diffusers/v0.19.2/en/_app/pages/using-diffusers/controlling_generation.mdx-hf-doc-builder.js

import{S as _g,i as Eg,s as wg,e as i,k as f,w as E,t as l,M as bg,c as r,d as a,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 xg}from"../../chunks/vendor-hf-doc-builder.js";import{T as gg}from"../../chunks/Tip-hf-doc-builder.js";import{I as $}from"../../chunks/IconCopyLink-hf-doc-builder.js";function Pg(fa){let g,D,_,A,J;return{c(){g=i("p"),D=l(`Pix2Pix Zero is the first model that allows \u201Czero-shot\u201D image editing. This means that the model
can edit an image in less than a minute on a consumer GPU as shown `),_=i("a"),A=l("here"),J=l("."),this.h()},l(T){g=r(T,"P",{});var O=o(g);D=s(O,`Pix2Pix Zero is the first model that allows \u201Czero-shot\u201D image editing. This means that the model
can edit an image in less than a minute on a consumer GPU as shown `),_=r(O,"A",{href:!0});var ce=o(_);A=s(ce,"here"),ce.forEach(a),J=s(O,"."),O.forEach(a),this.h()},h(){n(_,"href","../api/pipelines/stable_diffusion/pix2pix_zero#usage-example")},m(T,O){d(T,g,O),e(g,D),e(g,_),e(_,A),e(g,J)},d(T){T&&a(g)}}}function yg(fa){let g,D;return{c(){g=i("p"),D=l(`An important distinction between methods like InstructPix2Pix and Pix2Pix Zero is that the former
involves fine-tuning the pre-trained weights while the latter does not. This means that you can
apply Pix2Pix Zero to any of the available Stable Diffusion models.`)},l(_){g=r(_,"P",{});var A=o(g);D=s(A,`An important distinction between methods like InstructPix2Pix and Pix2Pix Zero is that the former
involves fine-tuning the pre-trained weights while the latter does not. This means that you can
apply Pix2Pix Zero to any of the available Stable Diffusion models.`),A.forEach(a)},m(_,A){d(_,g,A),e(g,D)},d(_){_&&a(g)}}}function Ag(fa){let g,D,_,A,J,T,O,ce,cs,en,ha,ms,tn,pa,vs,an,da,gs,rn,k,_s,jr,Es,ws,_t,bs,xs,Et,Ps,ys,on,ua,As,nn,ca,$s,ln,ma,Ts,sn,c,zr,va,Ds,ks,Hr,ga,Ss,Is,Zr,_a,Ns,Gs,Br,Ea,Cs,qs,Fr,wa,Rs,Ms,Wr,ba,Ls,js,Ur,xa,zs,Hs,Or,Pa,Zs,Bs,Vr,ya,Fs,Ws,Jr,Aa,Us,Os,Yr,$a,Vs,Js,Kr,Ta,Ys,Ks,Qr,Da,Qs,Xs,Xr,ka,ef,fn,Sa,tf,hn,me,eo,I,Ia,to,af,rf,Na,ao,of,nf,Ga,wt,lf,sf,ff,hf,Ca,io,pf,df,u,N,qa,Ra,uf,cf,Ma,mf,vf,La,gf,_f,S,Ef,wf,bf,xf,Pf,yf,Af,$f,G,ja,za,Tf,Df,Ha,kf,Sf,Za,If,Nf,ro,Gf,C,Ba,Fa,Cf,qf,Wa,Rf,Mf,Ua,Lf,jf,oo,zf,q,Oa,Va,Hf,Zf,Ja,Bf,Ff,Ya,Wf,Uf,no,Of,R,Ka,Qa,Vf,Jf,Xa,Yf,Kf,ei,Qf,Xf,lo,eh,M,ti,ai,th,ah,ii,ih,rh,ri,oh,nh,so,lh,L,oi,ni,sh,fh,li,hh,ph,si,dh,uh,fo,ch,j,fi,hi,mh,vh,pi,gh,_h,di,Eh,wh,ho,bh,z,ui,ci,xh,Ph,mi,yh,Ah,vi,$h,Th,po,Dh,H,gi,_i,kh,Sh,Ei,Ih,Nh,wi,Gh,Ch,V,qh,Rh,Mh,Lh,jh,zh,Z,bi,xi,Hh,Zh,Pi,Bh,Fh,yi,Wh,Uh,uo,Oh,B,Ai,$i,Vh,Jh,Ti,Yh,Kh,Di,Qh,Xh,co,ep,F,ki,Si,tp,ap,Ii,ip,rp,Ni,op,np,mo,lp,W,Gi,Ci,sp,fp,qi,hp,pp,Ri,dp,up,vo,cp,U,Mi,Li,mp,vp,ji,gp,_p,zi,Ep,wp,go,pn,Y,ve,_o,bt,bp,Eo,xp,dn,Hi,xt,Pp,un,K,Zi,yp,Ap,Pt,$p,Tp,cn,ge,Dp,Bi,kp,Sp,mn,Q,_e,wo,yt,Ip,bo,Np,vn,Fi,At,Gp,gn,$t,Wi,Cp,qp,_n,Ui,Rp,En,Oi,Mp,wn,Ee,Tt,Lp,Dt,jp,zp,Hp,kt,Zp,St,Bp,Fp,bn,we,xn,be,Wp,Vi,Up,Op,Pn,xe,Vp,Ji,Jp,Yp,yn,X,Pe,xo,It,Kp,Po,Qp,An,Yi,Nt,Xp,$n,Gt,Ki,ed,td,Tn,Qi,ad,Dn,ye,id,yo,rd,od,kn,Ae,nd,Xi,ld,sd,Sn,ee,$e,Ao,Ct,fd,$o,hd,In,er,qt,pd,Nn,tr,dd,Gn,ar,ud,Cn,ir,cd,qn,Te,md,rr,vd,gd,Rn,te,De,To,Rt,_d,Do,Ed,Mn,or,Mt,wd,Ln,Lt,nr,bd,xd,jn,lr,Pd,zn,ke,yd,sr,Ad,$d,Hn,ae,Se,ko,jt,Td,So,Dd,Zn,fr,zt,kd,Bn,Ht,hr,Sd,Id,Fn,pr,Nd,Wn,Ie,Gd,dr,Cd,qd,Un,Ne,On,ie,Ge,Io,Zt,Rd,No,Md,Vn,ur,Bt,Ld,Jn,Ce,jd,cr,zd,Hd,Yn,qe,Zd,mr,Bd,Fd,Kn,re,Re,Go,Ft,Wd,Co,Ud,Qn,vr,Od,Xn,oe,Me,qo,Wt,Vd,Ro,Jd,el,Ut,gr,Yd,Kd,tl,Le,Qd,_r,Xd,eu,al,ne,je,Mo,Ot,tu,Lo,au,il,Vt,Er,iu,ru,rl,ze,ou,wr,nu,lu,ol,le,He,jo,Jt,su,zo,fu,nl,br,Yt,hu,ll,se,xr,pu,du,Pr,uu,cu,sl,Ze,mu,yr,vu,gu,fl,fe,Be,Ho,Kt,_u,Zo,Eu,hl,Ar,wu,pl,Fe,bu,$r,xu,Pu,dl,he,We,Bo,Qt,yu,Fo,Au,ul,Xt,Tr,$u,Tu,cl,Ue,Du,Dr,ku,Su,ml,pe,Oe,Wo,ea,Iu,Uo,Nu,vl,kr,ta,Gu,gl,Ve,Cu,Sr,qu,Ru,_l,Je,Mu,Ir,Lu,ju,El,de,Ye,Oo,aa,zu,Vo,Hu,wl,Nr,ia,Zu,bl,ra,Gr,Bu,Fu,xl,Ke,Wu,Cr,Uu,Ou,Pl,ue,Qe,Jo,oa,Vu,Yo,Ju,yl,qr,na,Yu,Al,la,Rr,Ku,Qu,$l,Xe,Xu,Mr,ec,tc,Tl;return T=new $({}),bt=new $({}),yt=new $({}),we=new gg({props:{$$slots:{default:[Pg]},$$scope:{ctx:fa}}}),It=new $({}),Ct=new $({}),Rt=new $({}),jt=new $({}),Ne=new gg({props:{$$slots:{default:[yg]},$$scope:{ctx:fa}}}),Zt=new $({}),Ft=new $({}),Wt=new $({}),Ot=new $({}),Jt=new $({}),Kt=new $({}),Qt=new $({}),ea=new $({}),aa=new $({}),oa=new $({}),{c(){g=i("meta"),D=f(),_=i("h1"),A=i("a"),J=i("span"),E(T.$$.fragment),O=f(),ce=i("span"),cs=l("Controlled generation"),en=f(),ha=i("p"),ms=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."),tn=f(),pa=i("p"),vs=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."),an=f(),da=i("p"),gs=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."),rn=f(),k=i("p"),_s=l("We will document some of the techniques "),jr=i("code"),Es=l("diffusers"),ws=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 "),_t=i("a"),bs=l("forum"),xs=l(" or a "),Et=i("a"),Ps=l("GitHub issue"),ys=l("."),on=f(),ua=i("p"),As=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."),nn=f(),ca=i("p"),$s=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."),ln=f(),ma=i("p"),Ts=l("Unless otherwise mentioned, these are techniques that work with existing models and don\u2019t require their own weights."),sn=f(),c=i("ol"),zr=i("li"),va=i("a"),Ds=l("Instruct Pix2Pix"),ks=f(),Hr=i("li"),ga=i("a"),Ss=l("Pix2Pix Zero"),Is=f(),Zr=i("li"),_a=i("a"),Ns=l("Attend and Excite"),Gs=f(),Br=i("li"),Ea=i("a"),Cs=l("Semantic Guidance"),qs=f(),Fr=i("li"),wa=i("a"),Rs=l("Self-attention Guidance"),Ms=f(),Wr=i("li"),ba=i("a"),Ls=l("Depth2Image"),js=f(),Ur=i("li"),xa=i("a"),zs=l("MultiDiffusion Panorama"),Hs=f(),Or=i("li"),Pa=i("a"),Zs=l("DreamBooth"),Bs=f(),Vr=i("li"),ya=i("a"),Fs=l("Textual Inversion"),Ws=f(),Jr=i("li"),Aa=i("a"),Us=l("ControlNet"),Os=f(),Yr=i("li"),$a=i("a"),Vs=l("Prompt Weighting"),Js=f(),Kr=i("li"),Ta=i("a"),Ys=l("Custom Diffusion"),Ks=f(),Qr=i("li"),Da=i("a"),Qs=l("Model Editing"),Xs=f(),Xr=i("li"),ka=i("a"),ef=l("DiffEdit"),fn=f(),Sa=i("p"),tf=l("For convenience, we provide a table to denote which methods are inference-only and which require fine-tuning/training."),hn=f(),me=i("table"),eo=i("thead"),I=i("tr"),Ia=i("th"),to=i("strong"),af=l("Method"),rf=f(),Na=i("th"),ao=i("strong"),of=l("Inference only"),nf=f(),Ga=i("th"),wt=i("strong"),lf=l("Requires training /"),sf=i("br"),ff=l(" fine-tuning"),hf=f(),Ca=i("th"),io=i("strong"),pf=l("Comments"),df=f(),u=i("tbody"),N=i("tr"),qa=i("td"),Ra=i("a"),uf=l("Instruct Pix2Pix"),cf=f(),Ma=i("td"),mf=l("\u2705"),vf=f(),La=i("td"),gf=l("\u274C"),_f=f(),S=i("td"),Ef=l("Can additionally be"),wf=i("br"),bf=l("fine-tuned for better "),xf=i("br"),Pf=l("performance on specific "),yf=i("br"),Af=l("edit instructions."),$f=f(),G=i("tr"),ja=i("td"),za=i("a"),Tf=l("Pix2Pix Zero"),Df=f(),Ha=i("td"),kf=l("\u2705"),Sf=f(),Za=i("td"),If=l("\u274C"),Nf=f(),ro=i("td"),Gf=f(),C=i("tr"),Ba=i("td"),Fa=i("a"),Cf=l("Attend and Excite"),qf=f(),Wa=i("td"),Rf=l("\u2705"),Mf=f(),Ua=i("td"),Lf=l("\u274C"),jf=f(),oo=i("td"),zf=f(),q=i("tr"),Oa=i("td"),Va=i("a"),Hf=l("Semantic Guidance"),Zf=f(),Ja=i("td"),Bf=l("\u2705"),Ff=f(),Ya=i("td"),Wf=l("\u274C"),Uf=f(),no=i("td"),Of=f(),R=i("tr"),Ka=i("td"),Qa=i("a"),Vf=l("Self-attention Guidance"),Jf=f(),Xa=i("td"),Yf=l("\u2705"),Kf=f(),ei=i("td"),Qf=l("\u274C"),Xf=f(),lo=i("td"),eh=f(),M=i("tr"),ti=i("td"),ai=i("a"),th=l("Depth2Image"),ah=f(),ii=i("td"),ih=l("\u2705"),rh=f(),ri=i("td"),oh=l("\u274C"),nh=f(),so=i("td"),lh=f(),L=i("tr"),oi=i("td"),ni=i("a"),sh=l("MultiDiffusion Panorama"),fh=f(),li=i("td"),hh=l("\u2705"),ph=f(),si=i("td"),dh=l("\u274C"),uh=f(),fo=i("td"),ch=f(),j=i("tr"),fi=i("td"),hi=i("a"),mh=l("DreamBooth"),vh=f(),pi=i("td"),gh=l("\u274C"),_h=f(),di=i("td"),Eh=l("\u2705"),wh=f(),ho=i("td"),bh=f(),z=i("tr"),ui=i("td"),ci=i("a"),xh=l("Textual Inversion"),Ph=f(),mi=i("td"),yh=l("\u274C"),Ah=f(),vi=i("td"),$h=l("\u2705"),Th=f(),po=i("td"),Dh=f(),H=i("tr"),gi=i("td"),_i=i("a"),kh=l("ControlNet"),Sh=f(),Ei=i("td"),Ih=l("\u2705"),Nh=f(),wi=i("td"),Gh=l("\u274C"),Ch=f(),V=i("td"),qh=l("A ControlNet can be "),Rh=i("br"),Mh=l("trained/fine-tuned on"),Lh=i("br"),jh=l("a custom conditioning."),zh=f(),Z=i("tr"),bi=i("td"),xi=i("a"),Hh=l("Prompt Weighting"),Zh=f(),Pi=i("td"),Bh=l("\u2705"),Fh=f(),yi=i("td"),Wh=l("\u274C"),Uh=f(),uo=i("td"),Oh=f(),B=i("tr"),Ai=i("td"),$i=i("a"),Vh=l("Custom Diffusion"),Jh=f(),Ti=i("td"),Yh=l("\u274C"),Kh=f(),Di=i("td"),Qh=l("\u2705"),Xh=f(),co=i("td"),ep=f(),F=i("tr"),ki=i("td"),Si=i("a"),tp=l("Model Editing"),ap=f(),Ii=i("td"),ip=l("\u2705"),rp=f(),Ni=i("td"),op=l("\u274C"),np=f(),mo=i("td"),lp=f(),W=i("tr"),Gi=i("td"),Ci=i("a"),sp=l("DiffEdit"),fp=f(),qi=i("td"),hp=l("\u2705"),pp=f(),Ri=i("td"),dp=l("\u274C"),up=f(),vo=i("td"),cp=f(),U=i("tr"),Mi=i("td"),Li=i("a"),mp=l("T2I-Adapter"),vp=f(),ji=i("td"),gp=l("\u2705"),_p=f(),zi=i("td"),Ep=l("\u274C"),wp=f(),go=i("td"),pn=f(),Y=i("h2"),ve=i("a"),_o=i("span"),E(bt.