diffusers/v0.18.2/en/_app/pages/using-diffusers/controlling_generation.mdx-hf-doc-builder.js

import{S as g2,i as _2,s as E2,e as a,k as f,w as E,t as l,M as w2,c as r,d as i,m as h,a as o,x as w,h as s,b as n,G as e,g as d,y as b,q as x,o as P,B as y,v as b2}from"../../chunks/vendor-hf-doc-builder.js";import{T as v2}from"../../chunks/Tip-hf-doc-builder.js";import{I as $}from"../../chunks/IconCopyLink-hf-doc-builder.js";function x2(ti){let g,D,_,A,V;return{c(){g=a("p"),D=l(`Pix2Pix Zero is the first model that allows \u201Czero-shot\u201D image editing. This means that the model
can edit an image in less than a minute on a consumer GPU as shown `),_=a("a"),A=l("here"),V=l("."),this.h()},l(T){g=r(T,"P",{});var U=o(g);D=s(U,`Pix2Pix Zero is the first model that allows \u201Czero-shot\u201D image editing. This means that the model
can edit an image in less than a minute on a consumer GPU as shown `),_=r(U,"A",{href:!0});var pe=o(_);A=s(pe,"here"),pe.forEach(i),V=s(U,"."),U.forEach(i),this.h()},h(){n(_,"href","../api/pipelines/stable_diffusion/pix2pix_zero#usage-example")},m(T,U){d(T,g,U),e(g,D),e(g,_),e(_,A),e(g,V)},d(T){T&&i(g)}}}function P2(ti){let g,D;return{c(){g=a("p"),D=l(`An important distinction between methods like InstructPix2Pix and Pix2Pix Zero is that the former
involves fine-tuning the pre-trained weights while the latter does not. This means that you can
apply Pix2Pix Zero to any of the available Stable Diffusion models.`)},l(_){g=r(_,"P",{});var A=o(g);D=s(A,`An important distinction between methods like InstructPix2Pix and Pix2Pix Zero is that the former
involves fine-tuning the pre-trained weights while the latter does not. This means that you can
apply Pix2Pix Zero to any of the available Stable Diffusion models.`),A.forEach(i)},m(_,A){d(_,g,A),e(g,D)},d(_){_&&i(g)}}}function y2(ti){let g,D,_,A,V,T,U,pe,Zl,qo,ii,Bl,Co,ai,Fl,Ro,ri,Wl,Mo,k,Ul,Pr,Ol,Vl,dt,Jl,Yl,ut,Kl,Ql,Lo,oi,Xl,jo,ni,es,zo,li,ts,Ho,u,yr,si,is,as,Ar,fi,rs,os,$r,hi,ns,ls,Tr,pi,ss,fs,Dr,di,hs,ps,kr,ui,ds,us,Sr,ci,cs,ms,Ir,mi,vs,gs,Nr,vi,_s,Es,Gr,gi,ws,bs,qr,_i,xs,Ps,Cr,Ei,ys,As,Rr,wi,$s,Ts,Mr,bi,Ds,Zo,xi,ks,Bo,de,Lr,I,Pi,jr,Ss,Is,yi,zr,Ns,Gs,Ai,ct,qs,Cs,Rs,Ms,$i,Hr,Ls,js,c,N,Ti,Di,zs,Hs,ki,Zs,Bs,Si,Fs,Ws,S,Us,Os,Vs,Js,Ys,Ks,Qs,Xs,G,Ii,Ni,ef,tf,Gi,af,rf,qi,of,nf,Zr,lf,q,Ci,Ri,sf,ff,Mi,hf,pf,Li,df,uf,Br,cf,C,ji,zi,mf,vf,Hi,gf,_f,Zi,Ef,wf,Fr,bf,R,Bi,Fi,xf,Pf,Wi,yf,Af,Ui,$f,Tf,Wr,Df,M,Oi,Vi,kf,Sf,Ji,If,Nf,Yi,Gf,qf,Ur,Cf,L,Ki,Qi,Rf,Mf,Xi,Lf,jf,ea,zf,Hf,Or,Zf,j,ta,ia,Bf,Ff,aa,Wf,Uf,ra,Of,Vf,Vr,Jf,z,oa,na,Yf,Kf,la,Qf,Xf,sa,eh,th,Jr,ih,H,fa,ha,ah,rh,pa,oh,nh,da,lh,sh,O,fh,hh,ph,dh,uh,ch,Z,ua,ca,mh,vh,ma,gh,_h,va,Eh,wh,Yr,bh,B,ga,_a,xh,Ph,Ea,yh,Ah,wa,$h,Th,Kr,Dh,F,ba,xa,kh,Sh,Pa,Ih,Nh,ya,Gh,qh,Qr,Ch,W,Aa,$a,Rh,Mh,Ta,Lh,jh,Da,zh,Hh,Xr,Fo,J,ue,eo,mt,Zh,to,Bh,Wo,ka,vt,Fh,Uo,Y,Sa,Wh,Uh,gt,Oh,Vh,Oo,ce,Jh,Ia,Yh,Kh,Vo,K,me,io,_t,Qh,ao,Xh,Jo,Na,Et,ep,Yo,wt,Ga,tp,ip,Ko,qa,ap,Qo,Ca,rp,Xo,ve,bt,op,xt,np,lp,sp,Pt,fp,yt,hp,pp,en,ge,tn,_e,dp,Ra,up,cp,an,Ee,mp,Ma,vp,gp,rn,Q,we,ro,At,_p,oo,Ep,on,La,$t,wp,nn,Tt,ja,bp,xp,ln,za,Pp,sn,be,yp,no,Ap,$p,fn,xe,Tp,Ha,Dp,kp,hn,X,Pe,lo,Dt,Sp,so,Ip,pn,Za,kt,Np,dn,Ba,Gp,un,Fa,qp,cn,Wa,Cp,mn,ye,Rp,Ua,Mp,Lp,vn,ee,Ae,fo,St,jp,ho,zp,gn,Oa,It,Hp,_n,Nt,Va,Zp,Bp,En,Ja,Fp,wn,$e,Wp,Ya,Up,Op,bn,te,Te,po,Gt,Vp,uo,Jp,xn,Ka,qt,Yp,Pn,Ct,Qa,Kp,Qp,yn,Xa,Xp,An,De,ed,er,td,id,$n,ke,Tn,ie,Se,co,Rt,ad,mo,rd,Dn,tr,Mt,od,kn,Ie,nd,ir,ld,sd,Sn,Ne,fd,ar,hd,pd,In,ae,Ge,vo,Lt,dd,go,ud,Nn,rr,cd,Gn,re,qe,_o,jt,md,Eo,vd,qn,zt,or,gd,_d,Cn,Ce,Ed,nr,wd,bd,Rn,oe,Re,wo,Ht,xd,bo,Pd,Mn,Zt,lr,yd,Ad,Ln,Me,$d,sr,Td,Dd,jn,ne,Le,xo,Bt,kd,Po,Sd,zn,fr,Ft,Id,Hn,Wt,hr,Nd,Gd,Zn,je,qd,pr,Cd,Rd,Bn,le,ze,yo,Ut,Md,Ao,Ld,Fn,dr,jd,Wn,He,zd,ur,Hd,Zd,Un,se,Ze,$o,Ot,Bd,To,Fd,On,Vt,cr,Wd,Ud,Vn,Be,Od,mr,Vd,Jd,Jn,fe,Fe,Do,Jt,Yd,ko,Kd,Yn,vr,Yt,Qd,Kn,We,Xd,gr,eu,tu,Qn,Ue,iu,_r,au,ru,Xn,he,Oe,So,Kt,ou,Io,nu,el,Er,Qt,lu,tl,Xt,wr,su,fu,il,Ve,hu,br,pu,du,al;return T=new $({}),mt=new $({}),_t=new $({}),ge=new v2({props:{$$slots:{default:[x2]},$$scope:{ctx:ti}}}),At=new $({}),Dt=new $({}),St=new $({}),Gt=new $({}),ke=new v2({props:{$$slots:{default:[P2]},$$scope:{ctx:ti}}}),Rt=new $({}),Lt=new $({}),jt=new $({}),Ht=new $({}),Bt=new $({}),Ut=new $({}),Ot=new $({}),Jt=new $({}),Kt=new $({}),{c(){g=a("meta"),D=f(),_=a("h1"),A=a("a"),V=a("span"),E(T.$$.fragment),U=f(),pe=a("span"),Zl=l("Controlled generation"),qo=f(),ii=a("p"),Bl=l("Controlling outputs generated by diffusion models has been long pursued by the community and is now an active research topic. In many popular diffusion models, subtle changes in inputs, both images and text prompts, can drastically change outputs. In an ideal world we want to be able to control how semantics are preserved and changed."),Co=f(),ai=a("p"),Fl=l("Most examples of preserving semantics reduce to being able to accurately map a change in input to a change in output. I.e. adding an adjective to a subject in a prompt preserves the entire image, only modifying the changed subject. Or, image variation of a particular subject preserves the subject\u2019s pose."),Ro=f(),ri=a("p"),Wl=l("Additionally, there are qualities of generated images that we would like to influence beyond semantic preservation. I.e. in general, we would like our outputs to be of good quality, adhere to a particular style, or be realistic."),Mo=f(),k=a("p"),Ul=l("We will document some of the techniques "),Pr=a("code"),Ol=l("diffusers"),Vl=l(" supports to control generation of diffusion models. Much is cutting edge research and can be quite nuanced. If something needs clarifying or you have a suggestion, don\u2019t hesitate to open a discussion on the "),dt=a("a"),Jl=l("forum"),Yl=l(" or a "),ut=a("a"),Kl=l("GitHub issue"),Ql=l("."),Lo=f(),oi=a("p"),Xl=l("We provide a high level explanation of how the generation can be controlled as well as a snippet of the technicals. For more in depth explanations on the technicals, the original papers which are linked from the pipelines are always the best resources."),jo=f(),ni=a("p"),es=l("Depending on the use case, one should choose a technique accordingly. In many cases, these techniques can be combined. For example, one can combine Textual Inversion with SEGA to provide more semantic guidance to the outputs generated using Textual Inversion."),zo=f(),li=a("p"),ts=l("Unless otherwise mentioned, these are techniques that work with existing models and don\u2019t require their own weights."),Ho=f(),u=a("ol"),yr=a("li"),si=a("a"),is=l("Instruct Pix2Pix"),as=f(),Ar=a("li"),fi=a("a"),rs=l("Pix2Pix Zero"),os=f(),$r=a("li"),hi=a("a"),ns=l("Attend and Excite"),ls=f(),Tr=a("li"),pi=a("a"),ss=l("Semantic Guidance"),fs=f(),Dr=a("li"),di=a("a"),hs=l("Self-attention Guidance"),ps=f(),kr=a("li"),ui=a("a"),ds=l("Depth2Image"),us=f(),Sr=a("li"),ci=a("a"),cs=l("MultiDiffusion Panorama"),ms=f(),Ir=a("li"),mi=a("a"),vs=l("DreamBooth"),gs=f(),Nr=a("li"),vi=a("a"),_s=l("Textual Inversion"),Es=f(),Gr=a("li"),gi=a("a"),ws=l("ControlNet"),bs=f(),qr=a("li"),_i=a("a"),xs=l("Prompt Weighting"),Ps=f(),Cr=a("li"),Ei=a("a"),ys=l("Custom Diffusion"),As=f(),Rr=a("li"),wi=a("a"),$s=l("Model Editing"),Ts=f(),Mr=a("li"),bi=a("a"),Ds=l("DiffEdit"),Zo=f(),xi=a("p"),ks=l("For convenience, we provide a table to denote which methods are inference-only and which require fine-tuning/training."),Bo=f(),de=a("table"),Lr=a("thead"),I=a("tr"),Pi=a("th"),jr=a("strong"),Ss=l("Method"),Is=f(),yi=a("th"),zr=a("strong"),Ns=l("Inference only"),Gs=f(),Ai=a("th"),ct=a("strong"),qs=l("Requires training /"),Cs=a("br"),Rs=l(" fine-tuning"),Ms=f(),$i=a("th"),Hr=a("strong"),Ls=l("Comments"),js=f(),c=a("tbody"),N=a("tr"),Ti=a("td"),Di=a("a"),zs=l("Instruct Pix2Pix"),Hs=f(),ki=a("td"),Zs=l("\u2705"),Bs=f(),Si=a("td"),Fs=l("\u274C"),Ws=f(),S=a("td"),Us=l("Can additionally be"),Os=a("br"),Vs=l("fine-tuned for better "),Js=a("br"),Ys=l("performance on specific "),Ks=a("br"),Qs=l("edit instructions."),Xs=f(),G=a("tr"),Ii=a("td"),Ni=a("a"),ef=l("Pix2Pix Zero"),tf=f(),Gi=a("td"),af=l("\u2705"),rf=f(),qi=a("td"),of=l("\u274C"),nf=f(),Zr=a("td"),lf=f(),q=a("tr"),Ci=a("td"),Ri=a("a"),sf=l("Attend and Excite"),ff=f(),Mi=a("td"),hf=l("\u2705"),pf=f(),Li=a("td"),df=l("\u274C"),uf=f(),Br=a("td"),cf=f(),C=a("tr"),ji=a("td"),zi=a("a"),mf=l("Semantic Guidance"),vf=f(),Hi=a("td"),gf=l("\u2705"),_f=f(),Zi=a("td"),Ef=l("\u274C"),wf=f(),Fr=a("td"),bf=f(),R=a("tr"),Bi=a("td"),Fi=a("a"),xf=l("Self-attention Guidance"),Pf=f(),Wi=a("td"),yf=l("\u2705"),Af=f(),Ui=a("td"),$f=l("\u274C"),Tf=f(),Wr=a("td"),Df=f(),M=a("tr"),Oi=a("td"),Vi=a("a"),kf=l("Depth2Image"),Sf=f(),Ji=a("td"),If=l("\u2705"),Nf=f(),Yi=a("td"),Gf=l("\u274C"),qf=f(),Ur=a("td"),Cf=f(),L=a("tr"),Ki=a("td"),Qi=a("a"),Rf=l("MultiDiffusion Panorama"),Mf=f(),Xi=a("td"),Lf=l("\u2705"),jf=f(),ea=a("td"),zf=l("\u274C"),Hf=f(),Or=a("td"),Zf=f(),j=a("tr"),ta=a("td"),ia=a("a"),Bf=l("DreamBooth"),Ff=f(),aa=a("td"),Wf=l("\u274C"),Uf=f(),ra=a("td"),Of=l("\u2705"),Vf=f(),Vr=a("td"),Jf=f(),z=a("tr"),oa=a("td"),na=a("a"),Yf=l("Textual Inversion"),Kf=f(),la=a("td"),Qf=l("\u274C"),Xf=f(),sa=a("td"),eh=l("\u2705"),th=f(),Jr=a("td"),ih=f(),H=a("tr"),fa=a("td"),ha=a("a"),ah=l("ControlNet"),rh=f(),pa=a("td"),oh=l("\u2705"),nh=f(),da=a("td"),lh=l("\u274C"),sh=f(),O=a("td"),fh=l("A ControlNet can be "),hh=a("br"),ph=l("trained/fine-tuned on"),dh=a("br"),uh=l("a custom conditioning."),ch=f(),Z=a("tr"),ua=a("td"),ca=a("a"),mh=l("Prompt Weighting"),vh=f(),ma=a("td"),gh=l("\u2705"),_h=f(),va=a("td"),Eh=l("\u274C"),wh=f(),Yr=a("td"),bh=f(),B=a("tr"),ga=a("td"),_a=a("a"),xh=l("Custom Diffusion"),Ph=f(),Ea=a("td"),yh=l("\u274C"),Ah=f(),wa=a("td"),$h=l("\u2705"),Th=f(),Kr=a("td"),Dh=f(),F=a("tr"),ba=a("td"),xa=a("a"),kh=l("Model Editing"),Sh=f(),Pa=a("td"),Ih=l("\u2705"),Nh=f(),ya=a("td"),Gh=l("\u274C"),qh=f(),Qr=a("td"),Ch=f(),W=a("tr"),Aa=a("td"),$a=a("a"),Rh=l("DiffEdit"),Mh=f(),Ta=a("td"),Lh=l("\u2705"),jh=f(),Da=a("td"),zh=l("\u274C"),Hh=f(),Xr=a("td"),Fo=f(),J=a("h2"),ue=a("a"),eo=a("span"),E(mt.$$.fragment),Zh=f(),to=a("span"),Bh=l("Instruct Pix2Pix"),Wo=f(),ka=a("p"),vt=a("a"),Fh=l("Paper"),Uo=f(),Y=a("p"),Sa=a("a"),Wh=l("Instruct Pix2Pix"),Uh=l(` is fine-tuned from stable diffusion to support editing input images. It takes as inputs an image and a prompt describing an edit, and it outputs the edited image.
