Stable Diffusion
Stable Diffusion xeetu bind-ci-nataal la bu ubbeeku, Stability AI moo ko genne ci 2022, muy defar ay nataal ci dindi ndànk-ndànk bruit ci barab bu ñu tàmbalee.
Résumé
Being open and runnable on consumer GPUs, it sparked a massive community of tools, fine-tunes, and apps.
Plongeur bu xóot
Modèle diffusion yi dañu jàng ni ñuy delloosi ab bruit. Ci diiru tàggat yaram, nataal dëgg yi dañuy yokk bruit bu bari jéego ak jéego ba ñuy nekk static; model bi dafay jàng xam luy waaja am ak dindi bruit boobu. Ngir defar, dafay tàmbali ci bruit bu sell ak dindi bruit yu bari ba keroog nataal bu méngoo feeñ, di tegtal sa bind. Li gëna am solo ci Stable Diffusion mooy wàllu 'latent' bi: ludul liggéey ci pixel yu mat sëkk, dafay kompresse nataal yi ci barab bu gëna ndaw bu nëbbu ci jëfandikoo autoencoder bu wuute, def fa denoising bu yeex, ba noppi dekodewaat ci pixel yi. Loolu moo tax mu mëna dox ci GPU gaming bu gëna nekk ci santu done. Encodeur text (CLIP ci version yu njëkk yi) dafay soppi sa prompt ci tegtal, U-Net mooy def denoising bi. Poids yu ubbeeku yi may nañu ControlNet, LoRA ak jumtukaayi sos lu bari.
Gis-gis xarala
Diffusion stabil xeetu diffusion bu nëbbu la. Benn autoencoder dafay wàññi nataalu 512x512 ci griy bu nëbbu bu dëgër, dagg xayma bi bu baax. U-Net dañu ko tàggat ngir mu mëna wax luy xew ci bruit bi ñuy yokk ci timestep bu nekk, lépp di aju ci mbind mi ñuy dugal ci cross-attention. Njàngale bu amul benn classifier daf lay may nga xool ni nataal bi di toppee li ñu la sant ci jaxase waxtaan yu am sart ak yu amul sart. Ci inference, sampler (lu melni DDIM wala Euler) dafay jël limu jéego yiñ tànn ngir dindi bruit; jéego yu bari dañuy tekki njariñ yu gëna set ci njëgu gaawaay.
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 diffusion bu dëgër
Ecosystem ubbeeku bi dafay wéy di gaaw: architecture yu bees yi (lu ci melni diffusion bu sukkandiko ci transformateur ak sampler yu néew wala distiye yu gëna gaaw) dañu dagg generation ci fukki-fukki jéego dem ci benn wala ñaar, loolu mooy tax ñu mëna defar ci jamono dëgg. Xaarandil rendu mbind bu gëna dëgër, gëna gaaw ci topp, ak soppali nataal bu jaar yoon, boole ci wideo ak extension 3D. Poid yu ubbeeku yi dina ñu wéy di gëna xëcc ay fine-tunes yu yam, waaye dina ñu gëna yokk waxtaan yi ci wàllu tàggat-done, deepfakes, ak watermarking, kon gis-gis ak jumtukaayi provenance dina ñu màgg ci wetu model yi.
Doxal ci àdduna dëgg
Artist yi ak hobbyist yi defar konsept art ak ay tegtal ci seen GPU bopp ak LoRA fine-tunes
Jëfandikoo ControlNet ngir tënk ab jamono bu am yax pose, kàrtu xóotaayu, wala jëmmal ci catu composition bu leer
Inpainting ak outpainting ngir soppali ay nataal, dindi ay mbir, wala yaatal seen bu weesu limu njëkka am
Istijoo yi ak defarkati jeu indie yi defar ay textures, tablo humeur ak ay variation ci anam wu gaaw te yomb
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 Stable 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
Gis bi ci topp
Diffusion wideo bu dëgër
Laaj yi ñuy faral di laaj
What is Stable Diffusion?
Stable Diffusion xeetu bind-ci-nataal la bu ubbeeku, Stability AI moo ko genne ci 2022, muy defar ay nataal ci dindi ndànk-ndànk bruit ci barab bu ñu tàmbalee. Nimu ubbeeku te mën nañu ko jëfandikoo ci GPU yi, dafa jur jumtukaay yu bari, ay fine-tune ak ay app.
Ci niveau bu kawe, naka la modelu diffusion di defaree ab nataal?
Modèle diffusion yi dañu jàng ni ñuy delloosi ab bruit: tàmbalee ci bruit, ñuy baamtu denoise ba keroog ñuy génne ab nataal bu lëkkaloo.
Lan moo tax Stable Diffusion mëna dox ci GPU konsomatër?
Diffusion bu dëgër xeetu diffusion bu nëbbu la: autoencoder dafay kompresse nataal yi suko defee denoising bu diis bi di daw ci barab bu nëbbu bu gëna ndaw.
Lan mooy njeextalu sa bind ci nataal biñ defar?
Encodeur text dafay soppi prompt bi ci ay embedding yuy indi U-Net ci cross-attention, di doxal resultaa bi.
Lan la tegtal bu amul klassifikatër di may nga doxal?
Njàngale bu amul klassifikatër dafay jaxase waxtaan yu am sart ak yu amul sart, loolu dina tax nga mëna yéeg wala wàcci topp gi ci saasi.
Lan moo waral tasaaroo bu dëgër jur jumtukaay yu bari yu melni ControlNet ak LoRA?
Poid yu ubbeeku yi may askan wi mu gëna defar ak yaatal model bi, defar ay yokk-yokk yu melni ControlNet ak adaptëru LoRA yu woyof.