$$.fragment),bp=f(),Eo=i("span"),xp=l("Instruct Pix2Pix"),dn=f(),Hi=i("p"),xt=i("a"),Pp=l("Paper"),un=f(),K=i("p"),Zi=i("a"),yp=l("Instruct Pix2Pix"),Ap=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 `),Pt=i("a"),$p=l("InstructGPT"),Tp=l("-like prompts."),cn=f(),ge=i("p"),Dp=l("See "),Bi=i("a"),kp=l("here"),Sp=l(" for more information on how to use it."),mn=f(),Q=i("h2"),_e=i("a"),wo=i("span"),E(yt.$$.fragment),Ip=f(),bo=i("span"),Np=l("Pix2Pix Zero"),vn=f(),Fi=i("p"),At=i("a"),Gp=l("Paper"),gn=f(),$t=i("p"),Wi=i("a"),Cp=l("Pix2Pix Zero"),qp=l(" allows modifying an image so that one concept or subject is translated to another one while preserving general image semantics."),_n=f(),Ui=i("p"),Rp=l("The denoising process is guided from one conceptual embedding towards another conceptual embedding. The intermediate latents are optimized during the denoising process to push the attention maps towards reference attention maps. The reference attention maps are from the denoising process of the input image and are used to encourage semantic preservation."),En=f(),Oi=i("p"),Mp=l("Pix2Pix Zero can be used both to edit synthetic images as well as real images."),wn=f(),Ee=i("ul"),Tt=i("li"),Lp=l(`To edit synthetic images, one first generates an image given a caption.
Next, we generate image captions for the concept that shall be edited and for the new target concept. We can use a model like `),Dt=i("a"),jp=l("Flan-T5"),zp=l(" for this purpose. Then, \u201Cmean\u201D prompt embeddings for both the source and target concepts are created via the text encoder. Finally, the pix2pix-zero algorithm is used to edit the synthetic image."),Hp=f(),kt=i("li"),Zp=l("To edit a real image, one first generates an image caption using a model like "),St=i("a"),Bp=l("BLIP"),Fp=l(". Then one applies ddim inversion on the prompt and image to generate \u201Cinverse\u201D latents. Similar to before, \u201Cmean\u201D prompt embeddings for both source and target concepts are created and finally the pix2pix-zero algorithm in combination with the \u201Cinverse\u201D latents is used to edit the image."),bn=f(),E(we.$$.fragment),xn=f(),be=i("p"),Wp=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 `),Vi=i("a"),Up=l("StableDiffusionPipeline"),Op=l("."),Pn=f(),xe=i("p"),Vp=l("See "),Ji=i("a"),Jp=l("here"),Yp=l(" for more information on how to use it."),yn=f(),X=i("h2"),Pe=i("a"),xo=i("span"),E(It.$$.fragment),Kp=f(),Po=i("span"),Qp=l("Attend and Excite"),An=f(),Yi=i("p"),Nt=i("a"),Xp=l("Paper"),$n=f(),Gt=i("p"),Ki=i("a"),ed=l("Attend and Excite"),td=l(" allows subjects in the prompt to be faithfully represented in the final image."),Tn=f(),Qi=i("p"),ad=l("A set of token indices are given as input, corresponding to the subjects in the prompt that need to be present in the image. During denoising, each token index is guaranteed to have a minimum attention threshold for at least one patch of the image. The intermediate latents are iteratively optimized during the denoising process to strengthen the attention of the most neglected subject token until the attention threshold is passed for all subject tokens."),Dn=f(),ye=i("p"),id=l("Like Pix2Pix Zero, Attend and Excite also involves a mini optimization loop (leaving the pre-trained weights untouched) in its pipeline and can require more memory than the usual "),yo=i("code"),rd=l("StableDiffusionPipeline"),od=l("."),kn=f(),Ae=i("p"),nd=l("See "),Xi=i("a"),ld=l("here"),sd=l(" for more information on how to use it."),Sn=f(),ee=i("h2"),$e=i("a"),Ao=i("span"),E(Ct.$$.fragment),fd=f(),$o=i("span"),hd=l("Semantic Guidance (SEGA)"),In=f(),er=i("p"),qt=i("a"),pd=l("Paper"),Nn=f(),tr=i("p"),dd=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."),Gn=f(),ar=i("p"),ud=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(),ir=i("p"),cd=l("Unlike Pix2Pix Zero or Attend and Excite, SEGA directly interacts with the diffusion process instead of performing any explicit gradient-based optimization."),qn=f(),Te=i("p"),md=l("See "),rr=i("a"),vd=l("here"),gd=l(" for more information on how to use it."),Rn=f(),te=i("h2"),De=i("a"),To=i("span"),E(Rt.$$.fragment),_d=f(),Do=i("span"),Ed=l("Self-attention Guidance (SAG)"),Mn=f(),or=i("p"),Mt=i("a"),wd=l("Paper"),Ln=f(),Lt=i("p"),nr=i("a"),bd=l("Self-attention Guidance"),xd=l(" improves the general quality of images."),jn=f(),lr=i("p"),Pd=l("SAG provides guidance from predictions not conditioned on high-frequency details to fully conditioned images. The high frequency details are extracted out of the UNet self-attention maps."),zn=f(),ke=i("p"),yd=l("See "),sr=i("a"),Ad=l("here"),$d=l(" for more information on how to use it."),Hn=f(),ae=i("h2"),Se=i("a"),ko=i("span"),E(jt.$$.fragment),Td=f(),So=i("span"),Dd=l("Depth2Image"),Zn=f(),fr=i("p"),zt=i("a"),kd=l("Project"),Bn=f(),Ht=i("p"),hr=i("a"),Sd=l("Depth2Image"),Id=l(" is fine-tuned from Stable Diffusion to better preserve semantics for text guided image variation."),Fn=f(),pr=i("p"),Nd=l("It conditions on a monocular depth estimate of the original image."),Wn=f(),Ie=i("p"),Gd=l("See "),dr=i("a"),Cd=l("here"),qd=l(" for more information on how to use it."),Un=f(),E(Ne.$$.fragment),On=f(),ie=i("h2"),Ge=i("a"),Io=i("span"),E(Zt.$$.fragment),Rd=f(),No=i("span"),Md=l("MultiDiffusion Panorama"),Vn=f(),ur=i("p"),Bt=i("a"),Ld=l("Paper"),Jn=f(),Ce=i("p"),jd=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.
`),cr=i("a"),zd=l("MultiDiffusion Panorama"),Hd=l(" allows to generate high-quality images at arbitrary aspect ratios (e.g., panoramas)."),Yn=f(),qe=i("p"),Zd=l("See "),mr=i("a"),Bd=l("here"),Fd=l(" for more information on how to use it to generate panoramic images."),Kn=f(),re=i("h2"),Re=i("a"),Go=i("span"),E(Ft.$$.fragment),Wd=f(),Co=i("span"),Ud=l("Fine-tuning your own models"),Qn=f(),vr=i("p"),Od=l("In addition to pre-trained models, Diffusers has training scripts for fine-tuning models on user-provided data."),Xn=f(),oe=i("h2"),Me=i("a"),qo=i("span"),E(Wt.$$.fragment),Vd=f(),Ro=i("span"),Jd=l("DreamBooth"),el=f(),Ut=i("p"),gr=i("a"),Yd=l("DreamBooth"),Kd=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."),tl=f(),Le=i("p"),Qd=l("See "),_r=i("a"),Xd=l("here"),eu=l(" for more information on how to use it."),al=f(),ne=i("h2"),je=i("a"),Mo=i("span"),E(Ot.$$.fragment),tu=f(),Lo=i("span"),au=l("Textual Inversion"),il=f(),Vt=i("p"),Er=i("a"),iu=l("Textual Inversion"),ru=l(" fine-tunes a model to teach it about a new concept. I.e. a few pictures of a style of artwork can be used to generate images in that style."),rl=f(),ze=i("p"),ou=l("See "),wr=i("a"),nu=l("here"),lu=l(" for more information on how to use it."),ol=f(),le=i("h2"),He=i("a"),jo=i("span"),E(Jt.$$.fragment),su=f(),zo=i("span"),fu=l("ControlNet"),nl=f(),br=i("p"),Yt=i("a"),hu=l("Paper"),ll=f(),se=i("p"),xr=i("a"),pu=l("ControlNet"),du=l(` is an auxiliary network which adds an extra condition.