Instruct Pix2Pix has been explicitly trained to work well with `),gt=a("a"),Oh=l("InstructGPT"),Vh=l("-like prompts."),Oo=f(),ce=a("p"),Jh=l("See "),Ia=a("a"),Yh=l("here"),Kh=l(" for more information on how to use it."),Vo=f(),K=a("h2"),me=a("a"),io=a("span"),E(_t.$$.fragment),Qh=f(),ao=a("span"),Xh=l("Pix2Pix Zero"),Jo=f(),Na=a("p"),Et=a("a"),ep=l("Paper"),Yo=f(),wt=a("p"),Ga=a("a"),tp=l("Pix2Pix Zero"),ip=l(" allows modifying an image so that one concept or subject is translated to another one while preserving general image semantics."),Ko=f(),qa=a("p"),ap=l("The denoising process is guided from one conceptual embedding towards another conceptual embedding. The intermediate latents are optimized during the denoising process to push the attention maps towards reference attention maps. The reference attention maps are from the denoising process of the input image and are used to encourage semantic preservation."),Qo=f(),Ca=a("p"),rp=l("Pix2Pix Zero can be used both to edit synthetic images as well as real images."),Xo=f(),ve=a("ul"),bt=a("li"),op=l(`To edit synthetic images, one first generates an image given a caption.
Next, we generate image captions for the concept that shall be edited and for the new target concept. We can use a model like `),xt=a("a"),np=l("Flan-T5"),lp=l(" for this purpose. Then, \u201Cmean\u201D prompt embeddings for both the source and target concepts are created via the text encoder. Finally, the pix2pix-zero algorithm is used to edit the synthetic image."),sp=f(),Pt=a("li"),fp=l("To edit a real image, one first generates an image caption using a model like "),yt=a("a"),hp=l("BLIP"),pp=l(". Then one applies ddim inversion on the prompt and image to generate \u201Cinverse\u201D latents. Similar to before, \u201Cmean\u201D prompt embeddings for both source and target concepts are created and finally the pix2pix-zero algorithm in combination with the \u201Cinverse\u201D latents is used to edit the image."),en=f(),E(ge.$$.fragment),tn=f(),_e=a("p"),dp=l(`As mentioned above, Pix2Pix Zero includes optimizing the latents (and not any of the UNet, VAE, or the text encoder) to steer the generation toward a specific concept. This means that the overall
pipeline might require more memory than a standard `),Ra=a("a"),up=l("StableDiffusionPipeline"),cp=l("."),an=f(),Ee=a("p"),mp=l("See "),Ma=a("a"),vp=l("here"),gp=l(" for more information on how to use it."),rn=f(),Q=a("h2"),we=a("a"),ro=a("span"),E(At.$$.fragment),_p=f(),oo=a("span"),Ep=l("Attend and Excite"),on=f(),La=a("p"),$t=a("a"),wp=l("Paper"),nn=f(),Tt=a("p"),ja=a("a"),bp=l("Attend and Excite"),xp=l(" allows subjects in the prompt to be faithfully represented in the final image."),ln=f(),za=a("p"),Pp=l("A set of token indices are given as input, corresponding to the subjects in the prompt that need to be present in the image. During denoising, each token index is guaranteed to have a minimum attention threshold for at least one patch of the image. The intermediate latents are iteratively optimized during the denoising process to strengthen the attention of the most neglected subject token until the attention threshold is passed for all subject tokens."),sn=f(),be=a("p"),yp=l("Like Pix2Pix Zero, Attend and Excite also involves a mini optimization loop (leaving the pre-trained weights untouched) in its pipeline and can require more memory than the usual "),no=a("code"),Ap=l("StableDiffusionPipeline"),$p=l("."),fn=f(),xe=a("p"),Tp=l("See "),Ha=a("a"),Dp=l("here"),kp=l(" for more information on how to use it."),hn=f(),X=a("h2"),Pe=a("a"),lo=a("span"),E(Dt.$$.fragment),Sp=f(),so=a("span"),Ip=l("Semantic Guidance (SEGA)"),pn=f(),Za=a("p"),kt=a("a"),Np=l("Paper"),dn=f(),Ba=a("p"),Gp=l("SEGA allows applying or removing one or more concepts from an image. The strength of the concept can also be controlled. I.e. the smile concept can be used to incrementally increase or decrease the smile of a portrait."),un=f(),Fa=a("p"),qp=l("Similar to how classifier free guidance provides guidance via empty prompt inputs, SEGA provides guidance on conceptual prompts. Multiple of these conceptual prompts can be applied simultaneously. Each conceptual prompt can either add or remove their concept depending on if the guidance is applied positively or negatively."),cn=f(),Wa=a("p"),Cp=l("Unlike Pix2Pix Zero or Attend and Excite, SEGA directly interacts with the diffusion process instead of performing any explicit gradient-based optimization."),mn=f(),ye=a("p"),Rp=l("See "),Ua=a("a"),Mp=l("here"),Lp=l(" for more information on how to use it."),vn=f(),ee=a("h2"),Ae=a("a"),fo=a("span"),E(St.$$.fragment),jp=f(),ho=a("span"),zp=l("Self-attention Guidance (SAG)"),gn=f(),Oa=a("p"),It=a("a"),Hp=l("Paper"),_n=f(),Nt=a("p"),Va=a("a"),Zp=l("Self-attention Guidance"),Bp=l(" improves the general quality of images."),En=f(),Ja=a("p"),Fp=l("SAG provides guidance from predictions not conditioned on high-frequency details to fully conditioned images. The high frequency details are extracted out of the UNet self-attention maps."),wn=f(),$e=a("p"),Wp=l("See "),Ya=a("a"),Up=l("here"),Op=l(" for more information on how to use it."),bn=f(),te=a("h2"),Te=a("a"),po=a("span"),E(Gt.$$.fragment),Vp=f(),uo=a("span"),Jp=l("Depth2Image"),xn=f(),Ka=a("p"),qt=a("a"),Yp=l("Project"),Pn=f(),Ct=a("p"),Qa=a("a"),Kp=l("Depth2Image"),Qp=l(" is fine-tuned from Stable Diffusion to better preserve semantics for text guided image variation."),yn=f(),Xa=a("p"),Xp=l("It conditions on a monocular depth estimate of the original image."),An=f(),De=a("p"),ed=l("See "),er=a("a"),td=l("here"),id=l(" for more information on how to use it."),$n=f(),E(ke.$$.fragment),Tn=f(),ie=a("h2"),Se=a("a"),co=a("span"),E(Rt.$$.fragment),ad=f(),mo=a("span"),rd=l("MultiDiffusion Panorama"),Dn=f(),tr=a("p"),Mt=a("a"),od=l("Paper"),kn=f(),Ie=a("p"),nd=l(`MultiDiffusion defines a new generation process over a pre-trained diffusion model. This process binds together multiple diffusion generation methods that can be readily applied to generate high quality and diverse images. Results adhere to user-provided controls, such as desired aspect ratio (e.g., panorama), and spatial guiding signals, ranging from tight segmentation masks to bounding boxes.
`),ir=a("a"),ld=l("MultiDiffusion Panorama"),sd=l(" allows to generate high-quality images at arbitrary aspect ratios (e.g., panoramas)."),Sn=f(),Ne=a("p"),fd=l("See "),ar=a("a"),hd=l("here"),pd=l(" for more information on how to use it to generate panoramic images."),In=f(),ae=a("h2"),Ge=a("a"),vo=a("span"),E(Lt.$$.fragment),dd=f(),go=a("span"),ud=l("Fine-tuning your own models"),Nn=f(),rr=a("p"),cd=l("In addition to pre-trained models, Diffusers has training scripts for fine-tuning models on user-provided data."),Gn=f(),re=a("h2"),qe=a("a"),_o=a("span"),E(jt.$$.fragment),md=f(),Eo=a("span"),vd=l("DreamBooth"),qn=f(),zt=a("p"),or=a("a"),gd=l("DreamBooth"),_d=l(" fine-tunes a model to teach it about a new subject. I.e. a few pictures of a person can be used to generate images of that person in different styles."),Cn=f(),Ce=a("p"),Ed=l("See "),nr=a("a"),wd=l("here"),bd=l(" for more information on how to use it."),Rn=f(),oe=a("h2"),Re=a("a"),wo=a("span"),E(Ht.$$.fragment),xd=f(),bo=a("span"),Pd=l("Textual Inversion"),Mn=f(),Zt=a("p"),lr=a("a"),yd=l("Textual Inversion"),Ad=l(" fine-tunes a model to teach it about a new concept. I.e. a few pictures of a style of artwork can be used to generate images in that style."),Ln=f(),Me=a("p"),$d=l("See "),sr=a("a"),Td=l("here"),Dd=l(" for more information on how to use it."),jn=f(),ne=a("h2"),Le=a("a"),xo=a("span"),E(Bt.$$.fragment),kd=f(),Po=a("span"),Sd=l("ControlNet"),zn=f(),fr=a("p"),Ft=a("a"),Id=l("Paper"),Hn=f(),Wt=a("p"),hr=a("a"),Nd=l("ControlNet"),Gd=l(` is an auxiliary network which adds an extra condition.
