Soppi-Fàttaliku-Xaajal Soppi
Prompt-to-Prompt dafay soppali nataal biñ defar ci soppali bindam ci di jëfandikoowaat kàrtu attention bi ci biir model bi, kon soppi benn baat dafay wecci élément boobu boole ci denc yeneen scene yi.
Résumé
It is editing through words, not pixels.
Plongeur bu xóot
Gaaw-ci-Gaaw (Hertz et al., 2022) xarala la bu amul tàggat ngir soppali mbind ci xeetu diffusion. Li gëna am solo mooy kàrtu cross-attention yi, ñooy wax model bi ban gox ci nataal bi baat bu nekk wara am njeexital, dañuy kode jëmmal barabu seen bi. Sooy defaraat ab nataal ak ab laaj buñu soppeeku tuuti, anam wi dafay dugal kàrtu bàyyi xel bu laaj bu njëkk bi ci doxalin bu bees bi. Soo wecci benn baat, lu melni 'welo' ak 'moto', dangay wecci mbir mi boole ci baña yàq composition bi ak fond bi. Yokk benn baat dafay dugal xel ci jeton yi soppikuwul, suko defee attribut bu bees feeñ te doo soppi lépp. Mën nga itam xoolaat token bi ngir gëna dooleel wala wàññi njeextalam. Ndax soxlawul ñu defar ko bu baax wala mask, mu nekkoon lu am solo ci pexe soppali yu bari ci ginaaw, lu ci melni defar done yu InstructPix2Pix.
Gis-gis xarala
Bu ñuy dindi bruit, cross-attention dafay xayma, ci token bu nekk, kàrtu barab buy wane fimu nekk ci nataal bi. Prompt-to-Prompt dafay kopie kàrt yii ci jamonoy njëkk ba ci biñ soppali ngir ay token yuñ bokk. Ngir weccoo baat, dafay xëcc xel ci diggante ay token yu méngoo; ngir yokk kàddu yi, dafay denc kàrt yu yàgg yi, te du bàyyi ludul token yu bees ñu am xel mu bees; reweighting dafay eskale valeur yiy bàyyi xel ci token bi, di gëna dooleel wala di mute influence visuel bi.
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 soppali-gaawaay-gaawaay Cross-Attention
Manipulation cross-attention leegi dafay jàppale famiy jumtukaayi defar yuñ mëna saytu, te xalaat yi dañuy dem ba ci saytu attention ci architecture yu bees yi ak diffusion wideo ngir soppali yuy méngoo ci diir bu gàtt. Xaarandil lëkkaloo bu gëna dëgër ak soppali nataal dëgg jaaraleko ci soppi, jëfandikoo bu gëna dëgër ci coppite yu mag ci jëmmal, ak boole ak xeeti tegtal suko defee pexe yu bàyyi xel yi di dox te kenn duko gis ci biir interfaasu làkk wiñ wax.
Doxal ci àdduna dëgg
Designer bi dafay soppi 'oto bu xonk ci mbedd mi' def ko 'oto bu baxa ci mbedd mi' ba noppi mu baña soppi jëmmu seen bi.
Benn nataalkat dafay soppi baat 'neige' ngir defar paysage buy gëna sedd ci anam yu bari.
Léebkat bi dafay wecci 'gaynde' ak 'tigre' ngir am pose bu nuróo ak fonndaasioŋ ngir fiche personage bi.
Gëstukat bi daf koy jëfandikoo ngir defar nataal yuñ boole balaa/ganaw, muy done yuy tàggat editër biy topp tegtal yi.
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 Prompt-to-Prompt Cross-Attention Editing 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
Fàttaliku bu jàkkarloo
Laaj yi ñuy faral di laaj
What is Prompt-to-Prompt Cross-Attention Editing?
Prompt-to-Prompt dafay soppali nataal biñ defar ci soppali bindam ci di jëfandikoowaat kàrtu attention bi ci biir model bi, kon soppi benn baat dafay wecci élément boobu boole ci denc yeneen scene yi. Dafay soppali ci kàddu, du ci pixel.
Ban leeral ci biir la Prompt-to-Prompt di jëfandikoowaat ngir baña yàq benn seen?
Kartu cross-attention yi dañuy kode fi baat bu nekk di amee njeexital ci nataal bi, kon soo leen jëfandikoowaat dina tax jëmmal bi nekk benn ci jamonoy soppali.
Ndax Prompt-to-Prompt dafay laaj nga defar model bi bu baax?
Dafay jaay xel ci waxtu inference te soxlawul beneen tàggat.
Lan lay def suñu delloosi bàyyi xel ci benn token?
Escaleer valeur attention bu token dafay gëna am doole wala mute ni konsept bi di feeñee.
Lan moo waral ñu jàpp ni Prompt-to-Prompt mooy pexem gëna am solo?
Manipulation attention bu amul tàggat nekkoon na luñu jëfandikoowaat ci gëstub soppali bi ci topp.