`),Pr=i("a"),uu=l("ControlNet"),cu=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.`),sl=f(),Ze=i("p"),mu=l("See "),yr=i("a"),vu=l("here"),gu=l(" for more information on how to use it."),fl=f(),fe=i("h2"),Be=i("a"),Ho=i("span"),E(Kt.$$.fragment),_u=f(),Zo=i("span"),Eu=l("Prompt Weighting"),hl=f(),Ar=i("p"),wu=l(`Prompt weighting is a simple technique that puts more attention weight on certain parts of the text
input.`),pl=f(),Fe=i("p"),bu=l("For a more in-detail explanation and examples, see "),$r=i("a"),xu=l("here"),Pu=l("."),dl=f(),he=i("h2"),We=i("a"),Bo=i("span"),E(Qt.$$.fragment),yu=f(),Fo=i("span"),Au=l("Custom Diffusion"),ul=f(),Xt=i("p"),Tr=i("a"),$u=l("Custom Diffusion"),Tu=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.`),cl=f(),Ue=i("p"),Du=l("For more details, check out our "),Dr=i("a"),ku=l("official doc"),Su=l("."),ml=f(),pe=i("h2"),Oe=i("a"),Wo=i("span"),E(ea.$$.fragment),Iu=f(),Uo=i("span"),Nu=l("Model Editing"),vl=f(),kr=i("p"),ta=i("a"),Gu=l("Paper"),gl=f(),Ve=i("p"),Cu=l("The "),Sr=i("a"),qu=l("text-to-image model editing pipeline"),Ru=l(` helps you mitigate some of the incorrect implicit assumptions a pre-trained text-to-image
diffusion model might make about the subjects present in the input prompt. For example, if you prompt Stable Diffusion to generate images for \u201CA pack of roses\u201D, the roses in the generated images
are more likely to be red. This pipeline helps you change that assumption.`),_l=f(),Je=i("p"),Mu=l("To know more details, check out the "),Ir=i("a"),Lu=l("official doc"),ju=l("."),El=f(),de=i("h2"),Ye=i("a"),Oo=i("span"),E(aa.$$.fragment),zu=f(),Vo=i("span"),Hu=l("DiffEdit"),wl=f(),Nr=i("p"),ia=i("a"),Zu=l("Paper"),bl=f(),ra=i("p"),Gr=i("a"),Bu=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.`),xl=f(),Ke=i("p"),Wu=l("To know more details, check out the "),Cr=i("a"),Uu=l("official doc"),Ou=l("."),Pl=f(),ue=i("h2"),Qe=i("a"),Jo=i("span"),E(oa.$$.fragment),Vu=f(),Yo=i("span"),Ju=l("T2I-Adapter"),yl=f(),qr=i("p"),na=i("a"),Yu=l("Paper"),Al=f(),la=i("p"),Rr=i("a"),Ku=l("T2I-Adapter"),Qu=l(` is an auxiliary network which adds an extra condition.
There are 8 canonical pre-trained adapters trained on different conditionings such as edge detection, sketch,
depth maps, and semantic segmentations.`),$l=f(),Xe=i("p"),Xu=l("See "),Mr=i("a"),ec=l("here"),tc=l(" for more information on how to use it."),this.h()},l(t){const p=bg('[data-svelte="svelte-1phssyn"]',document.head);g=r(p,"META",{name:!0,content:!0}),p.forEach(a),D=h(t),_=r(t,"H1",{class:!0});var sa=o(_);A=r(sa,"A",{id:!0,class:!0,href:!0});var Ko=o(A);J=r(Ko,"SPAN",{});var pc=o(J);w(T.$$.fragment,pc),pc.forEach(a),Ko.forEach(a),O=h(sa),ce=r(sa,"SPAN",{});var dc=o(ce);cs=s(dc,"Controlled generation"),dc.forEach(a),sa.forEach(a),en=h(t),ha=r(t,"P",{});var uc=o(ha);ms=s(uc,"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."),uc.forEach(a),tn=h(t),pa=r(t,"P",{});var cc=o(pa);vs=s(cc,"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."),cc.forEach(a),an=h(t),da=r(t,"P",{});var mc=o(da);gs=s(mc,"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."),mc.forEach(a),rn=h(t),k=r(t,"P",{});var et=o(k);_s=s(et,"We will document some of the techniques "),jr=r(et,"CODE",{});var vc=o(jr);Es=s(vc,"diffusers"),vc.forEach(a),ws=s(et," 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 "),_t=r(et,"A",{href:!0,rel:!0});var gc=o(_t);bs=s(gc,"forum"),gc.forEach(a),xs=s(et," or a "),Et=r(et,"A",{href:!0,rel:!0});var _c=o(Et);Ps=s(_c,"GitHub issue"),_c.forEach(a),ys=s(et,"."),et.forEach(a),on=h(t),ua=r(t,"P",{});var Ec=o(ua);As=s(Ec,"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."),Ec.forEach(a),nn=h(t),ca=r(t,"P",{});var wc=o(ca);$s=s(wc,"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."),wc.forEach(a),ln=h(t),ma=r(t,"P",{});var bc=o(ma);Ts=s(bc,"Unless otherwise mentioned, these are techniques that work with existing models and don\u2019t require their own weights."),bc.forEach(a),sn=h(t),c=r(t,"OL",{});var v=o(c);zr=r(v,"LI",{});var xc=o(zr);va=r(xc,"A",{href:!0});var Pc=o(va);Ds=s(Pc,"Instruct Pix2Pix"),Pc.forEach(a),xc.forEach(a),ks=h(v),Hr=r(v,"LI",{});var yc=o(Hr);ga=r(yc,"A",{href:!0});var Ac=o(ga);Ss=s(Ac,"Pix2Pix Zero"),Ac.forEach(a),yc.forEach(a),Is=h(v),Zr=r(v,"LI",{});var $c=o(Zr);_a=r($c,"A",{href:!0});var Tc=o(_a);Ns=s(Tc,"Attend and Excite"),Tc.forEach(a),$c.forEach(a),Gs=h(v),Br=r(v,"LI",{});var Dc=o(Br);Ea=r(Dc,"A",{href:!0});var kc=o(Ea);Cs=s(kc,"Semantic Guidance"),kc.forEach(a),Dc.forEach(a),qs=h(v),Fr=r(v,"LI",{});var Sc=o(Fr);wa=r(Sc,"A",{href:!0});var Ic=o(wa);Rs=s(Ic,"Self-attention Guidance"),Ic.forEach(a),Sc.forEach(a),Ms=h(v),Wr=r(v,"LI",{});var Nc=o(Wr);ba=r(Nc,"A",{href:!0});var Gc=o(ba);Ls=s(Gc,"Depth2Image"),Gc.forEach(a),Nc.forEach(a),js=h(v),Ur=r(v,"LI",{});var Cc=o(Ur);xa=r(Cc,"A",{href:!0});var qc=o(xa);zs=s(qc,"MultiDiffusion Panorama"),qc.forEach(a),Cc.forEach(a),Hs=h(v),Or=r(v,"LI",{});var Rc=o(Or);Pa=r(Rc,"A",{href:!0});var Mc=o(Pa);Zs=s(Mc,"DreamBooth"),Mc.forEach(a),Rc.forEach(a),Bs=h(v),Vr=r(v,"LI",{});var Lc=o(Vr);ya=r(Lc,"A",{href:!0});var jc=o(ya);Fs=s(jc,"Textual Inversion"),jc.forEach(a),Lc.forEach(a),Ws=h(v),Jr=r(v,"LI",{});var zc=o(Jr);Aa=r(zc,"A",{href:!0});var Hc=o(Aa);Us=s(Hc,"ControlNet"),Hc.forEach(a),zc.forEach(a),Os=h(v),Yr=r(v,"LI",{});var Zc=o(Yr);$a=r(Zc,"A",{href:!0});var Bc=o($a);Vs=s(Bc,"Prompt Weighting"),Bc.forEach(a),Zc.forEach(a),Js=h(v),Kr=r(v,"LI",{});var Fc=o(Kr);Ta=r(Fc,"A",{href:!0});var Wc=o(Ta);Ys=s(Wc,"Custom Diffusion"),Wc.forEach(a),Fc.forEach(a),Ks=h(v),Qr=r(v,"LI",{});var Uc=o(Qr);Da=r(Uc,"A",{href:!0});var Oc=o(Da);Qs=s(Oc,"Model Editing"),Oc.forEach(a),Uc.forEach(a),Xs=h(v),Xr=r(v,"LI",{});var Vc=o(Xr);ka=r(Vc,"A",{href:!0});var Jc=o(ka);ef=s(Jc,"DiffEdit"),Jc.forEach(a),Vc.forEach(a),v.forEach(a),fn=h(t),Sa=r(t,"P",{});var Yc=o(Sa);tf=s(Yc,"For convenience, we provide a table to denote which methods are inference-only and which require fine-tuning/training."),Yc.forEach(a),hn=h(t),me=r(t,"TABLE",{});var Dl=o(me);eo=r(Dl,"THEAD",{});var Kc=o(eo);I=r(Kc,"TR",{});var tt=o(I);Ia=r(tt,"TH",{align:!0});var Qc=o(Ia);to=r(Qc,"STRONG",{});var Xc=o(to);af=s(Xc,"Method"),Xc.forEach(a),Qc.forEach(a),rf=h(tt),Na=r(tt,"TH",{align:!0});var em=o(Na);ao=r(em,"STRONG",{});var tm=o(ao);of=s(tm,"Inference only"),tm.forEach(a),em.forEach(a),nf=h(tt),Ga=r(tt,"TH",{align:!0});var am=o(Ga);wt=r(am,"STRONG",{});var kl=o(wt);lf=s(kl,"Requires training /"),sf=r(kl,"BR",{}),ff=s(kl," fine-tuning"),kl.forEach(a),am.forEach(a),hf=h(tt),Ca=r(tt,"TH",{align:!0});var im=o(Ca);io=r(im,"STRONG",{});var rm=o(io);pf=s(rm,"Comments"),rm.forEach(a),im.forEach(a),tt.forEach(a),Kc.forEach(a),df=h(Dl),u=r(Dl,"TBODY",{});var m=o(u);N=r(m,"TR",{});var at=o(N);qa=r(at,"TD",{align:!0});var