There are 8 canonical pre-trained ControlNets trained on different conditionings such as edge detection, scribbles,
depth maps, and semantic segmentations.`),Zn=f(),je=a("p"),qd=l("See "),pr=a("a"),Cd=l("here"),Rd=l(" for more information on how to use it."),Bn=f(),le=a("h2"),ze=a("a"),yo=a("span"),E(Ut.$$.fragment),Md=f(),Ao=a("span"),Ld=l("Prompt Weighting"),Fn=f(),dr=a("p"),jd=l(`Prompt weighting is a simple technique that puts more attention weight on certain parts of the text
input.`),Wn=f(),He=a("p"),zd=l("For a more in-detail explanation and examples, see "),ur=a("a"),Hd=l("here"),Zd=l("."),Un=f(),se=a("h2"),Ze=a("a"),$o=a("span"),E(Ot.$$.fragment),Bd=f(),To=a("span"),Fd=l("Custom Diffusion"),On=f(),Vt=a("p"),cr=a("a"),Wd=l("Custom Diffusion"),Ud=l(` only fine-tunes the cross-attention maps of a pre-trained
text-to-image diffusion model. It also allows for additionally performing textual inversion. It supports
multi-concept training by design. Like DreamBooth and Textual Inversion, Custom Diffusion is also used to
teach a pre-trained text-to-image diffusion model about new concepts to generate outputs involving the
concept(s) of interest.`),Vn=f(),Be=a("p"),Od=l("For more details, check out our "),mr=a("a"),Vd=l("official doc"),Jd=l("."),Jn=f(),fe=a("h2"),Fe=a("a"),Do=a("span"),E(Jt.$$.fragment),Yd=f(),ko=a("span"),Kd=l("Model Editing"),Yn=f(),vr=a("p"),Yt=a("a"),Qd=l("Paper"),Kn=f(),We=a("p"),Xd=l("The "),gr=a("a"),eu=l("text-to-image model editing pipeline"),tu=l(` helps you mitigate some of the incorrect implicit assumptions a pre-trained text-to-image
diffusion model might make about the subjects present in the input prompt. For example, if you prompt Stable Diffusion to generate images for \u201CA pack of roses\u201D, the roses in the generated images
are more likely to be red. This pipeline helps you change that assumption.`),Qn=f(),Ue=a("p"),iu=l("To know more details, check out the "),_r=a("a"),au=l("official doc"),ru=l("."),Xn=f(),he=a("h2"),Oe=a("a"),So=a("span"),E(Kt.$$.fragment),ou=f(),Io=a("span"),nu=l("DiffEdit"),el=f(),Er=a("p"),Qt=a("a"),lu=l("Paper"),tl=f(),Xt=a("p"),wr=a("a"),su=l("DiffEdit"),fu=l(` allows for semantic editing of input images along with
input prompts while preserving the original input images as much as possible.`),il=f(),Ve=a("p"),hu=l("To know more details, check out the "),br=a("a"),pu=l("official doc"),du=l("."),this.h()},l(t){const p=w2('[data-svelte="svelte-1phssyn"]',document.head);g=r(p,"META",{name:!0,content:!0}),p.forEach(i),D=h(t),_=r(t,"H1",{class:!0});var ei=o(_);A=r(ei,"A",{id:!0,class:!0,href:!0});var No=o(A);V=r(No,"SPAN",{});var xu=o(V);w(T.$$.fragment,xu),xu.forEach(i),No.forEach(i),U=h(ei),pe=r(ei,"SPAN",{});var Pu=o(pe);Zl=s(Pu,"Controlled generation"),Pu.forEach(i),ei.forEach(i),qo=h(t),ii=r(t,"P",{});var yu=o(ii);Bl=s(yu,"Controlling outputs generated by diffusion models has been long pursued by the community and is now an active research topic. In many popular diffusion models, subtle changes in inputs, both images and text prompts, can drastically change outputs. In an ideal world we want to be able to control how semantics are preserved and changed."),yu.forEach(i),Co=h(t),ai=r(t,"P",{});var Au=o(ai);Fl=s(Au,"Most examples of preserving semantics reduce to being able to accurately map a change in input to a change in output. I.e. adding an adjective to a subject in a prompt preserves the entire image, only modifying the changed subject. Or, image variation of a particular subject preserves the subject\u2019s pose."),Au.forEach(i),Ro=h(t),ri=r(t,"P",{});var $u=o(ri);Wl=s($u,"Additionally, there are qualities of generated images that we would like to influence beyond semantic preservation. I.e. in general, we would like our outputs to be of good quality, adhere to a particular style, or be realistic."),$u.forEach(i),Mo=h(t),k=r(t,"P",{});var Je=o(k);Ul=s(Je,"We will document some of the techniques "),Pr=r(Je,"CODE",{});var Tu=o(Pr);Ol=s(Tu,"diffusers"),Tu.forEach(i),Vl=s(Je," supports to control generation of diffusion models. Much is cutting edge research and can be quite nuanced. If something needs clarifying or you have a suggestion, don\u2019t hesitate to open a discussion on the "),dt=r(Je,"A",{href:!0,rel:!0});var Du=o(dt);Jl=s(Du,"forum"),Du.forEach(i),Yl=s(Je," or a "),ut=r(Je,"A",{href:!0,rel:!0});var ku=o(ut);Kl=s(ku,"GitHub issue"),ku.forEach(i),Ql=s(Je,"."),Je.forEach(i),Lo=h(t),oi=r(t,"P",{});var Su=o(oi);Xl=s(Su,"We provide a high level explanation of how the generation can be controlled as well as a snippet of the technicals. For more in depth explanations on the technicals, the original papers which are linked from the pipelines are always the best resources."),Su.forEach(i),jo=h(t),ni=r(t,"P",{});var Iu=o(ni);es=s(Iu,"Depending on the use case, one should choose a technique accordingly. In many cases, these techniques can be combined. For example, one can combine Textual Inversion with SEGA to provide more semantic guidance to the outputs generated using Textual Inversion."),Iu.forEach(i),zo=h(t),li=r(t,"P",{});var Nu=o(li);ts=s(Nu,"Unless otherwise mentioned, these are techniques that work with existing models and don\u2019t require their own weights."),Nu.forEach(i),Ho=h(t),u=r(t,"OL",{});var m=o(u);yr=r(m,"LI",{});var Gu=o(yr);si=r(Gu,"A",{href:!0});var qu=o(si);is=s(qu,"Instruct Pix2Pix"),qu.forEach(i),Gu.forEach(i),as=h(m),Ar=r(m,"LI",{});var Cu=o(Ar);fi=r(Cu,"A",{href:!0});var Ru=o(fi);rs=s(Ru,"Pix2Pix Zero"),Ru.forEach(i),Cu.forEach(i),os=h(m),$r=r(m,"LI",{});var Mu=o($r);hi=r(Mu,"A",{href:!0});var Lu=o(hi);ns=s(Lu,"Attend and Excite"),Lu.forEach(i),Mu.forEach(i),ls=h(m),Tr=r(m,"LI",{});var ju=o(Tr);pi=r(ju,"A",{href:!0});var zu=o(pi);ss=s(zu,"Semantic Guidance"),zu.forEach(i),ju.forEach(i),fs=h(m),Dr=r(m,"LI",{});var Hu=o(Dr);di=r(Hu,"A",{href:!0});var Zu=o(di);hs=s(Zu,"Self-attention Guidance"),Zu.forEach(i),Hu.forEach(i),ps=h(m),kr=r(m,"LI",{});var Bu=o(kr);ui=r(Bu,"A",{href:!0});var Fu=o(ui);ds=s(Fu,"Depth2Image"),Fu.forEach(i),Bu.forEach(i),us=h(m),Sr=r(m,"LI",{});var Wu=o(Sr);ci=r(Wu,"A",{href:!0});var Uu=o(ci);cs=s(Uu,"MultiDiffusion Panorama"),Uu.forEach(i),Wu.forEach(i),ms=h(m),Ir=r(m,"LI",{});var Ou=o(Ir);mi=r(Ou,"A",{href:!0});var Vu=o(mi);vs=s(Vu,"DreamBooth"),Vu.forEach(i),Ou.forEach(i),gs=h(m),Nr=r(m,"LI",{});var Ju=o(Nr);vi=r(Ju,"A",{href:!0});var Yu=o(vi);_s=s(Yu,"Textual Inversion"),Yu.forEach(i),Ju.forEach(i),Es=h(m),Gr=r(m,"LI",{});var Ku=o(Gr);gi=r(Ku,"A",{href:!0});var Qu=o(gi);ws=s(Qu,"ControlNet"),Qu.forEach(i),Ku.forEach(i),bs=h(m),qr=r(m,"LI",{});var Xu=o(qr);_i=r(Xu,"A",{href:!0});var ec=o(_i);xs=s(ec,"Prompt Weighting"),ec.forEach(i),Xu.forEach(i),Ps=h(m),Cr=r(m,"LI",{});var tc=o(Cr);Ei=r(tc,"A",{href:!0});var ic=o(Ei);ys=s(ic,"Custom Diffusion"),ic.forEach(i),tc.forEach(i),As=h(m),Rr=r(m,"LI",{});var ac=o(Rr);wi=r(ac,"A",{href:!0});var rc=o(wi);$s=s(rc,"Model Editing"),rc.forEach(i),ac.forEach(i),Ts=h(m),Mr=r(m,"LI",{});var oc=o(Mr);bi=r(oc,"A",{href:!0});var nc=o(bi);Ds=s(nc,"DiffEdit"),nc.forEach(i),oc.forEach(i),m.forEach(i),Zo=h(t),xi=r(t,"P",{});var lc=o(xi);ks=s(lc,"For convenience, we provide a table to denote which methods are inference-only and which require fine-tuning/training."),lc.forEach(i),Bo=h(t),de=r(t,"TABLE",{});var rl=o(de);Lr=r(rl,"THEAD",{});var sc=o(Lr);I=r(sc,"TR",{});var Ye=o(I);Pi=r(Ye,"TH",{align:!0});var fc=o(Pi);jr=r(fc,"STRONG",{});var hc=o(jr);Ss=s(hc,"Method"),hc.forEach(i),fc.forEach(i),Is=h(Ye),yi=r(Ye,"TH",{align:!0});var pc=o(yi);zr=r(pc,"STRONG",{});var dc=o(zr);Ns=s(dc,"Inference only"),dc.forEach(i),pc.forEach(i),Gs=h(Ye),Ai=r(Ye,"TH",{align:!0});var uc=o(Ai);ct=r(uc,"STRONG",{});var ol=o(ct);qs=s(ol,"Requires training /"),Cs=r(ol,"BR",{}),Rs=s(ol," fine-tuning"),ol.forEach(i),uc.forEach(i),Ms=h(Ye),$i=r(Ye,"TH",{align:!0});var cc=o($i);Hr=r(cc,"STRONG",{});var mc=o(Hr);Ls=s(mc,"Comments"),mc.forEach(i),cc.forEach(i),Ye.forEach(i),sc.forEach(i),js=h(rl),c=r(rl,"TBODY",{});var v=o(c);N=r(v,"TR",{});var Ke=o(N);Ti=r(Ke,"TD",{align:!0});var vc=o(Ti);Di=r(vc,"A",{href:!0});var