Modèlu deggoo bu nëbbu
Modèlu Consistence Latent (LCMs) xarala la buy may generatëri nataali diffusion ñu defar nataal yu baax ci benn wala ñeenti jéego, duñu def ay fukki-fukki nataal yuñy faral di def.
Résumé
They make near-real-time, interactive image generation practical even on modest hardware.
Plongeur bu xóot
Royuwaayi diffusion yu nëbbu yuñ miin lu melni diffusion stabil dañuy tàmbalee ci bruit ak denoise ñuy baamtu, ñuy faral di soxla 20 ba 50 jàngat reso ngir defar benn nataal, te daa yeex. LCMs, Luo ak ay naataango dugal ko ci 2023, jëfandikoo distilaasioŋ bu méngoo ci barab bu nëbbu bu xeetu diffusion buñ tàggat bu njëkk. Xalaat bu am solo bi: tàggat reso ndongo yi ñu tëb directement ci resultaa bu sell bi ci bepp poñ ci trajectoire denoising, suko defee benn tontu bi ñuy yegg ci benn jéego bu mag bu njëkkoon jël yu bari yu ndaw. Lépp soo ko boolee mu am nataal yu ñaw ci diggante 1 ba 4 jéego. Benn pexem àndado, LCM-LoRA, dafay boole gaawaay bi ni adaptëru plug-in bu ndaw buñu mëna wàcce ci modelu Stable Diffusion buñ defar bu baax te duñu tàggataat reso bi yépp.
Gis-gis xarala
Modèle yiy méngoo dañuy tëral màndarga « méngoo seen bopp » : ñaari poñ yu nekk ci benn yoonu denoising (probabilite-flow ODE trajectoire) dañu wara méngoo ak benn nataal bu mujj bu sell. Ndongo dañu koy jëlee ci xeetu diffusion jàngalekat bi ngir mëna satisfaire lii, jàng xam luy waaja xew ci njeexte trajectoire bi ci saasi. Liggéey ci barab bu nëbbu buñ komprime moo gën pixel yi dafay tax distillation bi yomb. Ndax benn jàngat mën na tëb ci trajectoire bi, sampling iteratif bu diis bi dafay daanu ci ay jéego yu néew.
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 lëkkaloo bu nëbbu
Leegi nit ñi dañuy gëna bari ay jéego yu néew, ñu am ñu koy wuutu lu melni SDXL-Turbo, setal LCM, ak pexe distillation adversarial yuy push kalite ci benn wala ñaari jéego. Xaarandil lii dina mëna indi coppite ci nataal yi ci saasi, broos-as-you-go, defar kadre wideo ci jamono dëgg, ak defar ci aparey yi ci telefon yi. Frontier bi dafay tëj bërëb bu ndaw bi ci kalite bi ak diffusion bu mat te bari jéego ak yaatal distillation bu méngoo ci wideo ak 3D, fu sakkanal yi ci dagg lim jéego yi gëna dramatik.
Doxal ci àdduna dëgg
Jumtukaayi toile ci jamono dëgg yuy yeesal nataal biñ defar sooy bind wala desine, ak ab yeexal bu jege amul
Doxal nataalu diffusion bu dëgër ci ordinatër portable wala GPU telefon ci diir bu gàtt
Daaneel benn adaptatëru LCM-LoRA ci benn model buñ defar bu baax ngir gaaw ko ci saasi te doo tàggataat
Defar ay nataal yu bari ci anam wu yomb ngir jàngat jëmmal ci dagg jéego yu tàmbalee ci ~30 wàcci ba ci ~4.
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 Consistency 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 yu nëbbu
Laaj yi ñuy faral di laaj
What is Latent Consistency Models?
Modèlu Consistence Latent (LCMs) xarala la buy may generatëri nataali diffusion ñu defar nataal yu baax ci benn wala ñeenti jéego, duñu def ay fukki-fukki nataal yuñy faral di def. Dañuy defar nataal yuy weccoo xalaat ci jamono dëgg, ba ci aparey yu ndaw.
Lan mooy njariñ li gëna mag ci xeetu méngoo bu nëbbu ci kaw diffusion bu nëbbu buñ miin?
LCMs yi dañuy dindi jéego yu bari yiy dindi bruit ci diffusion standard ci lu tollu ci benn wala ñeent, loolu mooy tax ñu mëna defar ci jamono dëgg.
Ban moomeel 'mën-mën sa bopp' la xeetu mën-mën yi di tëral?
Konsistans mooy poñ yi ci benn trajectoire ODE biy dem ci benn kàrt bu mujj ba ci benn sorti bu sell.
Ci ban barab la LCM yi di defee seen distilaasioŋ?
Liggéey ci barab bu nëbbu, ni ko Stable Diffusion di defee, dafay tax distilaasioŋ bi gëna am njariñ.
Lu la LCM-LoRA may nga def?
LCM-LoRA dafay ëmb gaawaay bi ni adaptatër bu woyof buy wàcci ci model diffusion stabil yuñ defar bu baax.
Naka lay def ba xeetu méngoo mëna am ay jéego yu néew?
Reseau ndongo yi dañu leen distiye ngir mëna wax luy waaja xew ci njeexte trajectoire denoising ci benn tëb, du ci jéego yu ndaw yu bari.