om=o(qa);Ra=r(om,"A",{href:!0});var nm=o(Ra);uf=s(nm,"Instruct Pix2Pix"),nm.forEach(a),om.forEach(a),cf=h(at),Ma=r(at,"TD",{align:!0});var lm=o(Ma);mf=s(lm,"\u2705"),lm.forEach(a),vf=h(at),La=r(at,"TD",{align:!0});var sm=o(La);gf=s(sm,"\u274C"),sm.forEach(a),_f=h(at),S=r(at,"TD",{align:!0});var it=o(S);Ef=s(it,"Can additionally be"),wf=r(it,"BR",{}),bf=s(it,"fine-tuned for better "),xf=r(it,"BR",{}),Pf=s(it,"performance on specific "),yf=r(it,"BR",{}),Af=s(it,"edit instructions."),it.forEach(a),at.forEach(a),$f=h(m),G=r(m,"TR",{});var rt=o(G);ja=r(rt,"TD",{align:!0});var fm=o(ja);za=r(fm,"A",{href:!0});var hm=o(za);Tf=s(hm,"Pix2Pix Zero"),hm.forEach(a),fm.forEach(a),Df=h(rt),Ha=r(rt,"TD",{align:!0});var pm=o(Ha);kf=s(pm,"\u2705"),pm.forEach(a),Sf=h(rt),Za=r(rt,"TD",{align:!0});var dm=o(Za);If=s(dm,"\u274C"),dm.forEach(a),Nf=h(rt),ro=r(rt,"TD",{align:!0}),o(ro).forEach(a),rt.forEach(a),Gf=h(m),C=r(m,"TR",{});var ot=o(C);Ba=r(ot,"TD",{align:!0});var um=o(Ba);Fa=r(um,"A",{href:!0});var cm=o(Fa);Cf=s(cm,"Attend and Excite"),cm.forEach(a),um.forEach(a),qf=h(ot),Wa=r(ot,"TD",{align:!0});var mm=o(Wa);Rf=s(mm,"\u2705"),mm.forEach(a),Mf=h(ot),Ua=r(ot,"TD",{align:!0});var vm=o(Ua);Lf=s(vm,"\u274C"),vm.forEach(a),jf=h(ot),oo=r(ot,"TD",{align:!0}),o(oo).forEach(a),ot.forEach(a),zf=h(m),q=r(m,"TR",{});var nt=o(q);Oa=r(nt,"TD",{align:!0});var gm=o(Oa);Va=r(gm,"A",{href:!0});var _m=o(Va);Hf=s(_m,"Semantic Guidance"),_m.forEach(a),gm.forEach(a),Zf=h(nt),Ja=r(nt,"TD",{align:!0});var Em=o(Ja);Bf=s(Em,"\u2705"),Em.forEach(a),Ff=h(nt),Ya=r(nt,"TD",{align:!0});var wm=o(Ya);Wf=s(wm,"\u274C"),wm.forEach(a),Uf=h(nt),no=r(nt,"TD",{align:!0}),o(no).forEach(a),nt.forEach(a),Of=h(m),R=r(m,"TR",{});var lt=o(R);Ka=r(lt,"TD",{align:!0});var bm=o(Ka);Qa=r(bm,"A",{href:!0});var xm=o(Qa);Vf=s(xm,"Self-attention Guidance"),xm.forEach(a),bm.forEach(a),Jf=h(lt),Xa=r(lt,"TD",{align:!0});var Pm=o(Xa);Yf=s(Pm,"\u2705"),Pm.forEach(a),Kf=h(lt),ei=r(lt,"TD",{align:!0});var ym=o(ei);Qf=s(ym,"\u274C"),ym.forEach(a),Xf=h(lt),lo=r(lt,"TD",{align:!0}),o(lo).forEach(a),lt.forEach(a),eh=h(m),M=r(m,"TR",{});var st=o(M);ti=r(st,"TD",{align:!0});var Am=o(ti);ai=r(Am,"A",{href:!0});var $m=o(ai);th=s($m,"Depth2Image"),$m.forEach(a),Am.forEach(a),ah=h(st),ii=r(st,"TD",{align:!0});var Tm=o(ii);ih=s(Tm,"\u2705"),Tm.forEach(a),rh=h(st),ri=r(st,"TD",{align:!0});var Dm=o(ri);oh=s(Dm,"\u274C"),Dm.forEach(a),nh=h(st),so=r(st,"TD",{align:!0}),o(so).forEach(a),st.forEach(a),lh=h(m),L=r(m,"TR",{});var ft=o(L);oi=r(ft,"TD",{align:!0});var km=o(oi);ni=r(km,"A",{href:!0});var Sm=o(ni);sh=s(Sm,"MultiDiffusion Panorama"),Sm.forEach(a),km.forEach(a),fh=h(ft),li=r(ft,"TD",{align:!0});var Im=o(li);hh=s(Im,"\u2705"),Im.forEach(a),ph=h(ft),si=r(ft,"TD",{align:!0});var Nm=o(si);dh=s(Nm,"\u274C"),Nm.forEach(a),uh=h(ft),fo=r(ft,"TD",{align:!0}),o(fo).forEach(a),ft.forEach(a),ch=h(m),j=r(m,"TR",{});var ht=o(j);fi=r(ht,"TD",{align:!0});var Gm=o(fi);hi=r(Gm,"A",{href:!0});var Cm=o(hi);mh=s(Cm,"DreamBooth"),Cm.forEach(a),Gm.forEach(a),vh=h(ht),pi=r(ht,"TD",{align:!0});var qm=o(pi);gh=s(qm,"\u274C"),qm.forEach(a),_h=h(ht),di=r(ht,"TD",{align:!0});var Rm=o(di);Eh=s(Rm,"\u2705"),Rm.forEach(a),wh=h(ht),ho=r(ht,"TD",{align:!0}),o(ho).forEach(a),ht.forEach(a),bh=h(m),z=r(m,"TR",{});var pt=o(z);ui=r(pt,"TD",{align:!0});var Mm=o(ui);ci=r(Mm,"A",{href:!0});var Lm=o(ci);xh=s(Lm,"Textual Inversion"),Lm.forEach(a),Mm.forEach(a),Ph=h(pt),mi=r(pt,"TD",{align:!0});var jm=o(mi);yh=s(jm,"\u274C"),jm.forEach(a),Ah=h(pt),vi=r(pt,"TD",{align:!0});var zm=o(vi);$h=s(zm,"\u2705"),zm.forEach(a),Th=h(pt),po=r(pt,"TD",{align:!0}),o(po).forEach(a),pt.forEach(a),Dh=h(m),H=r(m,"TR",{});var dt=o(H);gi=r(dt,"TD",{align:!0});var Hm=o(gi);_i=r(Hm,"A",{href:!0});var Zm=o(_i);kh=s(Zm,"ControlNet"),Zm.forEach(a),Hm.forEach(a),Sh=h(dt),Ei=r(dt,"TD",{align:!0});var Bm=o(Ei);Ih=s(Bm,"\u2705"),Bm.forEach(a),Nh=h(dt),wi=r(dt,"TD",{align:!0});var Fm=o(wi);Gh=s(Fm,"\u274C"),Fm.forEach(a),Ch=h(dt),V=r(dt,"TD",{align:!0});var Lr=o(V);qh=s(Lr,"A ControlNet can be "),Rh=r(Lr,"BR",{}),Mh=s(Lr,"trained/fine-tuned on"),Lh=r(Lr,"BR",{}),jh=s(Lr,"a custom conditioning."),Lr.forEach(a),dt.forEach(a),zh=h(m),Z=r(m,"TR",{});var ut=o(Z);bi=r(ut,"TD",{align:!0});var Wm=o(bi);xi=r(Wm,"A",{href:!0});var Um=o(xi);Hh=s(Um,"Prompt Weighting"),Um.forEach(a),Wm.forEach(a),Zh=h(ut),Pi=r(ut,"TD",{align:!0});var Om=o(Pi);Bh=s(Om,"\u2705"),Om.forEach(a),Fh=h(ut),yi=r(ut,"TD",{align:!0});var Vm=o(yi);Wh=s(Vm,"\u274C"),Vm.forEach(a),Uh=h(ut),uo=r(ut,"TD",{align:!0}),o(uo).forEach(a),ut.forEach(a),Oh=h(m),B=r(m,"TR",{});var ct=o(B);Ai=r(ct,"TD",{align:!0});var Jm=o(Ai);$i=r(Jm,"A",{href:!0});var Ym=o($i);Vh=s(Ym,"Custom Diffusion"),Ym.forEach(a),Jm.forEach(a),Jh=h(ct),Ti=r(ct,"TD",{align:!0});var Km=o(Ti);Yh=s(Km,"\u274C"),Km.forEach(a),Kh=h(ct),Di=r(ct,"TD",{align:!0});var Qm=o(Di);Qh=s(Qm,"\u2705"),Qm.forEach(a),Xh=h(ct),co=r(ct,"TD",{align:!0}),o(co).forEach(a),ct.forEach(a),ep=h(m),F=r(m,"TR",{});var mt=o(F);ki=r(mt,"TD",{align:!0});var Xm=o(ki);Si=r(Xm,"A",{href:!0});var ev=o(Si);tp=s(ev,"Model Editing"),ev.forEach(a),Xm.forEach(a),ap=h(mt),Ii=r(mt,"TD",{align:!0});var tv=o(Ii);ip=s(tv,"\u2705"),tv.forEach(a),rp=h(mt),Ni=r(mt,"TD",{align:!0});var av=o(Ni);op=s(av,"\u274C"),av.forEach(a),np=h(mt),mo=r(mt,"TD",{align:!0}),o(mo).forEach(a),mt.forEach(a),lp=h(m),W=r(m,"TR",{});var vt=o(W);Gi=r(vt,"TD",{align:!0});var iv=o(Gi);Ci=r(iv,"A",{href:!0});var rv=o(Ci);sp=s(rv,"DiffEdit"),rv.forEach(a),iv.forEach(a),fp=h(vt),qi=r(vt,"TD",{align:!0});var ov=o(qi);hp=s(ov,"\u2705"),ov.forEach(a),pp=h(vt),Ri=r(vt,"TD",{align:!0});var nv=o(Ri);dp=s(nv,"\u274C"),nv.forEach(a),up=h(vt),vo=r(vt,"TD",{align:!0}),o(vo).forEach(a),vt.forEach(a),cp=h(m),U=r(m,"TR",{});var gt=o(U);Mi=r(gt,"TD",{align:!0});var lv=o(Mi);Li=r(lv,"A",{href:!0});var sv=o(Li);mp=s(sv,"T2I-Adapter"),sv.forEach(a),lv.forEach(a),vp=h(gt),ji=r(gt,"TD",{align:!0});var fv=o(ji);gp=s(fv,"\u2705"),fv.forEach(a),_p=h(gt),zi=r(gt,"TD",{align:!0});var hv=o(zi);Ep=s(hv,"\u274C"),hv.forEach(a),wp=h(gt),go=r(gt,"TD",{align:!0}),o(go).forEach(a),gt.forEach(a),m.forEach(a),Dl.forEach(a),pn=h(t),Y=r(t,"H2",{class:!0});var Sl=o(Y);ve=r(Sl,"A",{id:!0,class:!0,href:!0});var pv=o(ve);_o=r(pv,"SPAN",{});var dv=o(_o);w(bt.$$.fragment,dv),dv.forEach(a),pv.forEach(a),bp=h(Sl),Eo=r(Sl,"SPAN",{});var uv=o(Eo);xp=s(uv,"Instruct Pix2Pix"),uv.forEach(a),Sl.forEach(a),dn=h(t),Hi=r(t,"P",{});var cv=o(Hi);xt=r(cv,"A",{href:!0,rel:!0});var mv=o(xt);Pp=s(mv,"Paper"),mv.forEach(a),cv.forEach(a),un=h(t),K=r(t,"P",{});var Qo=o(K);Zi=r(Qo,"A",{href:!0});var vv=o(Zi);yp=s(vv,"Instruct Pix2Pix"),vv.forEach(a),Ap=s(Qo,` is fine-tuned from stable diffusion to support editing input images. It takes as inputs an image and a prompt describing an edit, and it outputs the edited image.