gc=o(Di);zs=s(gc,"Instruct Pix2Pix"),gc.forEach(i),vc.forEach(i),Hs=h(Ke),ki=r(Ke,"TD",{align:!0});var _c=o(ki);Zs=s(_c,"\u2705"),_c.forEach(i),Bs=h(Ke),Si=r(Ke,"TD",{align:!0});var Ec=o(Si);Fs=s(Ec,"\u274C"),Ec.forEach(i),Ws=h(Ke),S=r(Ke,"TD",{align:!0});var Qe=o(S);Us=s(Qe,"Can additionally be"),Os=r(Qe,"BR",{}),Vs=s(Qe,"fine-tuned for better "),Js=r(Qe,"BR",{}),Ys=s(Qe,"performance on specific "),Ks=r(Qe,"BR",{}),Qs=s(Qe,"edit instructions."),Qe.forEach(i),Ke.forEach(i),Xs=h(v),G=r(v,"TR",{});var Xe=o(G);Ii=r(Xe,"TD",{align:!0});var wc=o(Ii);Ni=r(wc,"A",{href:!0});var bc=o(Ni);ef=s(bc,"Pix2Pix Zero"),bc.forEach(i),wc.forEach(i),tf=h(Xe),Gi=r(Xe,"TD",{align:!0});var xc=o(Gi);af=s(xc,"\u2705"),xc.forEach(i),rf=h(Xe),qi=r(Xe,"TD",{align:!0});var Pc=o(qi);of=s(Pc,"\u274C"),Pc.forEach(i),nf=h(Xe),Zr=r(Xe,"TD",{align:!0}),o(Zr).forEach(i),Xe.forEach(i),lf=h(v),q=r(v,"TR",{});var et=o(q);Ci=r(et,"TD",{align:!0});var yc=o(Ci);Ri=r(yc,"A",{href:!0});var Ac=o(Ri);sf=s(Ac,"Attend and Excite"),Ac.forEach(i),yc.forEach(i),ff=h(et),Mi=r(et,"TD",{align:!0});var $c=o(Mi);hf=s($c,"\u2705"),$c.forEach(i),pf=h(et),Li=r(et,"TD",{align:!0});var Tc=o(Li);df=s(Tc,"\u274C"),Tc.forEach(i),uf=h(et),Br=r(et,"TD",{align:!0}),o(Br).forEach(i),et.forEach(i),cf=h(v),C=r(v,"TR",{});var tt=o(C);ji=r(tt,"TD",{align:!0});var Dc=o(ji);zi=r(Dc,"A",{href:!0});var kc=o(zi);mf=s(kc,"Semantic Guidance"),kc.forEach(i),Dc.forEach(i),vf=h(tt),Hi=r(tt,"TD",{align:!0});var Sc=o(Hi);gf=s(Sc,"\u2705"),Sc.forEach(i),_f=h(tt),Zi=r(tt,"TD",{align:!0});var Ic=o(Zi);Ef=s(Ic,"\u274C"),Ic.forEach(i),wf=h(tt),Fr=r(tt,"TD",{align:!0}),o(Fr).forEach(i),tt.forEach(i),bf=h(v),R=r(v,"TR",{});var it=o(R);Bi=r(it,"TD",{align:!0});var Nc=o(Bi);Fi=r(Nc,"A",{href:!0});var Gc=o(Fi);xf=s(Gc,"Self-attention Guidance"),Gc.forEach(i),Nc.forEach(i),Pf=h(it),Wi=r(it,"TD",{align:!0});var qc=o(Wi);yf=s(qc,"\u2705"),qc.forEach(i),Af=h(it),Ui=r(it,"TD",{align:!0});var Cc=o(Ui);$f=s(Cc,"\u274C"),Cc.forEach(i),Tf=h(it),Wr=r(it,"TD",{align:!0}),o(Wr).forEach(i),it.forEach(i),Df=h(v),M=r(v,"TR",{});var at=o(M);Oi=r(at,"TD",{align:!0});var Rc=o(Oi);Vi=r(Rc,"A",{href:!0});var Mc=o(Vi);kf=s(Mc,"Depth2Image"),Mc.forEach(i),Rc.forEach(i),Sf=h(at),Ji=r(at,"TD",{align:!0});var Lc=o(Ji);If=s(Lc,"\u2705"),Lc.forEach(i),Nf=h(at),Yi=r(at,"TD",{align:!0});var jc=o(Yi);Gf=s(jc,"\u274C"),jc.forEach(i),qf=h(at),Ur=r(at,"TD",{align:!0}),o(Ur).forEach(i),at.forEach(i),Cf=h(v),L=r(v,"TR",{});var rt=o(L);Ki=r(rt,"TD",{align:!0});var zc=o(Ki);Qi=r(zc,"A",{href:!0});var Hc=o(Qi);Rf=s(Hc,"MultiDiffusion Panorama"),Hc.forEach(i),zc.forEach(i),Mf=h(rt),Xi=r(rt,"TD",{align:!0});var Zc=o(Xi);Lf=s(Zc,"\u2705"),Zc.forEach(i),jf=h(rt),ea=r(rt,"TD",{align:!0});var Bc=o(ea);zf=s(Bc,"\u274C"),Bc.forEach(i),Hf=h(rt),Or=r(rt,"TD",{align:!0}),o(Or).forEach(i),rt.forEach(i),Zf=h(v),j=r(v,"TR",{});var ot=o(j);ta=r(ot,"TD",{align:!0});var Fc=o(ta);ia=r(Fc,"A",{href:!0});var Wc=o(ia);Bf=s(Wc,"DreamBooth"),Wc.forEach(i),Fc.forEach(i),Ff=h(ot),aa=r(ot,"TD",{align:!0});var Uc=o(aa);Wf=s(Uc,"\u274C"),Uc.forEach(i),Uf=h(ot),ra=r(ot,"TD",{align:!0});var Oc=o(ra);Of=s(Oc,"\u2705"),Oc.forEach(i),Vf=h(ot),Vr=r(ot,"TD",{align:!0}),o(Vr).forEach(i),ot.forEach(i),Jf=h(v),z=r(v,"TR",{});var nt=o(z);oa=r(nt,"TD",{align:!0});var Vc=o(oa);na=r(Vc,"A",{href:!0});var Jc=o(na);Yf=s(Jc,"Textual Inversion"),Jc.forEach(i),Vc.forEach(i),Kf=h(nt),la=r(nt,"TD",{align:!0});var Yc=o(la);Qf=s(Yc,"\u274C"),Yc.forEach(i),Xf=h(nt),sa=r(nt,"TD",{align:!0});var Kc=o(sa);eh=s(Kc,"\u2705"),Kc.forEach(i),th=h(nt),Jr=r(nt,"TD",{align:!0}),o(Jr).forEach(i),nt.forEach(i),ih=h(v),H=r(v,"TR",{});var lt=o(H);fa=r(lt,"TD",{align:!0});var Qc=o(fa);ha=r(Qc,"A",{href:!0});var Xc=o(ha);ah=s(Xc,"ControlNet"),Xc.forEach(i),Qc.forEach(i),rh=h(lt),pa=r(lt,"TD",{align:!0});var em=o(pa);oh=s(em,"\u2705"),em.forEach(i),nh=h(lt),da=r(lt,"TD",{align:!0});var tm=o(da);lh=s(tm,"\u274C"),tm.forEach(i),sh=h(lt),O=r(lt,"TD",{align:!0});var xr=o(O);fh=s(xr,"A ControlNet can be "),hh=r(xr,"BR",{}),ph=s(xr,"trained/fine-tuned on"),dh=r(xr,"BR",{}),uh=s(xr,"a custom conditioning."),xr.forEach(i),lt.forEach(i),ch=h(v),Z=r(v,"TR",{});var st=o(Z);ua=r(st,"TD",{align:!0});var im=o(ua);ca=r(im,"A",{href:!0});var am=o(ca);mh=s(am,"Prompt Weighting"),am.forEach(i),im.forEach(i),vh=h(st),ma=r(st,"TD",{align:!0});var rm=o(ma);gh=s(rm,"\u2705"),rm.forEach(i),_h=h(st),va=r(st,"TD",{align:!0});var om=o(va);Eh=s(om,"\u274C"),om.forEach(i),wh=h(st),Yr=r(st,"TD",{align:!0}),o(Yr).forEach(i),st.forEach(i),bh=h(v),B=r(v,"TR",{});var ft=o(B);ga=r(ft,"TD",{align:!0});var nm=o(ga);_a=r(nm,"A",{href:!0});var lm=o(_a);xh=s(lm,"Custom Diffusion"),lm.forEach(i),nm.forEach(i),Ph=h(ft),Ea=r(ft,"TD",{align:!0});var sm=o(Ea);yh=s(sm,"\u274C"),sm.forEach(i),Ah=h(ft),wa=r(ft,"TD",{align:!0});var fm=o(wa);$h=s(fm,"\u2705"),fm.forEach(i),Th=h(ft),Kr=r(ft,"TD",{align:!0}),o(Kr).forEach(i),ft.forEach(i),Dh=h(v),F=r(v,"TR",{});var ht=o(F);ba=r(ht,"TD",{align:!0});var hm=o(ba);xa=r(hm,"A",{href:!0});var pm=o(xa);kh=s(pm,"Model Editing"),pm.forEach(i),hm.forEach(i),Sh=h(ht),Pa=r(ht,"TD",{align:!0});var dm=o(Pa);Ih=s(dm,"\u2705"),dm.forEach(i),Nh=h(ht),ya=r(ht,"TD",{align:!0});var um=o(ya);Gh=s(um,"\u274C"),um.forEach(i),qh=h(ht),Qr=r(ht,"TD",{align:!0}),o(Qr).forEach(i),ht.forEach(i),Ch=h(v),W=r(v,"TR",{});var pt=o(W);Aa=r(pt,"TD",{align:!0});var cm=o(Aa);$a=r(cm,"A",{href:!0});var mm=o($a);Rh=s(mm,"DiffEdit"),mm.forEach(i),cm.forEach(i),Mh=h(pt),Ta=r(pt,"TD",{align:!0});var vm=o(Ta);Lh=s(vm,"\u2705"),vm.forEach(i),jh=h(pt),Da=r(pt,"TD",{align:!0});var gm=o(Da);zh=s(gm,"\u274C"),gm.forEach(i),Hh=h(pt),Xr=r(pt,"TD",{align:!0}),o(Xr).forEach(i),pt.forEach(i),v.forEach(i),rl.forEach(i),Fo=h(t),J=r(t,"H2",{class:!0});var nl=o(J);ue=r(nl,"A",{id:!0,class:!0,href:!0});var _m=o(ue);eo=r(_m,"SPAN",{});var Em=o(eo);w(mt.$$.fragment,Em),Em.forEach(i),_m.forEach(i),Zh=h(nl),to=r(nl,"SPAN",{});var wm=o(to);Bh=s(wm,"Instruct Pix2Pix"),wm.forEach(i),nl.forEach(i),Wo=h(t),ka=r(t,"P",{});var bm=o(ka);vt=r(bm,"A",{href:!0,rel:!0});var xm=o(vt);Fh=s(xm,"Paper"),xm.forEach(i),bm.forEach(i),Uo=h(t),Y=r(t,"P",{});var Go=o(Y);Sa=r(Go,"A",{href:!0});var Pm=o(Sa);Wh=s(Pm,"Instruct Pix2Pix"),Pm.forEach(i),Uh=s(Go,` is fine-tuned from stable diffusion to support editing input images. It takes as inputs an image and a prompt describing an edit, and it outputs the edited image.
Instruct Pix2Pix has been explicitly trained to work well with `),gt=r(Go,"A",{href:!0,rel:!0});var ym=o(gt);Oh=s(ym,"InstructGPT"),ym.forEach(i),Vh=s(Go,"-like prompts."),Go.forEach(i),Oo=h(t),ce=r(t,"P",{});var ll=o(ce);Jh=s(ll,"See "),Ia=r(ll,"A",{href:!0});var Am=o(Ia);Yh=s(Am,"here"),Am.forEach(i),Kh=s(ll," for more information on how to use it."),ll.forEach(i),Vo=h(t),K=r(t,"H2",{class:!0});var sl=o(K);me=r(sl,"A",{id:!0,class:!0,href:!0});var $m=o(me);io=r($m,"SPAN",{});var Tm=o(io);w(_t.$$.fragment,Tm),Tm.forEach(i),$m.forEach(i),Qh=h(sl),ao=r(sl,"SPAN",{});var Dm=o(ao);Xh=s(Dm,"Pix2Pix Zero"),Dm.forEach(i),sl.forEach(i),Jo=h(t),Na=r(t,"P",{});var km=o(Na);Et=r(km,"A",{href:!0,rel:!0});var Sm=o(Et);ep=s(Sm,"Paper"),Sm.forEach(i),km.forEach(i),Yo=h(t),wt=r(t,"P",{});var uu=o(wt);Ga=r(uu,"A",{href:!0});var Im=o(Ga);tp=s(Im,"Pix2Pix Zero"),Im.forEach(i),ip=s(uu," allows modifying an image so that one concept or subject is translated to another one while preserving general image semantics."),uu.forEach(i),Ko=h(t),qa=r(t,"P",{});var Nm=o(qa);ap=s(Nm,"The denoising process is guided from one conceptual embedding towards another conceptual embedding. The intermediate latents are optimized during the denoising process to push the attention maps towards reference attention maps. The reference attention maps are from the denoising process of the input image and are used to encourage semantic preservation."),Nm.forEach(i),Qo=h(t),Ca=r(t,"P",{});var Gm=o(Ca);rp=s(Gm,"Pix2Pix Zero can be used both to edit synthetic images as well as real images."),Gm.forEach(i),Xo=h(t),ve=r(t,"UL",{});var fl=o(ve);bt=r(fl,"LI",{});var hl=o(bt);op=s(hl,`To edit synthetic images, one first generates an image given a caption.
