, yana ọtụtụ ngwaọrụ okike.

Nghọta nka nka

Stable Diffusion bụ ụdị mgbasa ozi zoro ezo. Ihe autoencoder na-ebelata onyonyo 512x512 ka ọ bụrụ grid kọmpat, na-ebelata mkpokọta n'ike. A zụrụ U-Net ka ọ buru amụma mkpọtụ agbakwunyere n'oge ọ bụla, dabere na ederede agbakwunyere site na nlebara anya. Ntuzi aka na-enweghị nkewa na-ahapụ gị ka ị kpọọ otu onyonyo siri gbasoo ngwa ngwa site na ịgwakọta amụma nwere ọnọdụ na enweghị ọnọdụ. Na ntinye, onye sampler (dị ka DDIM ma ọ bụ Euler) na-ewere usoro nkwubi okwu ahọpụtara; usoro ndị ọzọ na-apụtakarị nsonaazụ dị ọcha na ọnụ ahịa ọsọ.

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 Mgbasa Stable

Usoro gburugburu ebe obibi mepere emepe na-aga n'ihu: ihe owuwu ọhụrụ (gụnyere mgbasa ozi na-agbanwe agbanwe na usoro ole na ole ma ọ bụ ndị na-agbapụta agbaze) belata ọgbọ site n'ọtụtụ usoro ruo otu ma ọ bụ abụọ, na-enye ike imepụta ihe dị nso. Na-atụ anya ntụgharị ederede siri ike karị, nrube isi ngwa ngwa ka mma, yana ndezi onyonyo enweghị nkebi, gbakwunyere vidiyo na ndọtị 3D. Ọdụdọ mepere emepe ga-aga n'ihu na-eme ka egwu egwu pụrụ iche, mana ha na-emekwa ka arụmụka gbasara nkwenye data-ọzụzụ, fakes na watermarking, yabụ nchọta na ngwa ngwa ga-eto n'akụkụ ụdị.

Mmejuputa n'ezie n'ụwa

Ndị na-ese ihe na ndị na-enwe ntụrụndụ na-emepụta nka na ihe atụ n'ime mpaghara na GPU nke ha nwere ọmarịcha ụda LoRA.

Iji ControlNet iji gbochie ọgbọ nwere skeleton pose, maapụ omimi, ma ọ bụ sketị ihu maka nhazi ziri ezi.

Ịtinye na eserese iji dezie foto, wepụ ihe, ma ọ bụ gbasaa ihe nkiri gafere oke mbụ ya

Ụlọ ihe nkiri egwuregwu Indie na ndị na-emepụta ihe na-emepụta textures, bọọdụ ọnọdụ, na mgbanwe akụrụngwa ngwa ngwa na ọnụ ala.

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

1

Kọwaa ụkpụrụ nnabata maka nkenke, icheta, na ụgwọ njehie.

2

Nwalee na data dabara na ọnọdụ mmepụta n'ezie.

3

Tinye nyocha mmadụ maka obere obi ike ma ọ bụ amụma mmetụta dị elu.

4

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 Stable Diffusion quiz

Instant feedback on every answer, and a shareable certificate with a verifiable ID once you pass a course.

Malite ajụjụ

Support free AI education. AI Understanding is a 501(c)(3) nonprofit — no ads, no paywall, ever. Make a donation

Ntuziaka na-esote

Mgbasa vidiyo kwụsiri ike

Ajụjụ a na-ajụkarị

Kedu ihe bụ Stable Diffusion?

Stable Diffusion bụ ihe ngosi nke mepere emepe nke ederede gaa na onyonyo, nke Stability AI wepụtara na 2022, nke na-ewepụta foto site na iji nke nta nke nta wepụ mkpọtụ site na mmalite mmalite. N'ịbụ ndị mepere emepe ma nwee ike ịgba ọsọ na GPU ndị na-azụ ahịa, ọ kpalitere nnukwu ngwa ọrụ, ụda ọma na ngwa.

N'ọkwa dị elu, kedu ka ụdị mgbasa ozi si ewepụta onyonyo?

Ụdị mgbasa ozi na-amụta ịgbanwe usoro mkpọtụ: na-amalite site na mkpọtụ, ha na-ekwu okwu ugboro ugboro ruo mgbe onyonyo nwere njikọ pụtara.

Kedu ihe na-eme Stable Diffusion rụọ ọrụ nke ọma iji na-agba ọsọ na GPU ndị ahịa?

Stable Diffusion bụ ụdị mgbasa ozi zoro ezo: ihe autoencoder na-akpakọ onyonyo ka nkwubi okwu ahụ na-agba ọsọ na obere oghere.

Kedu ka ngwa ngwa ederede gị si emetụta onyonyo ewepụtara?

Ihe ntinye ederede na-atụgharị ngwa ngwa ka ọ bụrụ ntinye nke na-edozi U-Net site na nlebara anya, na-eduzi nsonaazụ ya.

Kedu ihe ntuziaka na-enweghị classifier na-ahapụ gị ijikwa?

Ntuzi aka na-enweghị nkewa na-agwakọta amụma nwere ọnọdụ yana enweghị ọnọdụ, na-ahapụ gị ka ị gbanwee nrubeisi ozugbo elu ma ọ bụ ala.

Kedu ihe kpatara Stable Diffusion ji kpalite nnukwu akụrụngwa dị ka ControlNet na LoRA?

Mepee ihe ọ̀tụ̀tụ̀ dị arọ na-eme ka ndị obodo na-emegharị ma gbatịa ihe nlereanya ahụ, na-emepụta ihe mgbakwunye dị ka ControlNet na ihe nkwụnye LoRA dị fechaa.