Instruct Pix2Pix has been explicitly trained to work well with `),Pt=r(Qo,"A",{href:!0,rel:!0});var gv=o(Pt);$p=s(gv,"InstructGPT"),gv.forEach(a),Tp=s(Qo,"-like prompts."),Qo.forEach(a),cn=h(t),ge=r(t,"P",{});var Il=o(ge);Dp=s(Il,"See "),Bi=r(Il,"A",{href:!0});var _v=o(Bi);kp=s(_v,"here"),_v.forEach(a),Sp=s(Il," for more information on how to use it."),Il.forEach(a),mn=h(t),Q=r(t,"H2",{class:!0});var Nl=o(Q);_e=r(Nl,"A",{id:!0,class:!0,href:!0});var Ev=o(_e);wo=r(Ev,"SPAN",{});var wv=o(wo);w(yt.$$.fragment,wv),wv.forEach(a),Ev.forEach(a),Ip=h(Nl),bo=r(Nl,"SPAN",{});var bv=o(bo);Np=s(bv,"Pix2Pix Zero"),bv.forEach(a),Nl.forEach(a),vn=h(t),Fi=r(t,"P",{});var xv=o(Fi);At=r(xv,"A",{href:!0,rel:!0});var Pv=o(At);Gp=s(Pv,"Paper"),Pv.forEach(a),xv.forEach(a),gn=h(t),$t=r(t,"P",{});var ac=o($t);Wi=r(ac,"A",{href:!0});var yv=o(Wi);Cp=s(yv,"Pix2Pix Zero"),yv.forEach(a),qp=s(ac," allows modifying an image so that one concept or subject is translated to another one while preserving general image semantics."),ac.forEach(a),_n=h(t),Ui=r(t,"P",{});var Av=o(Ui);Rp=s(Av,"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."),Av.forEach(a),En=h(t),Oi=r(t,"P",{});var $v=o(Oi);Mp=s($v,"Pix2Pix Zero can be used both to edit synthetic images as well as real images."),$v.forEach(a),wn=h(t),Ee=r(t,"UL",{});var Gl=o(Ee);Tt=r(Gl,"LI",{});var Cl=o(Tt);Lp=s(Cl,`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 `),Dt=r(Cl,"A",{href:!0,rel:!0});var Tv=o(Dt);jp=s(Tv,"Flan-T5"),Tv.forEach(a),zp=s(Cl," 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."),Cl.forEach(a),Hp=h(Gl),kt=r(Gl,"LI",{});var ql=o(kt);Zp=s(ql,"To edit a real image, one first generates an image caption using a model like "),St=r(ql,"A",{href:!0,rel:!0});var Dv=o(St);Bp=s(Dv,"BLIP"),Dv.forEach(a),Fp=s(ql,". 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."),ql.forEach(a),Gl.forEach(a),bn=h(t),w(we.$$.fragment,t),xn=h(t),be=r(t,"P",{});var Rl=o(be);Wp=s(Rl,`As mentioned above, Pix2Pix Zero includes optimizing the latents (and not any of the UNet, VAE, or the text encoder) to steer the generation toward a specific concept. This means that the overall
pipeline might require more memory than a standard `),Vi=r(Rl,"A",{href:!0});var kv=o(Vi);Up=s(kv,"StableDiffusionPipeline"),kv.forEach(a),Op=s(Rl,"."),Rl.forEach(a),Pn=h(t),xe=r(t,"P",{});var Ml=o(xe);Vp=s(Ml,"See "),Ji=r(Ml,"A",{href:!0});var Sv=o(Ji);Jp=s(Sv,"here"),Sv.forEach(a),Yp=s(Ml," for more information on how to use it."),Ml.forEach(a),yn=h(t),X=r(t,"H2",{class:!0});var Ll=o(X);Pe=r(Ll,"A",{id:!0,class:!0,href:!0});var Iv=o(Pe);xo=r(Iv,"SPAN",{});var Nv=o(xo);w(It.$$.fragment,Nv),Nv.forEach(a),Iv.forEach(a),Kp=h(Ll),Po=r(Ll,"SPAN",{});var Gv=o(Po);Qp=s(Gv,"Attend and Excite"),Gv.forEach(a),Ll.forEach(a),An=h(t),Yi=r(t,"P",{});var Cv=o(Yi);Nt=r(Cv,"A",{href:!0,rel:!0});var qv=o(Nt);Xp=s(qv,"Paper"),qv.forEach(a),Cv.forEach(a),$n=h(t),Gt=r(t,"P",{});var ic=o(Gt);Ki=r(ic,"A",{href:!0});var Rv=o(Ki);ed=s(Rv,"Attend and Excite"),Rv.forEach(a),td=s(ic," allows subjects in the prompt to be faithfully represented in the final image."),ic.forEach(a),Tn=h(t),Qi=r(t,"P",{});var Mv=o(Qi);ad=s(Mv,"A set of token indices are given as input, corresponding to the subjects in the prompt that need to be present in the image. During denoising, each token index is guaranteed to have a minimum attention threshold for at least one patch of the image. The intermediate latents are iteratively optimized during the denoising process to strengthen the attention of the most neglected subject token until the attention threshold is passed for all subject tokens."),Mv.forEach(a),Dn=h(t),ye=r(t,"P",{});var jl=o(ye);id=s(jl,"Like Pix2Pix Zero, Attend and Excite also involves a mini optimization loop (leaving the pre-trained weights untouched) in its pipeline and can require more memory than the usual "),yo=r(jl,"CODE",{});var Lv=o(yo);rd=s(Lv,"StableDiffusionPipeline"),Lv.forEach(a),od=s(jl,"."),jl.forEach(a),kn=h(t),Ae=r(t,"P",{});var zl=o(Ae);nd=s(zl,"See "),Xi=r(zl,"A",{href:!0});var jv=o(Xi);ld=s(jv,"here"),jv.forEach(a),sd=s(zl," for more information on how to use it."),zl.forEach(a),Sn=h(t),ee=r(t,"H2",{class:!0});var Hl=o(ee);$e=r(Hl,"A",{id:!0,class:!0,href:!0});var zv=o($e);Ao=r(zv,"SPAN",{});var Hv=o(Ao);w(Ct.$$.fragment,Hv),Hv.forEach(a),zv.forEach(a),fd=h(Hl),$o=r(Hl,"SPAN",{});var Zv=o($o);hd=s(Zv,"Semantic Guidance (SEGA)"),Zv.forEach(a),Hl.forEach(a),In=h(t),er=r(t,"P",{});var Bv=o(er);qt=r(Bv,"A",{href:!0,rel:!0});var Fv=o(qt);pd=s(Fv,"Paper"),Fv.forEach(a),Bv.forEach(a),Nn=h(t),tr=r(t,"P",{});var Wv=o(tr);dd=s(Wv,"SEGA allows applying or removing one or more concepts from an image. The strength of the concept can also be controlled. I.e. the smile concept can be used to incrementally increase or decrease the smile of a portrait."),Wv.forEach(a),Gn=h(t),ar=r(t,"P",{});var Uv=o(ar);ud=s(Uv,"Similar to how classifier free guidance provides guidance via empty prompt inputs, SEGA provides guidance on conceptual prompts. Multiple of these conceptual prompts can be applied simultaneously. Each conceptual prompt can either add or remove their concept depending on if the guidance is applied positively or negatively."),Uv.forEach(a),Cn=h(t),ir=r(t,"P",{});var Ov=o(ir);cd=s(Ov,"Unlike Pix2Pix Zero or Attend and Excite, SEGA directly interacts with the diffusion process instead of performing any explicit gradient-based optimization."),Ov.forEach(a),qn=h(t),Te=r(t,"P",{});var Zl=o(Te);md=s(Zl,"See "),rr=r(Zl,"A",{href:!0});var Vv=o(rr);vd=s(Vv,"here"),Vv.forEach(a),gd=s(Zl," for more information on how to use it."),Zl.forEach(a),Rn=h(t),te=r(t,"H2",{class:!0});var Bl=o(te);De=r(Bl,"A",{id:!0,class:!0,href:!0});var Jv=o(De);To=r(Jv,"SPAN",{});var Yv=o(To);w(Rt.$$.fragment,Yv),Yv.forEach(a),Jv.forEach(a),_d=h(Bl),Do=r(Bl,"SPAN",{});var Kv=o(Do);Ed=s(Kv,"Self-attention Guidance (SAG)"),Kv.forEach(a),Bl.forEach(a),Mn=h(t),or=r(t,"P",{});var Qv=o(or);Mt=r(Qv,"A",{href:!0,rel:!0});var Xv=o(Mt);wd=s(Xv,"Paper"),Xv.forEach(a),Qv.forEach(a),Ln=h(t),Lt=r(t,"P",{});var rc=o(Lt);nr=r(rc,"A",{href:!0});var e2=o(nr);bd=s(e2,"Self-attention Guidance"),e2.forEach(a),xd=s(rc," improves the general quality of images."),rc.forEach(a),jn=h(t),lr=r(t,"P",{});var t2=o(lr);Pd=s(t2,"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."),t2.forEach(a),zn=h(t),ke=r(t,"P",{});var Fl=o(ke);yd=s(Fl,"See "),sr=r(Fl,"A",{href:!0});var a2=o(sr);Ad=s(a2,"here"),a2.forEach(a),$d=s(Fl," for more information on how to use it."),Fl.forEach(a),Hn=h(t),ae=r(t,"H2",{class:!0});var Wl=o(ae);Se=r(Wl,"A",{id:!0,class:!0,href:!0});var i2=o(Se);ko=r(i2,"SPAN",{});var r2=o(ko);w(jt.$$.fragment,r2),r2.forEach(a),i2.forEach(a),Td=h(Wl),So=r(Wl,"SPAN",{});var o2=o(So);Dd=s(o2,"Depth2Image"),o2.forEach(a),Wl.forEach(a),Zn=h(t),fr=r(t,"P",{});var n2=o(fr);zt=r(n2,"A",{href:!0,rel:!0});var l2=o(zt);kd=s(l2,"Project"),l2.forEach(a),n2.forEach(a),Bn=h(t),Ht=r(t,"P",{});var oc=o(Ht);hr=r(oc,"A",{href:!0});var s2=o(hr);Sd=s(s2,"Depth2Image"),s2.forEach(a),Id=s(oc," is fine-tuned from Stable Diffusion to better preserve semantics for text guided image variation."),oc.forEach(a),Fn=h(t),pr=r(t,"P",{});var f2=o(pr);Nd=s(f2,"It conditions on a monocular depth estimate of the original image."),f2.forEach(a),Wn=h(t),Ie=r(t,"P",{});var Ul=o(Ie);Gd=s(Ul,"See "),dr=r(Ul,"A",{href:!0});var h2=o(dr);Cd=s(h2,"here"),h2.forEach(a),qd=s(Ul," for more information on how to use it."),Ul.forEach(a),Un=h(t),w(Ne.$$.fragment,t),On=h(t),ie=r(t,"H2",{class:!0});var Ol=o(ie);Ge=r(Ol,"A",{id:!0,class:!0,href:!0});var p2=o(Ge);Io=r(p2,"SPAN",{});var d2=o(Io);w(Zt.$$.fragment,d2),d2.forEach(a),p2.forEach(a),Rd=h(Ol),No=r(Ol,"SPAN",{});var u2=o(No);Md=s(u2,"MultiDiffusion Panorama"),u2.forEach(a),Ol.forEach(a),Vn=h(t),ur=r(t,"P",{});var c2=o(ur);Bt=r(c2,"A",{href:!0,rel:!0});var m2=o(Bt);Ld=s(m2,"Paper"),m2.forEach(a),c2.forEach(a),Jn=h(t),Ce=r(t,"P",{});var Vl=o(Ce);jd=s(Vl,`MultiDiffusion defines a new generation process over a pre-trained diffusion model. This process binds together multiple diffusion generation methods that can be readily applied to generate high quality and diverse images. Results adhere to user-provided controls, such as desired aspect ratio (e.g., panorama), and spatial guiding signals, ranging from tight segmentation masks to bounding boxes.
`),cr=r(Vl,"A",{href:!0});var v2=o(cr);zd=s(v2,"MultiDiffusion Panorama"),v2.forEach(a),Hd=s(Vl," allows to generate high-quality images at arbitrary aspect ratios (e.g., panoramas)."),Vl.forEach(a),Yn=h(t),qe=r(t,"P",{});var Jl=o(qe);Zd=s(Jl,"See "),mr=r(Jl,"A",{href:!0});var g2=o(mr);Bd=s(g2,"here"),g2.forEach(a),Fd=s(Jl," for more information on how to use it to generate panoramic images."),Jl.forEach(a),Kn=h(t),re=r(t,"H2",{class:!0});var Yl=o(re);Re=r(Yl,"A",{id:!0,class:!0,href:!0});var _2=o(Re);Go=r(_2,"SPAN",{});var E2=o(Go);w(Ft.$$.fragment,E2),E2.forEach(a),_2.forEach(a),Wd=h(Yl),Co=r(Yl,"SPAN",{});var w2=o(Co);Ud=s(w2,"Fine-tuning your own models"),w2.forEach(a),Yl.forEach(a),Qn=h(t),vr=r(t,"P",{});var b2=o(vr);Od=s(b2,"In addition to pre-trained models, Diffusers has training scripts for fine-tuning models on user-provided data."),b2.forEach(a),Xn=h(t),oe=r(t,"H2",{class:!0});var Kl=o(oe);Me=r(Kl,"A",{id:!0,class:!0,href:!0});var x2=o(Me);qo=r(x2,"SPAN",{});var P2=o(qo);w(Wt.$$.fragment,P2),P2.forEach(a),x2.forEach(a),Vd=h(Kl),Ro=r(Kl,"SPAN",{});var y2=o(Ro);Jd=s(y2,"DreamBooth"),y2.forEach(a),Kl.forEach(a),el=h(t),Ut=r(t,"P",{});var nc=o(Ut);gr=r(nc,"A",{href:!0});var A2=o(gr);Yd=s(A2,"DreamBooth"),A2.forEach(a),Kd=s(nc," fine-tunes a model to teach it about a new subject. I.e. a few pictures of a person can be used to generate images of that person in different styles."),nc.forEach(a),tl=h(t),Le=r(t,"P",{});var Ql=o(Le);Qd=s(Ql,"See "),_r=r(Ql,"A",{href:!0});var $2=o(_r);Xd=s($2,"here"),$2.forEach(a),eu=s(Ql," for more information on how to use it."),Ql.forEach(a),al=h(t),ne=r(t,"H2",{class:!0});var Xl=o(ne);je=r(Xl,"A",{id:!0,class:!0,href:!0});var T2=o(je);Mo=r(T2,"SPAN",{});var D2=o(Mo);w(Ot.$$.fragment,D2),D2.forEach(a),T2.forEach(a),tu=h(Xl),Lo=r(Xl,"SPAN",{});var k2=o(Lo);au=s(k2,"Textual Inversion"),k2.forEach(a),Xl.forEach(a),il=h(t),Vt=r(t,"P",{});var lc=o(Vt);Er=r(lc,"A",{href:!0});var S2=o(Er);iu=s(S2,"Textual Inversion"),S2.forEach(a),ru=s(lc," 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."),lc.forEach(a),rl=h(t),ze=r(t,"P",{});var es=o(ze);ou=s(es,"See "),wr=r(es,"A",{href:!0});var I2=o(wr);nu=s(I2,"here"),I2.forEach(a),lu=s(es," for more information on how to use it."),es.forEach(a),ol=h(t),le=r(t,"H2",{class:!0});var ts=o(le);He=r(ts,"A",{id:!0,class:!0,href:!0});var N2=o(He);jo=r(N2,"SPAN",{});var G2=o(jo);w(Jt.$$.fragment,G2),G2.forEach(a),N2.forEach(a),su=h(ts),zo=r(ts,"SPAN",{});var C2=o(zo);fu=s(C2,"ControlNet"),C2.forEach(a),ts.forEach(a),nl=h(t),br=r(t,"P",{});var q2=o(br);Yt=r(q2,"A",{href:!0,rel:!0});var R2=o(Yt);hu=s(R2,"Paper"),R2.forEach(a),q2.forEach(a),ll=h(t),se=r(t,"P",{});var Xo=o(se);xr=r(Xo,"A",{href:!0});var M2=o(xr);pu=s(M2,"ControlNet"),M2.forEach(a),du=s(Xo,` is an auxiliary network which adds an extra condition.