Next, we generate image captions for the concept that shall be edited and for the new target concept. We can use a model like `),xt=r(hl,"A",{href:!0,rel:!0});var qm=o(xt);np=s(qm,"Flan-T5"),qm.forEach(i),lp=s(hl," for this purpose. Then, \u201Cmean\u201D prompt embeddings for both the source and target concepts are created via the text encoder. Finally, the pix2pix-zero algorithm is used to edit the synthetic image."),hl.forEach(i),sp=h(fl),Pt=r(fl,"LI",{});var pl=o(Pt);fp=s(pl,"To edit a real image, one first generates an image caption using a model like "),yt=r(pl,"A",{href:!0,rel:!0});var Cm=o(yt);hp=s(Cm,"BLIP"),Cm.forEach(i),pp=s(pl,". Then one applies ddim inversion on the prompt and image to generate \u201Cinverse\u201D latents. Similar to before, \u201Cmean\u201D prompt embeddings for both source and target concepts are created and finally the pix2pix-zero algorithm in combination with the \u201Cinverse\u201D latents is used to edit the image."),pl.forEach(i),fl.forEach(i),en=h(t),w(ge.$$.fragment,t),tn=h(t),_e=r(t,"P",{});var dl=o(_e);dp=s(dl,`As mentioned above, Pix2Pix Zero includes optimizing the latents (and not any of the UNet, VAE, or the text encoder) to steer the generation toward a specific concept. This means that the overall
pipeline might require more memory than a standard `),Ra=r(dl,"A",{href:!0});var Rm=o(Ra);up=s(Rm,"StableDiffusionPipeline"),Rm.forEach(i),cp=s(dl,"."),dl.forEach(i),an=h(t),Ee=r(t,"P",{});var ul=o(Ee);mp=s(ul,"See "),Ma=r(ul,"A",{href:!0});var Mm=o(Ma);vp=s(Mm,"here"),Mm.forEach(i),gp=s(ul," for more information on how to use it."),ul.forEach(i),rn=h(t),Q=r(t,"H2",{class:!0});var cl=o(Q);we=r(cl,"A",{id:!0,class:!0,href:!0});var Lm=o(we);ro=r(Lm,"SPAN",{});var jm=o(ro);w(At.$$.fragment,jm),jm.forEach(i),Lm.forEach(i),_p=h(cl),oo=r(cl,"SPAN",{});var zm=o(oo);Ep=s(zm,"Attend and Excite"),zm.forEach(i),cl.forEach(i),on=h(t),La=r(t,"P",{});var Hm=o(La);$t=r(Hm,"A",{href:!0,rel:!0});var Zm=o($t);wp=s(Zm,"Paper"),Zm.forEach(i),Hm.forEach(i),nn=h(t),Tt=r(t,"P",{});var cu=o(Tt);ja=r(cu,"A",{href:!0});var Bm=o(ja);bp=s(Bm,"Attend and Excite"),Bm.forEach(i),xp=s(cu," allows subjects in the prompt to be faithfully represented in the final image."),cu.forEach(i),ln=h(t),za=r(t,"P",{});var Fm=o(za);Pp=s(Fm,"A set of token indices are given as input, corresponding to the subjects in the prompt that need to be present in the image. During denoising, each token index is guaranteed to have a minimum attention threshold for at least one patch of the image. The intermediate latents are iteratively optimized during the denoising process to strengthen the attention of the most neglected subject token until the attention threshold is passed for all subject tokens."),Fm.forEach(i),sn=h(t),be=r(t,"P",{});var ml=o(be);yp=s(ml,"Like Pix2Pix Zero, Attend and Excite also involves a mini optimization loop (leaving the pre-trained weights untouched) in its pipeline and can require more memory than the usual "),no=r(ml,"CODE",{});var Wm=o(no);Ap=s(Wm,"StableDiffusionPipeline"),Wm.forEach(i),$p=s(ml,"."),ml.forEach(i),fn=h(t),xe=r(t,"P",{});var vl=o(xe);Tp=s(vl,"See "),Ha=r(vl,"A",{href:!0});var Um=o(Ha);Dp=s(Um,"here"),Um.forEach(i),kp=s(vl," for more information on how to use it."),vl.forEach(i),hn=h(t),X=r(t,"H2",{class:!0});var gl=o(X);Pe=r(gl,"A",{id:!0,class:!0,href:!0});var Om=o(Pe);lo=r(Om,"SPAN",{});var Vm=o(lo);w(Dt.$$.fragment,Vm),Vm.forEach(i),Om.forEach(i),Sp=h(gl),so=r(gl,"SPAN",{});var Jm=o(so);Ip=s(Jm,"Semantic Guidance (SEGA)"),Jm.forEach(i),gl.forEach(i),pn=h(t),Za=r(t,"P",{});var Ym=o(Za);kt=r(Ym,"A",{href:!0,rel:!0});var Km=o(kt);Np=s(Km,"Paper"),Km.forEach(i),Ym.forEach(i),dn=h(t),Ba=r(t,"P",{});var Qm=o(Ba);Gp=s(Qm,"SEGA allows applying or removing one or more concepts from an image. The strength of the concept can also be controlled. I.e. the smile concept can be used to incrementally increase or decrease the smile of a portrait."),Qm.forEach(i),un=h(t),Fa=r(t,"P",{});var Xm=o(Fa);qp=s(Xm,"Similar to how classifier free guidance provides guidance via empty prompt inputs, SEGA provides guidance on conceptual prompts. Multiple of these conceptual prompts can be applied simultaneously. Each conceptual prompt can either add or remove their concept depending on if the guidance is applied positively or negatively."),Xm.forEach(i),cn=h(t),Wa=r(t,"P",{});var ev=o(Wa);Cp=s(ev,"Unlike Pix2Pix Zero or Attend and Excite, SEGA directly interacts with the diffusion process instead of performing any explicit gradient-based optimization."),ev.forEach(i),mn=h(t),ye=r(t,"P",{});var _l=o(ye);Rp=s(_l,"See "),Ua=r(_l,"A",{href:!0});var tv=o(Ua);Mp=s(tv,"here"),tv.forEach(i),Lp=s(_l," for more information on how to use it."),_l.forEach(i),vn=h(t),ee=r(t,"H2",{class:!0});var El=o(ee);Ae=r(El,"A",{id:!0,class:!0,href:!0});var iv=o(Ae);fo=r(iv,"SPAN",{});var av=o(fo);w(St.$$.fragment,av),av.forEach(i),iv.forEach(i),jp=h(El),ho=r(El,"SPAN",{});var rv=o(ho);zp=s(rv,"Self-attention Guidance (SAG)"),rv.forEach(i),El.forEach(i),gn=h(t),Oa=r(t,"P",{});var ov=o(Oa);It=r(ov,"A",{href:!0,rel:!0});var nv=o(It);Hp=s(nv,"Paper"),nv.forEach(i),ov.forEach(i),_n=h(t),Nt=r(t,"P",{});var mu=o(Nt);Va=r(mu,"A",{href:!0});var lv=o(Va);Zp=s(lv,"Self-attention Guidance"),lv.forEach(i),Bp=s(mu," improves the general quality of images."),mu.forEach(i),En=h(t),Ja=r(t,"P",{});var sv=o(Ja);Fp=s(sv,"SAG provides guidance from predictions not conditioned on high-frequency details to fully conditioned images. The high frequency details are extracted out of the UNet self-attention maps."),sv.forEach(i),wn=h(t),$e=r(t,"P",{});var wl=o($e);Wp=s(wl,"See "),Ya=r(wl,"A",{href:!0});var fv=o(Ya);Up=s(fv,"here"),fv.forEach(i),Op=s(wl," for more information on how to use it."),wl.forEach(i),bn=h(t),te=r(t,"H2",{class:!0});var bl=o(te);Te=r(bl,"A",{id:!0,class:!0,href:!0});var hv=o(Te);po=r(hv,"SPAN",{});var pv=o(po);w(Gt.$$.fragment,pv),pv.forEach(i),hv.forEach(i),Vp=h(bl),uo=r(bl,"SPAN",{});var dv=o(uo);Jp=s(dv,"Depth2Image"),dv.forEach(i),bl.forEach(i),xn=h(t),Ka=r(t,"P",{});var uv=o(Ka);qt=r(uv,"A",{href:!0,rel:!0});var cv=o(qt);Yp=s(cv,"Project"),cv.forEach(i),uv.forEach(i),Pn=h(t),Ct=r(t,"P",{});var vu=o(Ct);Qa=r(vu,"A",{href:!0});var mv=o(Qa);Kp=s(mv,"Depth2Image"),mv.forEach(i),Qp=s(vu," is fine-tuned from Stable Diffusion to better preserve semantics for text guided image variation."),vu.forEach(i),yn=h(t),Xa=r(t,"P",{});var vv=o(Xa);Xp=s(vv,"It conditions on a monocular depth estimate of the original image."),vv.forEach(i),An=h(t),De=r(t,"P",{});var xl=o(De);ed=s(xl,"See "),er=r(xl,"A",{href:!0});var gv=o(er);td=s(gv,"here"),gv.forEach(i),id=s(xl," for more information on how to use it."),xl.forEach(i),$n=h(t),w(ke.$$.fragment,t),Tn=h(t),ie=r(t,"H2",{class:!0});var Pl=o(ie);Se=r(Pl,"A",{id:!0,class:!0,href:!0});var _v=o(Se);co=r(_v,"SPAN",{});var Ev=o(co);w(Rt.$$.fragment,Ev),Ev.forEach(i),_v.forEach(i),ad=h(Pl),mo=r(Pl,"SPAN",{});var wv=o(mo);rd=s(wv,"MultiDiffusion Panorama"),wv.forEach(i),Pl.forEach(i),Dn=h(t),tr=r(t,"P",{});var bv=o(tr);Mt=r(bv,"A",{href:!0,rel:!0});var xv=o(Mt);od=s(xv,"Paper"),xv.forEach(i),bv.forEach(i),kn=h(t),Ie=r(t,"P",{});var yl=o(Ie);nd=s(yl,`MultiDiffusion defines a new generation process over a pre-trained diffusion model. This process binds together multiple diffusion generation methods that can be readily applied to generate high quality and diverse images. Results adhere to user-provided controls, such as desired aspect ratio (e.g., panorama), and spatial guiding signals, ranging from tight segmentation masks to bounding boxes.
`),ir=r(yl,"A",{href:!0});var Pv=o(ir);ld=s(Pv,"MultiDiffusion Panorama"),Pv.forEach(i),sd=s(yl," allows to generate high-quality images at arbitrary aspect ratios (e.g., panoramas)."),yl.forEach(i),Sn=h(t),Ne=r(t,"P",{});var Al=o(Ne);fd=s(Al,"See "),ar=r(Al,"A",{href:!0});var yv=o(ar);hd=s(yv,"here"),yv.forEach(i),pd=s(Al," for more information on how to use it to generate panoramic images."),Al.forEach(i),In=h(t),ae=r(t,"H2",{class:!0});var $l=o(ae);Ge=r($l,"A",{id:!0,class:!0,href:!0});var Av=o(Ge);vo=r(Av,"SPAN",{});var $v=o(vo);w(Lt.$$.fragment,$v),$v.forEach(i),Av.forEach(i),dd=h($l),go=r($l,"SPAN",{});var Tv=o(go);ud=s(Tv,"Fine-tuning your own models"),Tv.forEach(i),$l.forEach(i),Nn=h(t),rr=r(t,"P",{});var Dv=o(rr);cd=s(Dv,"In addition to pre-trained models, Diffusers has training scripts for fine-tuning models on user-provided data."),Dv.forEach(i),Gn=h(t),re=r(t,"H2",{class:!0});var Tl=o(re);qe=r(Tl,"A",{id:!0,class:!0,href:!0});var kv=o(qe);_o=r(kv,"SPAN",{});var Sv=o(_o);w(jt.$$.fragment,Sv),Sv.forEach(i),kv.forEach(i),md=h(Tl),Eo=r(Tl,"SPAN",{});var Iv=o(Eo);vd=s(Iv,"DreamBooth"),Iv.forEach(i),Tl.forEach(i),qn=h(t),zt=r(t,"P",{});var gu=o(zt);or=r(gu,"A",{href:!0});var Nv=o(or);gd=s(Nv,"DreamBooth"),Nv.forEach(i),_d=s(gu," fine-tunes a model to teach it about a new subject. I.e. a few pictures of a person can be used to generate images of that person in different styles."),gu.forEach(i),Cn=h(t),Ce=r(t,"P",{});var Dl=o(Ce);Ed=s(Dl,"See "),nr=r(Dl,"A",{href:!0});var Gv=o(nr);wd=s(Gv,"here"),Gv.forEach(i),bd=s(Dl," for more information on how to use it."),Dl.forEach(i),Rn=h(t),oe=r(t,"H2",{class:!0});var kl=o(oe);Re=r(kl,"A",{id:!0,class:!0,href:!0});var qv=o(Re);wo=r(qv,"SPAN",{});var Cv=o(wo);w(Ht.$$.fragment,Cv),Cv.forEach(i),qv.forEach(i),xd=h(kl),bo=r(kl,"SPAN",{});var Rv=o(bo);Pd=s(Rv,"Textual Inversion"),Rv.forEach(i),kl.forEach(i),Mn=h(t),Zt=r(t,"P",{});var _u=o(Zt);lr=r(_u,"A",{href:!0});var Mv=o(lr);yd=s(Mv,"Textual Inversion"),Mv.forEach(i),Ad=s(_u," fine-tunes a model to teach it about a new concept. I.e. a few pictures of a style of artwork can be used to generate images in that style."),_u.forEach(i),Ln=h(t),Me=r(t,"P",{});var Sl=o(Me);$d=s(Sl,"See "),sr=r(Sl,"A",{href:!0});var Lv=o(sr);Td=s(Lv,"here"),Lv.forEach(i),Dd=s(Sl," for more information on how to use it."),Sl.forEach(i),jn=h(t),ne=r(t,"H2",{class:!0});var Il=o(ne);Le=r(Il,"A",{id:!0,class:!0,href:!0});var jv=o(Le);xo=r(jv,"SPAN",{});var zv=o(xo);w(Bt.$$.fragment,zv),zv.forEach(i),jv.forEach(i),kd=h(Il),Po=r(Il,"SPAN",{});var Hv=o(Po);Sd=s(Hv,"ControlNet"),Hv.forEach(i),Il.forEach(i),zn=h(t),fr=r(t,"P",{});var Zv=o(fr);Ft=r(Zv,"A",{href:!0,rel:!0});var Bv=o(Ft);Id=s(Bv,"Paper"),Bv.forEach(i),Zv.forEach(i),Hn=h(t),Wt=r(t,"P",{});var Eu=o(Wt);hr=r(Eu,"A",{href:!0});var Fv=o(hr);Nd=s(Fv,"ControlNet"),Fv.forEach(i),Gd=s(Eu,` is an auxiliary network which adds an extra condition.