`),Pr=r(Xo,"A",{href:!0});var L2=o(Pr);uu=s(L2,"ControlNet"),L2.forEach(a),cu=s(Xo,` is an auxiliary network which adds an extra condition.
There are 8 canonical pre-trained ControlNets trained on different conditionings such as edge detection, scribbles,
depth maps, and semantic segmentations.`),Xo.forEach(a),sl=h(t),Ze=r(t,"P",{});var as=o(Ze);mu=s(as,"See "),yr=r(as,"A",{href:!0});var j2=o(yr);vu=s(j2,"here"),j2.forEach(a),gu=s(as," for more information on how to use it."),as.forEach(a),fl=h(t),fe=r(t,"H2",{class:!0});var is=o(fe);Be=r(is,"A",{id:!0,class:!0,href:!0});var z2=o(Be);Ho=r(z2,"SPAN",{});var H2=o(Ho);w(Kt.$$.fragment,H2),H2.forEach(a),z2.forEach(a),_u=h(is),Zo=r(is,"SPAN",{});var Z2=o(Zo);Eu=s(Z2,"Prompt Weighting"),Z2.forEach(a),is.forEach(a),hl=h(t),Ar=r(t,"P",{});var B2=o(Ar);wu=s(B2,`Prompt weighting is a simple technique that puts more attention weight on certain parts of the text
input.`),B2.forEach(a),pl=h(t),Fe=r(t,"P",{});var rs=o(Fe);bu=s(rs,"For a more in-detail explanation and examples, see "),$r=r(rs,"A",{href:!0});var F2=o($r);xu=s(F2,"here"),F2.forEach(a),Pu=s(rs,"."),rs.forEach(a),dl=h(t),he=r(t,"H2",{class:!0});var os=o(he);We=r(os,"A",{id:!0,class:!0,href:!0});var W2=o(We);Bo=r(W2,"SPAN",{});var U2=o(Bo);w(Qt.$$.fragment,U2),U2.forEach(a),W2.forEach(a),yu=h(os),Fo=r(os,"SPAN",{});var O2=o(Fo);Au=s(O2,"Custom Diffusion"),O2.forEach(a),os.forEach(a),ul=h(t),Xt=r(t,"P",{});var sc=o(Xt);Tr=r(sc,"A",{href:!0});var V2=o(Tr);$u=s(V2,"Custom Diffusion"),V2.forEach(a),Tu=s(sc,` only fine-tunes the cross-attention maps of a pre-trained
text-to-image diffusion model. It also allows for additionally performing textual inversion. It supports
multi-concept training by design. Like DreamBooth and Textual Inversion, Custom Diffusion is also used to
teach a pre-trained text-to-image diffusion model about new concepts to generate outputs involving the
concept(s) of interest.`),sc.forEach(a),cl=h(t),Ue=r(t,"P",{});var ns=o(Ue);Du=s(ns,"For more details, check out our "),Dr=r(ns,"A",{href:!0});var J2=o(Dr);ku=s(J2,"official doc"),J2.forEach(a),Su=s(ns,"."),ns.forEach(a),ml=h(t),pe=r(t,"H2",{class:!0});var ls=o(pe);Oe=r(ls,"A",{id:!0,class:!0,href:!0});var Y2=o(Oe);Wo=r(Y2,"SPAN",{});var K2=o(Wo);w(ea.$$.fragment,K2),K2.forEach(a),Y2.forEach(a),Iu=h(ls),Uo=r(ls,"SPAN",{});var Q2=o(Uo);Nu=s(Q2,"Model Editing"),Q2.forEach(a),ls.forEach(a),vl=h(t),kr=r(t,"P",{});var X2=o(kr);ta=r(X2,"A",{href:!0,rel:!0});var eg=o(ta);Gu=s(eg,"Paper"),eg.forEach(a),X2.forEach(a),gl=h(t),Ve=r(t,"P",{});var ss=o(Ve);Cu=s(ss,"The "),Sr=r(ss,"A",{href:!0});var tg=o(Sr);qu=s(tg,"text-to-image model editing pipeline"),tg.forEach(a),Ru=s(ss,` 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.`),ss.forEach(a),_l=h(t),Je=r(t,"P",{});var fs=o(Je);Mu=s(fs,"To know more details, check out the "),Ir=r(fs,"A",{href:!0});var ag=o(Ir);Lu=s(ag,"official doc"),ag.forEach(a),ju=s(fs,"."),fs.forEach(a),El=h(t),de=r(t,"H2",{class:!0});var hs=o(de);Ye=r(hs,"A",{id:!0,class:!0,href:!0});var ig=o(Ye);Oo=r(ig,"SPAN",{});var rg=o(Oo);w(aa.$$.fragment,rg),rg.forEach(a),ig.forEach(a),zu=h(hs),Vo=r(hs,"SPAN",{});var og=o(Vo);Hu=s(og,"DiffEdit"),og.forEach(a),hs.forEach(a),wl=h(t),Nr=r(t,"P",{});var ng=o(Nr);ia=r(ng,"A",{href:!0,rel:!0});var lg=o(ia);Zu=s(lg,"Paper"),lg.forEach(a),ng.forEach(a),bl=h(t),ra=r(t,"P",{});var fc=o(ra);Gr=r(fc,"A",{href:!0});var sg=o(Gr);Bu=s(sg,"DiffEdit"),sg.forEach(a),Fu=s(fc,` allows for semantic editing of input images along with
input prompts while preserving the original input images as much as possible.`),fc.forEach(a),xl=h(t),Ke=r(t,"P",{});var ps=o(Ke);Wu=s(ps,"To know more details, check out the "),Cr=r(ps,"A",{href:!0});var fg=o(Cr);Uu=s(fg,"official doc"),fg.forEach(a),Ou=s(ps,"."),ps.forEach(a),Pl=h(t),ue=r(t,"H2",{class:!0});var ds=o(ue);Qe=r(ds,"A",{id:!0,class:!0,href:!0});var hg=o(Qe);Jo=r(hg,"SPAN",{});var pg=o(Jo);w(oa.$$.fragment,pg),pg.forEach(a),hg.forEach(a),Vu=h(ds),Yo=r(ds,"SPAN",{});var dg=o(Yo);Ju=s(dg,"T2I-Adapter"),dg.forEach(a),ds.forEach(a),yl=h(t),qr=r(t,"P",{});var ug=o(qr);na=r(ug,"A",{href:!0,rel:!0});var cg=o(na);Yu=s(cg,"Paper"),cg.forEach(a),ug.forEach(a),Al=h(t),la=r(t,"P",{});var hc=o(la);Rr=r(hc,"A",{href:!0});var mg=o(Rr);Ku=s(mg,"T2I-Adapter"),mg.forEach(a),Qu=s(hc,` is an auxiliary network which adds an extra condition.
There are 8 canonical pre-trained adapters trained on different conditionings such as edge detection, sketch,
depth maps, and semantic segmentations.`),hc.forEach(a),$l=h(t),Xe=r(t,"P",{});var us=o(Xe);Xu=s(us,"See "),Mr=r(us,"A",{href:!0});var vg=o(Mr);ec=s(vg,"here"),vg.forEach(a),tc=s(us," for more information on how to use it."),us.forEach(a),this.h()},h(){n(g,"name","hf:doc:metadata"),n(g,"content",JSON.stringify($g)),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(_t,"href","https://discuss.huggingface.co/"),n(_t,"rel","nofollow"),n(Et,"href","https://github.com/huggingface/diffusers/issues"),n(Et,"rel","nofollow"),n(va,"href","#instruct-pix2pix"),n(ga,"href","#pix2pixzero"),n(_a,"href","#attend-and-excite"),n(Ea,"href","#semantic-guidance"),n(wa,"href","#self-attention-guidance"),n(ba,"href","#depth2image"),n(xa,"href","#multidiffusion-panorama"),n(Pa,"href","#dreambooth"),n(ya,"href","#textual-inversion"),n(Aa,"href","#controlnet"),n($a,"href","#prompt-weighting"),n(Ta,"href","#custom-diffusion"),n(Da,"href","#model-editing"),n(ka,"href","#diffedit"),n(Ia,"align","center"),n(Na,"align","center"),n(Ga,"align","center"),n(Ca,"align","center"),n(Ra,"href","#instruct-pix2pix"),n(qa,"align","center"),n(Ma,"align","center"),n(La,"align","center"),n(S,"align","center"),n(za,"href","#pix2pixzero"),n(ja,"align","center"),n(Ha,"align","center"),n(Za,"align","center"),n(ro,"align","center"),n(Fa,"href","#attend-and-excite"),n(Ba,"align","center"),n(Wa,"align","center"),n(Ua,"align","center"),n(oo,"align","center"),n(Va,"href","#semantic-guidance"),n(Oa,"align","center"),n(Ja,"align","center"),n(Ya,"align","center"),n(no,"align","center"),n(Qa,"href","#self-attention-guidance"),n(Ka,"align","center"),n(Xa,"align","center"),n(ei,"align","center"),n(lo,"align","center"),n(ai,"href","#depth2image"),n(ti,"align","center"),n(ii,"align","center"),n(ri,"align","center"),n(so,"align","center"),n(ni,"href","#multidiffusion-panorama"),n(oi,"align","center"),n(li,"align","center"),n(si,"align","center"),n(fo,"align","center"),n(hi,"href","#dreambooth"),n(fi,"align","center"),n(pi,"align","center"),n(di,"align","center"),n(ho,"align","center"),n(ci,"href","#textual-inversion"),n(ui,"align","center"),n(mi,"align","center"),n(vi,"align","center"),n(po,"align","center"),n(_i,"href","#controlnet"),n(gi,"align","center"),n(Ei,"align","center"),n(wi,"align","center"),n(V,"align","center"),n(xi,"href","#prompt-weighting"),n(bi,"align","center"),n(Pi,"align","center"),n(yi,"align","center"),n(uo,"align","center"),n($i,"href","#custom-diffusion"),n(Ai,"align","center"),n(Ti,"align","center"),n(Di,"align","center"),n(co,"align","center"),n(Si,"href","#model-editing"),n(ki,"align","center"),n(Ii,"align","center"),n(Ni,"align","center"),n(mo,"align","center"),n(Ci,"href","#diffedit"),n(Gi,"align","center"),n(qi,"align","center"),n(Ri,"align","center"),n(vo,"align","center"),n(Li,"href","#t2i-adapter"),n(Mi,"align","center"),n(ji,"align","center"),n(zi,"align","center"),n(go,"align","center"),n(ve,"id","instruct-pix2pix"),n(ve,"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(ve,"href","#instruct-pix2pix"),n(Y,"class","relative group"),n(xt,"href","https://arxiv.org/abs/2211.09800"),n(xt,"rel","nofollow"),n(Zi,"href","../api/pipelines/stable_diffusion/pix2pix"),n(Pt,"href","https://openai.com/blog/instruction-following/"),n(Pt,"rel","nofollow"),n(Bi,"href","../api/pipelines/stable_diffusion/pix2pix"),n(_e,"id","pix2pix-zero"),n(_e,"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(_e,"href","#pix2pix-zero"),n(Q,"class","relative group"),n(At,"href","https://arxiv.org/abs/2302.03027"),n(At,"rel","nofollow"),n(Wi,"href","../api/pipelines/stable_diffusion/pix2pix_zero"),n(Dt,"href","https://huggingface.co/docs/transformers/model_doc/flan-t5"),n(Dt,"rel","nofollow"),n(St,"href","https://huggingface.co/docs/transformers/model_doc/blip"),n(St,"rel","nofollow"),n(Vi,"href","../api/pipelines/stable_diffusion/text2img"),n(Ji,"href","../api/pipelines/stable_diffusion/pix2pix_zero"),n(Pe,"id","attend-and-excite"),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","#attend-and-excite"),n(X,"class","relative group"),n(Nt,"href","https://arxiv.org/abs/2301.13826"),n(Nt,"rel","nofollow"),n(Ki,"href","../api/pipelines/stable_diffusion/attend_and_excite"),n(Xi,"href","../api/pipelines/stable_diffusion/attend_and_excite"),n($e,"id","semantic-guidance-sega"),n($e,"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($e,"href","#semantic-guidance-sega"),n(ee,"class","relative group"),n(qt,"href","https://arxiv.org/abs/2301.12247"),n(qt,"rel","nofollow"),n(rr,"href","../api/pipelines/semantic_stable_diffusion"),n(De,"id","selfattention-guidance-sag"),n(De,"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(De,"href","#selfattention-guidance-sag"),n(te,"class","relative group"),n(Mt,"href","https://arxiv.org/abs/2210.00939"),n(Mt,"rel","nofollow"),n(nr,"href","../api/pipelines/stable_diffusion/self_attention_guidance"),n(sr,"href","../api/pipelines/stable_diffusion/self_attention_guidance"),n(Se,"id","depth2image"),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","#depth2image"),n(ae,"class","relative group"),n(zt,"href","https://huggingface.co/stabilityai/stable-diffusion-2-depth"),n(zt,"rel","nofollow"),n(hr,"href","../pipelines/stable_diffusion_2#depthtoimage"),n(dr,"href","../api/pipelines/stable_diffusion_2#depthtoimage"),n(Ge,"id","multidiffusion-panorama"),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","#multidiffusion-panorama"),n(ie,"class","relative group"),n(Bt,"href","https://arxiv.org/abs/2302.08113"),n(Bt,"rel","nofollow"),n(cr,"href","../api/pipelines/stable_diffusion/panorama"),n(mr,"href","../api/pipelines/stable_diffusion/panorama"),n(Re,"id","finetuning-your-own-models"),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","#finetuning-your-own-models"),n(re,"class","relative group"),n(Me,"id","dreambooth"),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","#dreambooth"),n(oe,"class","relative group"),n(gr,"href","../training/dreambooth"),n(_r,"href","../training/dreambooth"),n(je,"id","textual-inversion"),n(je,"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(je,"href","#textual-inversion"),n(ne,"class","relative group"),n(Er,"href","../training/text_inversion"),n(wr,"href","../training/text_inversion"),n(He,"id","controlnet"),n(He,"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(He,"href","#controlnet"),n(le,"class","relative group"),n(Yt,"href","https://arxiv.org/abs/2302.05543"),n(Yt,"rel","nofollow"),n(xr,"href","../api/pipelines/controlnet"),n(Pr,"href","../api/pipelines/controlnet"),n(yr,"href","../api/pipelines/controlnet"),n(Be,"id","prompt-weighting"),n(Be,"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(Be,"href","#prompt-weighting"),n(fe,"class","relative group"),n($r,"href","../using-diffusers/weighted_prompts"),n(We,"id","custom-diffusion"),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","#custom-diffusion"),n(he,"class","relative group"),n(Tr,"href","../training/custom_diffusion"),n(Dr,"href","../training/custom_diffusion"),n(Oe,"id","model-editing"),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","#model-editing"),n(pe,"class","relative group"),n(ta,"href","https://arxiv.org/abs/2303.08084"),n(ta,"rel","nofollow"),n(Sr,"href","../api/pipelines/stable_diffusion/model_editing"),n(Ir,"href","../api/pipelines/stable_diffusion/model_editing"),n(Ye,"id","diffedit"),n(Ye,"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(Ye,"href","#diffedit"),n(de,"class","relative group"),n(ia,"href","https://arxiv.org/abs/2210.11427"),n(ia,"rel","nofollow"),n(Gr,"href","../api/pipelines/stable_diffusion/diffedit"),n(Cr,"href","../api/pipelines/stable_diffusion/model_editing"),n(Qe,"id","t2iadapter"),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","#t2iadapter"),n(ue,"class","relative group"),n(na,"href","https://arxiv.org/abs/2302.08453"),n(na,"rel","nofollow"),n(Rr,"href","../api/pipelines/stable_diffusion/adapter"),n(Mr,"href","../api/pipelines/stable_diffusion/adapter")},m(t,p){e(document.head,g),d(t,D,p),d(t,_,p),e(_,A),e(A,J),b(T,J,null),e(_,O),e(_,ce),e(ce,cs),d(t,en,p),d(t,ha,p),e(ha,ms),d(t,tn,p),d(t,pa,p),e(pa,vs),d(t,an,p),d(t,da,p),e(da,gs),d(t,rn,p),d(t,k,p),e(k,_s),e(k,jr),e(jr,Es),e(k,ws),e(k,_t),e(_t,bs),e(k,xs),e(k,Et),e(Et,Ps),e(k,ys),d(t,on,p),d(t,ua,p),e(ua,As),d(t,nn,p),d(t,ca,p),e(ca,$s),d(t,ln,p),d(t,ma,p),e(ma,Ts),d(t,sn,p),d(t,c,p),e(c,zr),e(zr,va),e(va,Ds),e(c,ks),e(c,Hr),e(Hr,ga),e(ga,Ss),e(c,Is),e(c,Zr),e(Zr,_a),e(_a,Ns),e(c,Gs),e(c,Br),e(Br,Ea),e(Ea,Cs),e(c,qs),e(c,Fr),e(Fr,wa),e(wa,Rs),e(c,Ms),e(c,Wr),e(Wr,ba),e(ba,Ls),e(c,js),e(c,Ur),e(Ur,xa),e(xa,zs),e(c,Hs),e(c,Or),e(Or,Pa),e(Pa,Zs),e(c,Bs),e(c,Vr),e(Vr,ya),e(ya,Fs),e(c,Ws),e(c,Jr),e(Jr,Aa),e(Aa,Us),e(c,Os),e(c,Yr),e(Yr,$a),e($a,Vs),e(c,Js),e(c,Kr),e(Kr,Ta),e(Ta,Ys),e(c,Ks),e(c,Qr),e(Qr,Da),e(Da,Qs),e(c,Xs),e(c,Xr),e(Xr,ka),e(ka,ef),d(t,fn,p),d(t,Sa,p),e(Sa,tf),d(t,hn,p),d(t,me,p),e(me,eo),e(eo,I),e(I,Ia),e(Ia,to),e(to,af),e(I,rf),e(I,Na),e(Na,ao),e(ao,of),e(I,nf),e(I,Ga),e(Ga,wt),e(wt,lf),e(wt,sf),e(wt,ff),e(I,hf),e(I,Ca),e(Ca,io),e(io,pf),e(me,df),e(me,u),e(u,N),e(N,qa),e(qa,Ra),e(Ra,uf),e(N,cf),e(N,Ma),e(Ma,mf),e(N,vf),e(N,La),e(La,gf),e(N,_f),e(N,S),e(S,Ef),e(S,wf),e(S,bf),e(S,xf),e(S,Pf),e(S,yf),e(S,Af),e(u,$f),e(u,G),e(G,ja),e(ja,za),e(za,Tf),e(G,Df),e(G,Ha),e(Ha,kf),e(G,Sf),e(G,Za),e(Za,If),e(G,Nf),e(G,ro),e(u,Gf),e(u,C),e(C,Ba),e(Ba,Fa),e(Fa,Cf),e(C,qf),e(C,Wa),e(Wa,Rf),e(C,Mf),e(C,Ua),e(Ua,Lf),e(C,jf),e(C,oo),e(u,zf),e(u,q),e(q,Oa),e(Oa,Va),e(Va,Hf),e(q,Zf),e(q,Ja),e(Ja,Bf),e(q,Ff),e(q,Ya),e(Ya,Wf),e(q,Uf),e(q,no),e(u,Of),e(u,R),e(R,Ka),e(Ka,Qa),e(Qa,Vf),e(R,Jf),e(R,Xa),e(Xa,Yf),e(R,Kf),e(R,ei),e(ei,Qf),e(R,Xf),e(R,lo),e(u,eh),e(u,M),e(M,ti),e(ti,ai),e(ai,th),e(M,ah),e(M,ii),e(ii,ih),e(M,rh),e(M,ri),e(ri,oh),e(M,nh),e(M,so),e(u,lh),e(u,L),e(L,oi),e(oi,ni),e(ni,sh),e(L,fh),e(L,li),e(li,hh),e(L,ph),e(L,si),e(si,dh),e(L,uh),e(L,fo),e(u,ch),e(u,j),e(j,fi),e(fi,hi),e(hi,mh),e(j,vh),e(j,pi),e(pi,gh),e(j,_h),e(j,di),e(di,Eh),e(j,wh),e(j,ho),e(u,bh),e(u,z),e(z,ui),e(ui,ci),e(ci,xh),e(z,Ph),e(z,mi),e(mi,yh),e(z,Ah),e(z,vi),e(vi,$h),e(z,Th),e(z,po),e(u,Dh),e(u,H),e(H,gi),e(gi,_i),e(_i,kh),e(H,Sh),e(H,Ei),e(Ei,Ih),e(H,Nh),e(H,wi),e(wi,Gh),e(H,Ch),e(H,V),e(V,qh),e(V,Rh),e(V,Mh),e(V,Lh),e(V,jh),e(u,zh),e(u,Z),e(Z,bi),e(bi,xi),e(xi,Hh),e(Z,Zh),e(Z,Pi),e(Pi,Bh),e(Z,Fh),e(Z,yi),e(yi,Wh),e(Z,Uh),e(Z,uo),e(u,Oh),e(u,B),e(B,Ai),e(Ai,$i),e($i,Vh),e(B,Jh),e(B,Ti),e(Ti,Yh),e(B,Kh),e(B,Di),e(Di,Qh),e(B,Xh),e(B,co),e(u,ep),e(u,F),e(F,ki),e(ki,Si),e(Si,tp),e(F,ap),e(F,Ii),e(Ii,ip),e(F,rp),e(F,Ni),e(Ni,op),e(F,np),e(F,mo),e(u,lp),e(u,W),e(W,Gi),e(Gi,Ci),e(Ci,sp),e(W