There are 8 canonical pre-trained ControlNets trained on different conditionings such as edge detection, scribbles,
depth maps, and semantic segmentations.`),Eu.forEach(i),Zn=h(t),je=r(t,"P",{});var Nl=o(je);qd=s(Nl,"See "),pr=r(Nl,"A",{href:!0});var Wv=o(pr);Cd=s(Wv,"here"),Wv.forEach(i),Rd=s(Nl," for more information on how to use it."),Nl.forEach(i),Bn=h(t),le=r(t,"H2",{class:!0});var Gl=o(le);ze=r(Gl,"A",{id:!0,class:!0,href:!0});var Uv=o(ze);yo=r(Uv,"SPAN",{});var Ov=o(yo);w(Ut.$$.fragment,Ov),Ov.forEach(i),Uv.forEach(i),Md=h(Gl),Ao=r(Gl,"SPAN",{});var Vv=o(Ao);Ld=s(Vv,"Prompt Weighting"),Vv.forEach(i),Gl.forEach(i),Fn=h(t),dr=r(t,"P",{});var Jv=o(dr);jd=s(Jv,`Prompt weighting is a simple technique that puts more attention weight on certain parts of the text
input.`),Jv.forEach(i),Wn=h(t),He=r(t,"P",{});var ql=o(He);zd=s(ql,"For a more in-detail explanation and examples, see "),ur=r(ql,"A",{href:!0});var Yv=o(ur);Hd=s(Yv,"here"),Yv.forEach(i),Zd=s(ql,"."),ql.forEach(i),Un=h(t),se=r(t,"H2",{class:!0});var Cl=o(se);Ze=r(Cl,"A",{id:!0,class:!0,href:!0});var Kv=o(Ze);$o=r(Kv,"SPAN",{});var Qv=o($o);w(Ot.$$.fragment,Qv),Qv.forEach(i),Kv.forEach(i),Bd=h(Cl),To=r(Cl,"SPAN",{});var Xv=o(To);Fd=s(Xv,"Custom Diffusion"),Xv.forEach(i),Cl.forEach(i),On=h(t),Vt=r(t,"P",{});var wu=o(Vt);cr=r(wu,"A",{href:!0});var e2=o(cr);Wd=s(e2,"Custom Diffusion"),e2.forEach(i),Ud=s(wu,` only fine-tunes the cross-attention maps of a pre-trained
text-to-image diffusion model. It also allows for additionally performing textual inversion. It supports
multi-concept training by design. Like DreamBooth and Textual Inversion, Custom Diffusion is also used to
teach a pre-trained text-to-image diffusion model about new concepts to generate outputs involving the
concept(s) of interest.`),wu.forEach(i),Vn=h(t),Be=r(t,"P",{});var Rl=o(Be);Od=s(Rl,"For more details, check out our "),mr=r(Rl,"A",{href:!0});var t2=o(mr);Vd=s(t2,"official doc"),t2.forEach(i),Jd=s(Rl,"."),Rl.forEach(i),Jn=h(t),fe=r(t,"H2",{class:!0});var Ml=o(fe);Fe=r(Ml,"A",{id:!0,class:!0,href:!0});var i2=o(Fe);Do=r(i2,"SPAN",{});var a2=o(Do);w(Jt.$$.fragment,a2),a2.forEach(i),i2.forEach(i),Yd=h(Ml),ko=r(Ml,"SPAN",{});var r2=o(ko);Kd=s(r2,"Model Editing"),r2.forEach(i),Ml.forEach(i),Yn=h(t),vr=r(t,"P",{});var o2=o(vr);Yt=r(o2,"A",{href:!0,rel:!0});var n2=o(Yt);Qd=s(n2,"Paper"),n2.forEach(i),o2.forEach(i),Kn=h(t),We=r(t,"P",{});var Ll=o(We);Xd=s(Ll,"The "),gr=r(Ll,"A",{href:!0});var l2=o(gr);eu=s(l2,"text-to-image model editing pipeline"),l2.forEach(i),tu=s(Ll,` helps you mitigate some of the incorrect implicit assumptions a pre-trained text-to-image
diffusion model might make about the subjects present in the input prompt. For example, if you prompt Stable Diffusion to generate images for \u201CA pack of roses\u201D, the roses in the generated images
are more likely to be red. This pipeline helps you change that assumption.`),Ll.forEach(i),Qn=h(t),Ue=r(t,"P",{});var jl=o(Ue);iu=s(jl,"To know more details, check out the "),_r=r(jl,"A",{href:!0});var s2=o(_r);au=s(s2,"official doc"),s2.forEach(i),ru=s(jl,"."),jl.forEach(i),Xn=h(t),he=r(t,"H2",{class:!0});var zl=o(he);Oe=r(zl,"A",{id:!0,class:!0,href:!0});var f2=o(Oe);So=r(f2,"SPAN",{});var h2=o(So);w(Kt.$$.fragment,h2),h2.forEach(i),f2.forEach(i),ou=h(zl),Io=r(zl,"SPAN",{});var p2=o(Io);nu=s(p2,"DiffEdit"),p2.forEach(i),zl.forEach(i),el=h(t),Er=r(t,"P",{});var d2=o(Er);Qt=r(d2,"A",{href:!0,rel:!0});var u2=o(Qt);lu=s(u2,"Paper"),u2.forEach(i),d2.forEach(i),tl=h(t),Xt=r(t,"P",{});var bu=o(Xt);wr=r(bu,"A",{href:!0});var c2=o(wr);su=s(c2,"DiffEdit"),c2.forEach(i),fu=s(bu,` allows for semantic editing of input images along with
input prompts while preserving the original input images as much as possible.`),bu.forEach(i),il=h(t),Ve=r(t,"P",{});var Hl=o(Ve);hu=s(Hl,"To know more details, check out the "),br=r(Hl,"A",{href:!0});var m2=o(br);pu=s(m2,"official doc"),m2.forEach(i),du=s(Hl,"."),Hl.forEach(i),this.h()},h(){n(g,"name","hf:doc:metadata"),n(g,"content",JSON.stringify(A2)),n(A,"id","controlled-generation"),n(A,"class","header-link block pr-1.5 text-lg no-hover:hidden with-hover:absolute with-hover:p-1.5 with-hover:opacity-0 with-hover:group-hover:opacity-100 with-hover:right-full"),n(A,"href","#controlled-generation"),n(_,"class","relative group"),n(dt,"href","https://discuss.huggingface.co/"),n(dt,"rel","nofollow"),n(ut,"href","https://github.com/huggingface/diffusers/issues"),n(ut,"rel","nofollow"),n(si,"href","#instruct-pix2pix"),n(fi,"href","#pix2pixzero"),n(hi,"href","#attend-and-excite"),n(pi,"href","#semantic-guidance"),n(di,"href","#self-attention-guidance"),n(ui,"href","#depth2image"),n(ci,"href","#multidiffusion-panorama"),n(mi,"href","#dreambooth"),n(vi,"href","#textual-inversion"),n(gi,"href","#controlnet"),n(_i,"href","#prompt-weighting"),n(Ei,"href","#custom-diffusion"),n(wi,"href","#model-editing"),n(bi,"href","#diffedit"),n(Pi,"align","center"),n(yi,"align","center"),n(Ai,"align","center"),n($i,"align","center"),n(Di,"href","#instruct-pix2pix"),n(Ti,"align","center"),n(ki,"align","center"),n(Si,"align","center"),n(S,"align","center"),n(Ni,"href","#pix2pixzero"),n(Ii,"align","center"),n(Gi,"align","center"),n(qi,"align","center"),n(Zr,"align","center"),n(Ri,"href","#attend-and-excite"),n(Ci,"align","center"),n(Mi,"align","center"),n(Li,"align","center"),n(Br,"align","center"),n(zi,"href","#semantic-guidance"),n(ji,"align","center"),n(Hi,"align","center"),n(Zi,"align","center"),n(Fr,"align","center"),n(Fi,"href","#self-attention-guidance"),n(Bi,"align","center"),n(Wi,"align","center"),n(Ui,"align","center"),n(Wr,"align","center"),n(Vi,"href","#depth2image"),n(Oi,"align","center"),n(Ji,"align","center"),n(Yi,"align","center"),n(Ur,"align","center"),n(Qi,"href","#multidiffusion-panorama"),n(Ki,"align","center"),n(Xi,"align","center"),n(ea,"align","center"),n(Or,"align","center"),n(ia,"href","#dreambooth"),n(ta,"align","center"),n(aa,"align","center"),n(ra,"align","center"),n(Vr,"align","center"),n(na,"href","#textual-inversion"),n(oa,"align","center"),n(la,"align","center"),n(sa,"align","center"),n(Jr,"align","center"),n(ha,"href","#controlnet"),n(fa,"align","center"),n(pa,"align","center"),n(da,"align","center"),n(O,"align","center"),n(ca,"href","#prompt-weighting"),n(ua,"align","center"),n(ma,"align","center"),n(va,"align","center"),n(Yr,"align","center"),n(_a,"href","#custom-diffusion"),n(ga,"align","center"),n(Ea,"align","center"),n(wa,"align","center"),n(Kr,"align","center"),n(xa,"href","#model-editing"),n(ba,"align","center"),n(Pa,"align","center"),n(ya,"align","center"),n(Qr,"align","center"),n($a,"href","#diffedit"),n(Aa,"align","center"),n(Ta,"align","center"),n(Da,"align","center"),n(Xr,"align","center"),n(ue,"id","instruct-pix2pix"),n(ue,"class","header-link block pr-1.5 text-lg no-hover:hidden with-hover:absolute with-hover:p-1.5 with-hover:opacity-0 with-hover:group-hover:opacity-100 with-hover:right-full"),n(ue,"href","#instruct-pix2pix"),n(J,"class","relative group"),n(vt,"href","https://arxiv.org/abs/2211.09800"),n(vt,"rel","nofollow"),n(Sa,"href","../api/pipelines/stable_diffusion/pix2pix"),n(gt,"href","https://openai.com/blog/instruction-following/"),n(gt,"rel","nofollow"),n(Ia,"href","../api/pipelines/stable_diffusion/pix2pix"),n(me,"id","pix2pix-zero"),n(me,"class","header-link block pr-1.5 text-lg no-hover:hidden with-hover:absolute with-hover:p-1.5 with-hover:opacity-0 with-hover:group-hover:opacity-100 with-hover:right-full"),n(me,"href","#pix2pix-zero"),n(K,"class","relative