,fp),e(W,qi),e(qi,hp),e(W,pp),e(W,Ri),e(Ri,dp),e(W,up),e(W,vo),e(u,cp),e(u,U),e(U,Mi),e(Mi,Li),e(Li,mp),e(U,vp),e(U,ji),e(ji,gp),e(U,_p),e(U,zi),e(zi,Ep),e(U,wp),e(U,go),d(t,pn,p),d(t,Y,p),e(Y,ve),e(ve,_o),b(bt,_o,null),e(Y,bp),e(Y,Eo),e(Eo,xp),d(t,dn,p),d(t,Hi,p),e(Hi,xt),e(xt,Pp),d(t,un,p),d(t,K,p),e(K,Zi),e(Zi,yp),e(K,Ap),e(K,Pt),e(Pt,$p),e(K,Tp),d(t,cn,p),d(t,ge,p),e(ge,Dp),e(ge,Bi),e(Bi,kp),e(ge,Sp),d(t,mn,p),d(t,Q,p),e(Q,_e),e(_e,wo),b(yt,wo,null),e(Q,Ip),e(Q,bo),e(bo,Np),d(t,vn,p),d(t,Fi,p),e(Fi,At),e(At,Gp),d(t,gn,p),d(t,$t,p),e($t,Wi),e(Wi,Cp),e($t,qp),d(t,_n,p),d(t,Ui,p),e(Ui,Rp),d(t,En,p),d(t,Oi,p),e(Oi,Mp),d(t,wn,p),d(t,Ee,p),e(Ee,Tt),e(Tt,Lp),e(Tt,Dt),e(Dt,jp),e(Tt,zp),e(Ee,Hp),e(Ee,kt),e(kt,Zp),e(kt,St),e(St,Bp),e(kt,Fp),d(t,bn,p),b(we,t,p),d(t,xn,p),d(t,be,p),e(be,Wp),e(be,Vi),e(Vi,Up),e(be,Op),d(t,Pn,p),d(t,xe,p),e(xe,Vp),e(xe,Ji),e(Ji,Jp),e(xe,Yp),d(t,yn,p),d(t,X,p),e(X,Pe),e(Pe,xo),b(It,xo,null),e(X,Kp),e(X,Po),e(Po,Qp),d(t,An,p),d(t,Yi,p),e(Yi,Nt),e(Nt,Xp),d(t,$n,p),d(t,Gt,p),e(Gt,Ki),e(Ki,ed),e(Gt,td),d(t,Tn,p),d(t,Qi,p),e(Qi,ad),d(t,Dn,p),d(t,ye,p),e(ye,id),e(ye,yo),e(yo,rd),e(ye,od),d(t,kn,p),d(t,Ae,p),e(Ae,nd),e(Ae,Xi),e(Xi,ld),e(Ae,sd),d(t,Sn,p),d(t,ee,p),e(ee,$e),e($e,Ao),b(Ct,Ao,null),e(ee,fd),e(ee,$o),e($o,hd),d(t,In,p),d(t,er,p),e(er,qt),e(qt,pd),d(t,Nn,p),d(t,tr,p),e(tr,dd),d(t,Gn,p),d(t,ar,p),e(ar,ud),d(t,Cn,p),d(t,ir,p),e(ir,cd),d(t,qn,p),d(t,Te,p),e(Te,md),e(Te,rr),e(rr,vd),e(Te,gd),d(t,Rn,p),d(t,te,p),e(te,De),e(De,To),b(Rt,To,null),e(te,_d),e(te,Do),e(Do,Ed),d(t,Mn,p),d(t,or,p),e(or,Mt),e(Mt,wd),d(t,Ln,p),d(t,Lt,p),e(Lt,nr),e(nr,bd),e(Lt,xd),d(t,jn,p),d(t,lr,p),e(lr,Pd),d(t,zn,p),d(t,ke,p),e(ke,yd),e(ke,sr),e(sr,Ad),e(ke,$d),d(t,Hn,p),d(t,ae,p),e(ae,Se),e(Se,ko),b(jt,ko,null),e(ae,Td),e(ae,So),e(So,Dd),d(t,Zn,p),d(t,fr,p),e(fr,zt),e(zt,kd),d(t,Bn,p),d(t,Ht,p),e(Ht,hr),e(hr,Sd),e(Ht,Id),d(t,Fn,p),d(t,pr,p),e(pr,Nd),d(t,Wn,p),d(t,Ie,p),e(Ie,Gd),e(Ie,dr),e(dr,Cd),e(Ie,qd),d(t,Un,p),b(Ne,t,p),d(t,On,p),d(t,ie,p),e(ie,Ge),e(Ge,Io),b(Zt,Io,null),e(ie,Rd),e(ie,No),e(No,Md),d(t,Vn,p),d(t,ur,p),e(ur,Bt),e(Bt,Ld),d(t,Jn,p),d(t,Ce,p),e(Ce,jd),e(Ce,cr),e(cr,zd),e(Ce,Hd),d(t,Yn,p),d(t,qe,p),e(qe,Zd),e(qe,mr),e(mr,Bd),e(qe,Fd),d(t,Kn,p),d(t,re,p),e(re,Re),e(Re,Go),b(Ft,Go,null),e(re,Wd),e(re,Co),e(Co,Ud),d(t,Qn,p),d(t,vr,p),e(vr,Od),d(t,Xn,p),d(t,oe,p),e(oe,Me),e(Me,qo),b(Wt,qo,null),e(oe,Vd),e(oe,Ro),e(Ro,Jd),d(t,el,p),d(t,Ut,p),e(Ut,gr),e(gr,Yd),e(Ut,Kd),d(t,tl,p),d(t,Le,p),e(Le,Qd),e(Le,_r),e(_r,Xd),e(Le,eu),d(t,al,p),d(t,ne,p),e(ne,je),e(je,Mo),b(Ot,Mo,null),e(ne,tu),e(ne,Lo),e(Lo,au),d(t,il,p),d(t,Vt,p),e(Vt,Er),e(Er,iu),e(Vt,ru),d(t,rl,p),d(t,ze,p),e(ze,ou),e(ze,wr),e(wr,nu),e(ze,lu),d(t,ol,p),d(t,le,p),e(le,He),e(He,jo),b(Jt,jo,null),e(le,su),e(le,zo),e(zo,fu),d(t,nl,p),d(t,br,p),e(br,Yt),e(Yt,hu),d(t,ll,p),d(t,se,p),e(se,xr),e(xr,pu),e(se,du),e(se,Pr),e(Pr,uu),e(se,cu),d(t,sl,p),d(t,Ze,p),e(Ze,mu),e(Ze,yr),e(yr,vu),e(Ze,gu),d(t,fl,p),d(t,fe,p),e(fe,Be),e(Be,Ho),b(Kt,Ho,null),e(fe,_u),e(fe,Zo),e(Zo,Eu),d(t,hl,p),d(t,Ar,p),e(Ar,wu),d(t,pl,p),d(t,Fe,p),e(Fe,bu),e(Fe,$r),e($r,xu),e(Fe,Pu),d(t,dl,p),d(t,he,p),e(he,We),e(We,Bo),b(Qt,Bo,null),e(he,yu),e(he,Fo),e(Fo,Au),d(t,ul,p),d(t,Xt,p),e(Xt,Tr),e(Tr,$u),e(Xt,Tu),d(t,cl,p),d(t,Ue,p),e(Ue,Du),e(Ue,Dr),e(Dr,ku),e(Ue,Su),d(t,ml,p),d(t,pe,p),e(pe,Oe),e(Oe,Wo),b(ea,Wo,null),e(pe,Iu),e(pe,Uo),e(Uo,Nu),d(t,vl,p),d(t,kr,p),e(kr,ta),e(ta,Gu),d(t,gl,p),d(t,Ve,p),e(Ve,Cu),e(Ve,Sr),e(Sr,qu),e(Ve,Ru),d(t,_l,p),d(t,Je,p),e(Je,Mu),e(Je,Ir),e(Ir,Lu),e(Je,ju),d(t,El,p),d(t,de,p),e(de,Ye),e(Ye,Oo),b(aa,Oo,null),e(de,zu),e(de,Vo),e(Vo,Hu),d(t,wl,p),d(t,Nr,p),e(Nr,ia),e(ia,Zu),d(t,bl,p),d(t,ra,p),e(ra,Gr),e(Gr,Bu),e(ra,Fu),d(t,xl,p),d(t,Ke,p),e(Ke,Wu),e(Ke,Cr),e(Cr,Uu),e(Ke,Ou),d(t,Pl,p),d(t,ue,p),e(ue,Qe),e(Qe,Jo),b(oa,Jo,null),e(ue,Vu),e(ue,Yo),e(Yo,Ju),d(t,yl,p),d(t,qr,p),e(qr,na),e(na,Yu),d(t,Al,p),d(t,la,p),e(la,Rr),e(Rr,Ku),e(la,Qu),d(t,$l,p),d(t,Xe,p),e(Xe,Xu),e(Xe,Mr),e(Mr,ec),e(Xe,tc),Tl=!0},p(t,[p]){const sa={};p&2&&(sa.$$scope={dirty:p,ctx:t}),we.$set(sa);const Ko={};p&2&&(Ko.$$scope={dirty:p,ctx:t}),Ne.$set(Ko)},i(t){Tl||(x(T.$$.fragment,t),x(bt.$$.fragment,t),x(yt.$$.fragment,t),x(we.$$.fragment,t),x(It.$$.fragment,t),x(Ct.$$.fragment,t),x(Rt.$$.fragment,t),x(jt.$$.fragment,t),x(Ne.$$.fragment,t),x(Zt.$$.fragment,t),x(Ft.$$.fragment,t),x(Wt.$$.fragment,t),x(Ot.$$.fragment,t),x(Jt.$$.fragment,t),x(Kt.$$.fragment,t),x(Qt.$$.fragment,t),x(ea.$$.fragment,t),x(aa.$$.fragment,t),x(oa.$$.fragment,t),Tl=!0)},o(t){P(T.$$.fragment,t),P(bt.$$.fragment,t),P(yt.$$.fragment,t),P(we.$$.fragment,t),P(It.$$.fragment,t),P(Ct.$$.fragment,t),P(Rt.$$.fragment,t),P(jt.$$.fragment,t),P(Ne.$$.fragment,t),P(Zt.$$.fragment,t),P(Ft.$$.fragment,t),P(Wt.$$.fragment,t),P(Ot.$$.fragment,t),P(Jt.$$.fragment,t),P(Kt.$$.fragment,t),P(Qt.$$.fragment,t),P(ea.$$.fragment,t),P(aa.$$.fragment,t),P(oa.$$.fragment,t),Tl=!1},d(t){a(g),t&&a(D),t&&a(_),y(T),t&&a(en),t&&a(ha),t&&a(tn),t&&a(pa),t&&a(an),t&&a(da),t&&a(rn),t&&a(k),t&&a(on),t&&a(ua),t&&a(nn),t&&a(ca),t&&a(ln),t&&a(ma),t&&a(sn),t&&a(c),t&&a(fn),t&&a(Sa),t&&a(hn),t&&a(me),t&&a(pn),t&&a(Y),y(bt),t&&a(dn),t&&a(Hi),t&&a(un),t&&a(K),t&&a(cn),t&&a(ge),t&&a(mn),t&&a(Q),y(yt),t&&a(vn),t&&a(Fi),t&&a(gn),t&&a($t),t&&a(_n),t&&a(Ui),t&&a(En),t&&a(Oi),t&&a(wn),t&&a(Ee),t&&a(bn),y(we,t),t&&a(xn),t&&a(be),t&&a(Pn),t&&a(xe),t&&a(yn),t&&a(X),y(It),t&&a(An),t&&a(Yi),t&&a($n),t&&a(Gt),t&&a(Tn),t&&a(Qi),t&&a(Dn),t&&a(ye),t&&a(kn),t&&a(Ae),t&&a(Sn),t&&a(ee),y(Ct),t&&a(In),t&&a(er),t&&a(Nn),t&&a(tr),t&&a(Gn),t&&a(ar),t&&a(Cn),t&&a(ir),t&&a(qn),t&&a(Te),t&&a(Rn),t&&a(te),y(Rt),t&&a(Mn),t&&a(or),t&&a(Ln),t&&a(Lt),t&&a(jn),t&&a(lr),t&&a(zn),t&&a(ke),t&&a(Hn),t&&a(ae),y(jt),t&&a(Zn),t&&a(fr),t&&a(Bn),t&&a(Ht),t&&a(Fn),t&&a(pr),t&&a(Wn),t&&a(Ie),t&&a(Un),y(Ne,t),t&&a(On),t&&a(ie),y(Zt),t&&a(Vn),t&&a(ur),t&&a(Jn),t&&a(Ce),t&&a(Yn),t&&a(qe),t&&a(Kn),t&&a(re),y(Ft),t&&a(Qn),t&&a(vr),t&&a(Xn),t&&a(oe),y(Wt),t&&a(el),t&&a(Ut),t&&a(tl),t&&a(Le),t&&a(al),t&&a(ne),y(Ot),t&&a(il),t&&a(Vt),t&&a(rl),t&&a(ze),t&&a(ol),t&&a(le),y(Jt),t&&a(nl),t&&a(br),t&&a(ll),t&&a(se),t&&a(sl),t&&a(Ze),t&&a(fl),t&&a(fe),y(Kt),t&&a(hl),t&&a(Ar),t&&a(pl),t&&a(Fe),t&&a(dl),t&&a(he),y(Qt),t&&a(ul),t&&a(Xt),t&&a(cl),t&&a(Ue),t&&a(ml),t&&a(pe),y(ea),t&&a(vl),t&&a(kr),t&&a(gl),t&&a(Ve),t&&a(_l),t&&a(Je),t&&a(El),t&&a(de),y(aa),t&&a(wl),t&&a(Nr),t&&a(bl),t&&a(ra),t&&a(xl),t&&a(Ke),t&&a(Pl),t&&a(ue),y(oa),t&&a(yl),t&&a(qr),t&&a(Al),t&&a(la),t&&a($l),t&&a(Xe)}}}const $g={local:"controlled-generation",sections:[{local:"instruct-pix2pix",title:"Instruct Pix2Pix"},{local:"pix2pix-zero",title:"Pix2Pix Zero"},{local:"attend-and-excite",title:"Attend and Excite"},{local:"semantic-guidance-sega",title:"Semantic Guidance (SEGA)"},{local:"selfattention-guidance-sag",title:"Self-attention Guidance (SAG)"},{local:"depth2image",title:"Depth2Image"},{local:"multidiffusion-panorama",title:"MultiDiffusion Panorama"},{local:"finetuning-your-own-models",title:"Fine-tuning your own models"},{local:"dreambooth",title:"DreamBooth"},{local:"textual-inversion",title:"Textual Inversion"},{local:"controlnet",title:"ControlNet"},{local:"prompt-weighting",title:"Prompt Weighting"},{local:"custom-diffusion",title:"Custom Diffusion"},{local:"model-editing",title:"Model Editing"},{local:"diffedit",title:"DiffEdit"},{local:"t2iadapter",title:"T2I-Adapter"}],title:"Controlled generation"};function Tg(fa){return xg(()=>{new URLSearchParams(window.location.search).get("fw")}),[]}class Ig extends _g{constructor(g){super();Eg(this,g,Tg,Ag,wg,{})}}export{Ig as default,$g as metadata};