group"),n(Et,"href","https://arxiv.org/abs/2302.03027"),n(Et,"rel","nofollow"),n(Ga,"href","../api/pipelines/stable_diffusion/pix2pix_zero"),n(xt,"href","https://huggingface.co/docs/transformers/model_doc/flan-t5"),n(xt,"rel","nofollow"),n(yt,"href","https://huggingface.co/docs/transformers/model_doc/blip"),n(yt,"rel","nofollow"),n(Ra,"href","../api/pipelines/stable_diffusion/text2img"),n(Ma,"href","../api/pipelines/stable_diffusion/pix2pix_zero"),n(we,"id","attend-and-excite"),n(we,"class","header-link block pr-1.5 text-lg no-hover:hidden with-hover:absolute with-hover:p-1.5 with-hover:opacity-0 with-hover:group-hover:opacity-100 with-hover:right-full"),n(we,"href","#attend-and-excite"),n(Q,"class","relative group"),n($t,"href","https://arxiv.org/abs/2301.13826"),n($t,"rel","nofollow"),n(ja,"href","../api/pipelines/stable_diffusion/attend_and_excite"),n(Ha,"href","../api/pipelines/stable_diffusion/attend_and_excite"),n(Pe,"id","semantic-guidance-sega"),n(Pe,"class","header-link block pr-1.5 text-lg no-hover:hidden with-hover:absolute with-hover:p-1.5 with-hover:opacity-0 with-hover:group-hover:opacity-100 with-hover:right-full"),n(Pe,"href","#semantic-guidance-sega"),n(X,"class","relative group"),n(kt,"href","https://arxiv.org/abs/2301.12247"),n(kt,"rel","nofollow"),n(Ua,"href","../api/pipelines/semantic_stable_diffusion"),n(Ae,"id","selfattention-guidance-sag"),n(Ae,"class","header-link block pr-1.5 text-lg no-hover:hidden with-hover:absolute with-hover:p-1.5 with-hover:opacity-0 with-hover:group-hover:opacity-100 with-hover:right-full"),n(Ae,"href","#selfattention-guidance-sag"),n(ee,"class","relative group"),n(It,"href","https://arxiv.org/abs/2210.00939"),n(It,"rel","nofollow"),n(Va,"href","../api/pipelines/stable_diffusion/self_attention_guidance"),n(Ya,"href","../api/pipelines/stable_diffusion/self_attention_guidance"),n(Te,"id","depth2image"),n(Te,"class","header-link block pr-1.5 text-lg no-hover:hidden with-hover:absolute with-hover:p-1.5 with-hover:opacity-0 with-hover:group-hover:opacity-100 with-hover:right-full"),n(Te,"href","#depth2image"),n(te,"class","relative group"),n(qt,"href","https://huggingface.co/stabilityai/stable-diffusion-2-depth"),n(qt,"rel","nofollow"),n(Qa,"href","../pipelines/stable_diffusion_2#depthtoimage"),n(er,"href","../api/pipelines/stable_diffusion_2#depthtoimage"),n(Se,"id","multidiffusion-panorama"),n(Se,"class","header-link block pr-1.5 text-lg no-hover:hidden with-hover:absolute with-hover:p-1.5 with-hover:opacity-0 with-hover:group-hover:opacity-100 with-hover:right-full"),n(Se,"href","#multidiffusion-panorama"),n(ie,"class","relative group"),n(Mt,"href","https://arxiv.org/abs/2302.08113"),n(Mt,"rel","nofollow"),n(ir,"href","../api/pipelines/stable_diffusion/panorama"),n(ar,"href","../api/pipelines/stable_diffusion/panorama"),n(Ge,"id","finetuning-your-own-models"),n(Ge,"class","header-link block pr-1.5 text-lg no-hover:hidden with-hover:absolute with-hover:p-1.5 with-hover:opacity-0 with-hover:group-hover:opacity-100 with-hover:right-full"),n(Ge,"href","#finetuning-your-own-models"),n(ae,"class","relative group"),n(qe,"id","dreambooth"),n(qe,"class","header-link block pr-1.5 text-lg no-hover:hidden with-hover:absolute with-hover:p-1.5 with-hover:opacity-0 with-hover:group-hover:opacity-100 with-hover:right-full"),n(qe,"href","#dreambooth"),n(re,"class","relative group"),n(or,"href","../training/dreambooth"),n(nr,"href","../training/dreambooth"),n(Re,"id","textual-inversion"),n(Re,"class","header-link block pr-1.5 text-lg no-hover:hidden with-hover:absolute with-hover:p-1.5 with-hover:opacity-0 with-hover:group-hover:opacity-100 with-hover:right-full"),n(Re,"href","#textual-inversion"),n(oe,"class","relative group"),n(lr,"href","../training/text_inversion"),n(sr,"href","../training/text_inversion"),n(Le,"id","controlnet"),n(Le,"class","header-link block pr-1.5 text-lg no-hover:hidden with-hover:absolute with-hover:p-1.5 with-hover:opacity-0 with-hover:group-hover:opacity-100 with-hover:right-full"),n(Le,"href","#controlnet"),n(ne,"class","relative group"),n(Ft,"href","https://arxiv.org/abs/2302.05543"),n(Ft,"rel","nofollow"),n(hr,"href","../api/pipelines/stable_diffusion/controlnet"),n(pr,"href","../api/pipelines/stable_diffusion/controlnet"),n(ze,"id","prompt-weighting"),n(ze,"class","header-link block pr-1.5 text-lg no-hover:hidden with-hover:absolute with-hover:p-1.5 with-hover:opacity-0 with-hover:group-hover:opacity-100 with-hover:right-full"),n(ze,"href","#prompt-weighting"),n(le,"class","relative group"),n(ur,"href","../using-diffusers/weighted_prompts"),n(Ze,"id","custom-diffusion"),n(Ze,"class","header-link block pr-1.5 text-lg no-hover:hidden with-hover:absolute with-hover:p-1.5 with-hover:opacity-0 with-hover:group-hover:opacity-100 with-hover:right-full"),n(Ze,"href","#custom-diffusion"),n(se,"class","relative group"),n(cr,"href","../training/custom_diffusion"),n(mr,"href","../training/custom_diffusion"),n(Fe,"id","model-editing"),n(Fe,"class","header-link block pr-1.5 text-lg no-hover:hidden with-hover:absolute with-hover:p-1.5 with-hover:opacity-0 with-hover:group-hover:opacity-100 with-hover:right-full"),n(Fe,"href","#model-editing"),n(fe,"class","relative group"),n(Yt,"href","https://arxiv.org/abs/2303.08084"),n(Yt,"rel","nofollow"),n(gr,"href","../api/pipelines/stable_diffusion/model_editing"),n(_r,"href","../api/pipelines/stable_diffusion/model_editing"),n(Oe,"id","diffedit"),n(Oe,"class","header-link block pr-1.5 text-lg no-hover:hidden with-hover:absolute with-hover:p-1.5 with-hover:opacity-0 with-hover:group-hover:opacity-100 with-hover:right-full"),n(Oe,"href","#diffedit"),n(he,"class","relative group"),n(Qt,"href","https://arxiv.org/abs/2210.11427"),n(Qt,"rel","nofollow"),n(wr,"href","../api/pipelines/stable_diffusion/diffedit"),n(br,"href","../api/pipelines/stable_diffusion/model_editing")},m(t,p){e(document.head,g),d(t,D,p),d(t,_,p),e(_,A),e(A,V),b(T,V,null),e(_,U),e(_,pe),e(pe,Zl),d(t,qo,p),d(t,ii,p),e(ii,Bl),d(t,Co,p),d(t,ai,p),e(ai,Fl),d(t,Ro,p),d(t,ri,p),e(ri,Wl),d(t,Mo,p),d(t,k,p),e(k,Ul),e(k,Pr),e(Pr,Ol),e(k,Vl),e(k,dt),e(dt,Jl),e(k,Yl),e(k,ut),e(ut,Kl),e(k,Ql),d(t,Lo,p),d(t,oi,p),e(oi,Xl),d(t,jo,p),d(t,ni,p),e(ni,es),d(t,zo,p),d(t,li,p),e(li,ts),d(t,Ho,p),d(t,u,p),e(u,yr),e(yr,si),e(si,is),e(u,as),e(u,Ar),e(Ar,fi),e(fi,rs),e(u,os),e(u,$r),e($r,hi),e(hi,ns),e(u,ls),e(u,Tr),e(Tr,pi),e(pi,ss),e(u,fs),e(u,Dr),e(Dr,di),e(di,hs),e(u,ps),e(u,kr),e(kr,ui),e(ui,ds),e(u,us),e(u,Sr),e(Sr,ci),e(ci,cs),e(u,ms),e(u,Ir),e(Ir,mi),e(mi,vs),e(u,gs),e(u,Nr),e(Nr,vi),e(vi,_s),e(u,Es),e(u,Gr),e(Gr,gi),e(gi,ws),e(u,bs),e(u,qr),e(qr,_i),e(_i,xs),e(u,Ps),e(u,Cr),e(Cr,Ei),e(Ei,ys),e(u,As),e(u,Rr),e(Rr,wi),e(wi,$s),e(u,Ts),e(u,Mr),e(Mr,bi),e(bi,Ds),d(t,Zo,p),d(t,xi,p),e(xi,ks),d(t,Bo,p),d(t,de,p),e(de,Lr),e(Lr,I),e(I,Pi),e(Pi,jr),e(jr,Ss),e(I,Is),e(I,yi),e(yi,zr),e(zr,Ns),e(I,Gs),e(I,Ai),e(Ai,ct),e(ct,qs),e(ct,Cs),e(ct,Rs),e(I,Ms),e(I,$i),e($i,Hr),e(Hr,Ls),e(de,js),e(de,c),e(c,N),e(N,Ti),e(Ti,Di),e(Di,zs),e(N,Hs),e(N,ki),e(ki,Zs),e(N,Bs),e(N,Si),e(Si,Fs),e(N,Ws),e(N,S),e(S,Us),e(S,Os),e(S,Vs),e(S,Js),e(S,Ys),e(S,Ks),e(S,Qs),e(c,Xs),e(c,G),e(G,Ii),e(Ii,Ni),e(Ni,ef),e(G,tf),e(G,Gi),e(Gi,af),e(G,rf),e(G,qi),e(qi,of),e(G,nf),e(G,Zr),e(c,lf),e(c,q),e(q,Ci),e(Ci,Ri),e(Ri,sf),e(q,ff),e(q,Mi),e(Mi,hf),e(q,pf),e(q,Li),e(Li,df),e(q,uf),e(q,Br),e(c,cf),e(c,C),e(C,ji),e(ji,zi),e(zi,mf),e(C,vf),e(C,Hi),e(Hi,gf),e(C,_f),e(C,Zi),e(Zi,Ef),e(C,wf),e(C,Fr),e(c,bf),e(c,R),e(R,Bi),e(Bi,Fi),e(Fi,xf),e(R,Pf),e(R,Wi),e(Wi,yf),e(R,Af),e(R,Ui),e(Ui,$f),e(R,Tf),e(R,Wr),e(c,Df),e(c,M),e(M,Oi),e(Oi,Vi),e(Vi,kf),e(M,Sf),e(M,Ji),e(Ji,If),e(M,Nf),e(M,Yi),e(Yi,Gf),e(M,qf),e(M,Ur),e(c,Cf),e(c,L),e(L,Ki),e(Ki,Qi),e(Qi,Rf),e(L,Mf),e(L,Xi),e(Xi,Lf),e(L,jf),e(L,ea),e(ea,zf),e(L,Hf),e(L,Or),e(c,Zf),e(c,j),e(j,ta),e(ta,ia),e(ia,Bf),e(j,Ff),e(j,aa),e(aa,Wf),e(j,Uf),e(j,ra),e(ra,Of),e(j,Vf),e(j,Vr),e(c,Jf),e(c,z),e(z,oa),e(oa,na),e(na,Yf),e(z,Kf),e(z,la),e(la,Qf),e(z,Xf),e(z,sa),e(sa,eh),e(z,th),e(z,Jr),e(c,ih),e(c,H),e(H,fa),e(fa,ha),e(ha,ah),e(H,rh),e(H,pa),e(pa,oh),e(H,nh),e(H,da),e(da,lh),e(H,sh),e(H,O),e(O,fh),e(O,hh),e(O,ph),e(O,dh),e(O,uh),e(c,ch),e(c,Z),e(Z,ua),e(ua,ca),e(ca,mh),e(Z,vh),e(Z,ma),e(ma,gh),e(Z,_h),e(Z,va),e(va,Eh),e(Z,wh),e(Z,Yr),e(c,bh),e(c,B),e(B,ga),e(ga,_a),e(_a,xh),e(B,Ph),e(B,Ea),e(Ea,yh),e(B,Ah),e(B,wa),e(wa,$h),e(B,Th),e(B,Kr),e(c,Dh),e(c,F),e(F,ba),e(ba,xa),e(xa,kh),e(F,Sh),e(F,Pa),e(Pa,Ih),e(F,Nh),e(F,ya),e(ya,Gh),e(F,qh),e(F,Qr),e(c,Ch),e(c,W),e(W,Aa),e(Aa,$a),e($a,Rh),e(W,Mh),e(W,Ta),e(Ta,Lh),e(W,jh),e(W,Da),e(Da,zh),e(W,Hh),e(W,Xr),d(t,Fo,p),d(t,J,p),e(J,ue),e(ue,eo),b(mt,eo,null),e(J,Zh),e(J,to),e(to,Bh),d(t,Wo,p),d(t,ka,p),e(ka,vt),e(vt,Fh),d(t,Uo,p),d(t,Y,p),e(Y,Sa),e(Sa,Wh),e(Y,Uh),e(Y,gt),e(gt,Oh),e(Y,Vh),d(t,Oo,p),d(t,ce,p),e(ce,Jh),e(ce,Ia),e(Ia,Yh),e(ce,Kh),d(t,Vo,p),d(t,K,p),e(K,me),e(me,io),b(_t,io,null),e(K,Qh),e(K,ao),e(ao,Xh),d(t,Jo,p),d(t,Na,p),e(Na,Et),e(Et,ep),d(t,Yo,p),d(t,wt,p),e(wt,Ga),e(Ga,tp),e(wt,ip),d(t,Ko,p),d(t,qa,p),e(qa,ap),d(t,Qo,p),d(t,Ca,p),e(Ca,rp),d(t,Xo,p),d(t,ve,p),e(ve,bt),e(bt,op),e(bt,xt),e(xt,np),e(bt,lp),e(ve,sp),e(ve,Pt),e(Pt,fp),e(Pt,yt),e(yt,hp),e(Pt,pp),d(t,en,p),b(ge,t,p),d(t,tn,p),d(t,_e,p),e(_e,dp),e(_e,Ra),e(Ra,up),e(_e,cp),d(t,an,p),d(t,Ee,p),e(Ee,mp),e(Ee,Ma),e(Ma,vp),e(Ee,gp),d(t,rn,p),d(t,Q,p),e(Q,we),e(we,ro),b(At,ro,null),e(Q,_p),e(Q,oo),e(oo,Ep),d(t,on,p),d(t,La,p),e(La,$t),e($t,wp),d(t,nn,p),d(t,Tt,p),e(Tt,ja),e(ja,bp),e(Tt,xp),d(t,ln,p),d(t,za,p),e(za,Pp),d(t,sn,p),d(t,be,p),e(be,yp),e(be,no),e(no,Ap),e(be,$p),d(t,fn,p),d(t,xe,p),e(xe,Tp),e(xe,Ha),e(Ha,Dp),e(xe,kp),d(t,hn,p),d(t,X,p),e(X,Pe),e(Pe,lo),b(Dt,lo,null),e(X,Sp),e(X,so),e(so,Ip),d(t,pn,p),d(t,Za,p),e(Za,kt),e(kt,Np),d(t,dn,p),d(t,Ba,p),e(Ba,Gp),d(t,un,p),d(t,Fa,p),e(Fa,qp),d(t,cn,p),d(t,Wa,p),e(Wa,Cp),d(t,mn,p),d(t,ye,p),e(ye,Rp),e(ye,Ua),e(Ua,Mp),e(ye,Lp),d(t,vn,p),d(t,ee,p),e(ee,Ae),e(Ae,fo),b(St,fo,null),e(ee,jp),e(ee,ho),e(ho,zp),d(t,gn,p),d(t,Oa,p),e(Oa,It),e(It,Hp),d(t,_n,p),d(t,Nt,p),e(Nt,Va),e(Va,Zp),e(Nt,Bp),d(t,En,p),d(t,Ja,p),e(Ja,Fp),d(t,wn,p),d(t,$e,p),e($e,Wp),e($e,Ya),e(Ya,Up),e($e,Op),d(t,bn,p),d(t,te,p),e(te,Te),e(Te,po),b(Gt,po,null),e(te,Vp),e(te,uo),e(uo,Jp),d(t,xn,p),d(t,Ka,p),e(Ka,qt),e(qt,Yp),d(t,Pn,p),d(t,Ct,p),e(Ct,Qa),e(Qa,Kp),e(Ct,Qp),d(t,yn,p),d(t,Xa,p),e(Xa,Xp),d(t,An,p),d(t,De,p),e(De,ed),e(De,er),e(er,td),e(De,id),d(t,$n,p),b(ke,t,p),d(t,Tn,p),d(t,ie,p),e(ie,Se),e(Se,co),b(Rt,co,null),e(ie,ad),e(ie,mo),e(mo,rd),d(t,Dn,p),d(t,tr,p),e(tr,Mt),e(Mt,od),d(t,kn,p),d(t,Ie,p),e(Ie,nd),e(Ie,ir),e(ir,ld),e(Ie,sd),d(t,Sn,p),d(t,Ne,p),e(Ne,fd),e(Ne,ar),e(ar,hd),e(Ne,pd),d(t,In,p),d(t,ae,p),e(ae,Ge),e(Ge,vo),b(Lt,vo,null),e(ae,dd),e(ae,go),e(go,ud),d(t,Nn,p),d(t,rr,p),e(rr,cd),d(t,Gn,p),d(t,re,p),e(re,qe),e(qe,_o),b(jt,_o,null),e(re,md),e(re,Eo),e(Eo,vd),d(t,qn,p),d(t,zt,p),e(zt,or),e(or,gd),e(zt,_d),d(t,Cn,p),d(t,Ce,p),e(Ce,Ed),e(Ce,nr),e(nr,wd),e(Ce,bd),d(t,Rn,p),d(t,oe,p),e(oe,Re),e(Re,wo),b(Ht,wo,null),e(oe,xd),e(oe,bo),e(bo,Pd),d(t,Mn,p),d(t,Zt,p),e(Zt,lr),e(lr,yd),e(Zt,Ad),d(t,Ln,p),d(t,Me,p),e(Me,$d),e(Me,sr),e(sr,Td),e(Me,Dd),d(t,jn,p),d(t,ne,p),e(ne,Le),e(Le,xo),b(Bt,xo,null),e(ne,kd),e(ne,Po),e(Po,Sd),d(t,zn,p),d(t,fr,p),e(fr,Ft),e(Ft,Id),d(t,Hn,p),d(t,Wt,p),e(Wt,hr),e(hr,Nd),e(Wt,Gd),d(t,Zn,p),d(t,je,p),e(je,qd),e(je,pr),e(pr,Cd),e(je,Rd),d(t,Bn,p),d(t,le,p),e(le,ze),e(ze,yo),b(Ut,yo,null),e(le,Md),e(le,Ao),e(Ao,Ld),d(t,Fn,p),d(t,dr,p),e(dr,jd),d(t,Wn,p),d(t,He,p),e(He,zd),e(He,ur),e(ur,Hd),e(He,Zd),d(t,Un,p),d(t,se,p),e(se,Ze),e(Ze,$o),b(Ot,$o,null),e(se,Bd),e(se,To),e(To,Fd),d(t,On,p),d(t,Vt,p),e(Vt,cr),e(cr,Wd),e(Vt,Ud),d(t,Vn,p),d(t,Be,p),e(Be,Od),e(Be,mr),e(mr,Vd),e(Be,Jd),d(t,Jn,p),d(t,fe,p),e(fe,Fe),e(Fe,Do),b(Jt,Do,null),e(fe,Yd),e(fe,ko),e(ko,Kd),d(t,Yn,p),d(t,vr,p),e(vr,Yt),e(Yt,Qd),d(t,Kn,p),d(t,We,p),e(We,Xd),e(We,gr),e(gr,eu),e(We,tu),d(t,Qn,p),d(t,Ue,p),e(Ue,iu),e(Ue,_r),e(_r,au),e(Ue,ru),d(t,Xn,p),d(t,he,p),e(he,Oe),e(Oe,So),b(Kt,So,null),e(he,ou),e(he,Io),e(Io,nu),d(t,el,p),d(t,Er,p),e(Er,Qt),e(Qt,lu),d(t,tl,p),d(t,Xt,p),e(Xt,wr),e(wr,su),e(Xt,fu),d(t,il,p),d(t,Ve,p),e(Ve,hu),e(Ve,br),e(br,pu),e(Ve,du),al=!0},p(t,[p]){const ei={};p&2&&(ei.$$scope={dirty:p,ctx:t}),ge.$set(ei);const No={};p&2&&(No.$$scope={dirty:p,ctx:t}),ke.$set(No)},i(t){al||(x(T.$$.fragment,t),x(mt.$$.fragment,t),x(_t.$$.fragment,t),x(ge.$$.fragment,t),x(At.$$.fragment,t),x(Dt.$$.fragment,t),x(St.$$.fragment,t),x(Gt.$$.fragment,t),x(ke.$$.fragment,t),x(Rt.$$.fragment,t),x(Lt.$$.fragment,t),x(jt.$$.fragment,t),x(Ht.$$.fragment,t),x(Bt.$$.fragment,t),x(Ut.$$.fragment,t),x(Ot.$$.fragment,t),x(Jt.$$.fragment,t),x(Kt.$$.fragment,t),al=!0)},o(t){P(T.$$.fragment,t),P(mt.$$.fragment,t),P(_t.$$.fragment,t),P(ge.$$.fragment,t),P(At.$$.fragment,t),P(Dt.$$.fragment,t),P(St.$$.fragment,t),P(Gt.$$.fragment,t),P(ke.$$.fragment,t),P(Rt.$$.fragment,t),P(Lt.$$.fragment,t),P(jt.$$.fragment,t),P(Ht.$$.fragment,t),P(Bt.$$.fragment,t),P(Ut.$$.fragment,t),P(Ot.$$.fragment,t),P(Jt.$$.fragment,t),P(Kt.$$.fragment,t),al=!1},d(t){i(g),t&&i(D),t&&i(_),y(T),t&&i(qo),t&&i(ii),t&&i(Co),t&&i(ai),t&&i(Ro),t&&i(ri),t&&i(Mo),t&&i(k),t&&i(Lo),t&&i(oi),t&&i(jo),t&&i(ni),t&&i(zo),t&&i(li),t&&i(Ho),t&&i(u),t&&i(Zo),t&&i(xi),t&&i(Bo),t&&i(de),t&&i(Fo),t&&i(J),y(mt),t&&i(Wo),t&&i(ka),t&&i(Uo),t&&i(Y),t&&i(Oo),t&&i(ce),t&&i(Vo),t&&i(K),y(_t),t&&i(Jo),t&&i(Na),t&&i(Yo),t&&i(wt),t&&i(Ko),t&&i(qa),t&&i(Qo),t&&i(Ca),t&&i(Xo),t&&i(ve),t&&i(en),y(ge,t),t&&i(tn),t&&i(_e),t&&i(an),t&&i(Ee),t&&i(rn),t&&i(Q),y(At),t&&i(on),t&&i(La),t&&i(nn),t&&i(Tt),t&&i(ln),t&&i(za),t&&i(sn),t&&i(be),t&&i(fn),t&&i(xe),t&&i(hn),t&&i(X),y(Dt),t&&i(pn),t&&i(Za),t&&i(dn),t&&i(Ba),t&&i(un),t&&i(Fa),t&&i(cn),t&&i(Wa),t&&i(mn),t&&i(ye),t&&i(vn),t&&i(ee),y(St),t&&i(gn),t&&i(Oa),t&&i(_n),t&&i(Nt),t&&i(En),t&&i(Ja),t&&i(wn),t&&i($e),t&&i(bn),t&&i(te),y(Gt),t&&i(xn),t&&i(Ka),t&&i(Pn),t&&i(Ct),t&&i(yn),t&&i(Xa),t&&i(An),t&&i(De),t&&i($n),y(ke,t),t&&i(Tn),t&&i(ie),y(Rt),t&&i(Dn),t&&i(tr),t&&i(kn),t&&i(Ie),t&&i(Sn),t&&i(Ne),t&&i(In),t&&i(ae),y(Lt),t&&i(Nn),t&&i(rr),t&&i(Gn),t&&i(re),y(jt),t&&i(qn),t&&i(zt),t&&i(Cn),t&&i(Ce),t&&i(Rn),t&&i(oe),y(Ht),t&&i(Mn),t&&i(Zt),t&&i(Ln),t&&i(Me),t&&i(jn),t&&i(ne),y(Bt),t&&i(zn),t&&i(fr),t&&i(Hn),t&&i(Wt),t&&i(Zn),t&&i(je),t&&i(Bn),t&&i(le),y(Ut),t&&i(Fn),t&&i(dr),t&&i(Wn),t&&i(He),t&&i(Un),t&&i(se),y(Ot),t&&i(On),t&&i(Vt),t&&i(Vn),t&&i(Be),t&&i(Jn),t&&i(fe),y(Jt),t&&i(Yn),t&&i(vr),t&&i(Kn),t&&i(We),t&&i(Qn),t&&i(Ue),t&&i(Xn),t&&i(he),y(Kt),t&&i(el),t&&i(Er),t&&i(tl),t&&i(Xt),t&&i(il),t&&i(Ve)}}}const A2={local:"controlled-generation",sections:[{local:"instruct-pix2pix",title:"Instruct Pix2Pix"},{local:"pix2pix-zero",title:"Pix2Pix Zero"},{local:"attend-and-excite",title:"Attend and Excite"},{local:"semantic-guidance-sega",title:"Semantic Guidance (SEGA)"},{local:"selfattention-guidance-sag",title:"Self-attention Guidance (SAG)"},{local:"depth2image",title:"Depth2Image"},{local:"multidiffusion-panorama",title:"MultiDiffusion Panorama"},{local:"finetuning-your-own-models",title:"Fine-tuning your own models"},{local:"dreambooth",title:"DreamBooth"},{local:"textual-inversion",title:"Textual Inversion"},{local:"controlnet",title:"ControlNet"},{local:"prompt-weighting",title:"Prompt Weighting"},{local:"custom-diffusion",title:"Custom Diffusion "},{local:"model-editing",title:"Model Editing "},{local:"diffedit",title:"DiffEdit "}],title:"Controlled generation"};function $2(ti){return b2(()=>{new URLSearchParams(window.location.search).get("fw")}),[]}class S2 extends g2{constructor(g){super();_2(this,g,$2,y2,E2,{})}}export{S2 as default,A2 as metadata};