diffusers/main/en/_app/immutable/nodes/110.3dd5ed3a.js

import{s as N,o as F,n as J}from"../chunks/scheduler.8c3d61f6.js";import{S as K,i as Q,g as p,s as r,r as A,A as V,h,f as i,c as f,j as U,u as C,x as L,k as W,y as Y,a,v as k,d as z,t as E,w as q}from"../chunks/index.da70eac4.js";import{T as Z}from"../chunks/Tip.1d9b8c37.js";import{H as B,E as ee}from"../chunks/EditOnGithub.1e64e623.js";function te(w){let s,d='To learn how to use SDXL Turbo for various tasks, how to optimize performance, and other usage examples, take a look at the <a href="../../../using-diffusers/sdxl_turbo">SDXL Turbo</a> guide.',u,o,m='Check out the <a href="https://huggingface.co/stabilityai" rel="nofollow">Stability AI</a> Hub organization for the official base and refiner model checkpoints!';return{c(){s=p("p"),s.innerHTML=d,u=r(),o=p("p"),o.innerHTML=m},l(n){s=h(n,"P",{"data-svelte-h":!0}),L(s)!=="svelte-neqfre"&&(s.innerHTML=d),u=f(n),o=h(n,"P",{"data-svelte-h":!0}),L(o)!=="svelte-1n0z285"&&(o.innerHTML=m)},m(n,l){a(n,s,l),a(n,u,l),a(n,o,l)},p:J,d(n){n&&(i(s),i(u),i(o))}}}function ie(w){let s,d,u,o,m,n,l,j='Stable Diffusion XL (SDXL) Turbo was proposed in <a href="https://stability.ai/research/adversarial-diffusion-distillation" rel="nofollow">Adversarial Diffusion Distillation</a> by Axel Sauer, Dominik Lorenz, Andreas Blattmann, and Robin Rombach.',x,g,I="The abstract from the paper is:",D,b,R="<em>We introduce Adversarial Diffusion Distillation (ADD), a novel training approach that efficiently samples large-scale foundational image diffusion models in just 1–4 steps while maintaining high image quality. We use score distillation to leverage large-scale off-the-shelf image diffusion models as a teacher signal in combination with an adversarial loss to ensure high image fidelity even in the low-step regime of one or two sampling steps. Our analyses show that our model clearly outperforms existing few-step methods (GANs,Latent Consistency Models) in a single step and reaches the performance of state-of-the-art diffusion models (SDXL) in only four steps. ADD is the first method to unlock single-step, real-time image synthesis with foundation models.</em>",y,$,S,_,G='<li>SDXL Turbo uses the exact same architecture as <a href="./stable_diffusion_xl">SDXL</a>, which means it also has the same API. Please refer to the <a href="./stable_diffusion_xl">SDXL</a> API reference for more details.</li> <li>SDXL Turbo should disable guidance scale by setting <code>guidance_scale=0.0</code>.</li> <li>SDXL Turbo should use <code>timestep_spacing=&#39;trailing&#39;</code> for the scheduler and use between 1 and 4 steps.</li> <li>SDXL Turbo has been trained to generate images of size 512x512.</li> <li>SDXL Turbo is open-access, but not open-source meaning that one might have to buy a model license in order to use it for commercial applications. Make sure to read the <a href="https://huggingface.co/stabilityai/sdxl-turbo" rel="nofollow">official model card</a> to learn more.</li>',M,c,P,T,X,v,H;return m=new B({props:{title:"SDXL Turbo",local:"sdxl-turbo",headingTag:"h1"}}),$=new B({props:{title:"Tips",local:"tips",headingTag:"h2"}}),c=new Z({props:{$$slots:{default:[te]},$$scope:{ctx:w}}}),T=new ee({props:{source:"https://github.com/huggingface/diffusers/blob/main/docs/source/en/api/pipelines/stable_diffusion/sdxl_turbo.md"}}),{c(){s=p("meta"),d=r(),u=p("p"),o=r(),A(m.$$.fragment),n=r(),l=p("p"),l.innerHTML=j,x=r(),g=p("p"),g.textContent=I,D=r(),b=p("p"),b.innerHTML=R,y=r(),A($.$$.fragment),S=r(),_=p("ul"),_.innerHTML=G,M=r(),A(c.$$.fragment),P=r(),A(T.$$.fragment),X=r(),v=p("p"),this.h()},l(e){const t=V("svelte-u9bgzb",document.head);s=h(t,"META",{name:!0,content:!0}),t.forEach(i),d=f(e),u=h(e,"P",{}),U(u).forEach(i),o=f(e),C(m.$$.fragment,e),n=f(e),l=h(e,"P",{"data-svelte-h":!0}),L(l)!=="svelte-1lov3q2"&&(l.innerHTML=j),x=f(e),g=h(e,"P",{"data-svelte-h":!0}),L(g)!=="svelte-1cwsb16"&&(g.textContent=I),D=f(e),b=h(e,"P",{"data-svelte-h":!0}),L(b)!=="svelte-g4uhaw"&&(b.innerHTML=R),y=f(e),C($.$$.fragment,e),S=f(e),_=h(e,"UL",{"data-svelte-h":!0}),L(_)!=="svelte-1s9reyj"&&(_.innerHTML=G),M=f(e),C(c.$$.fragment,e),P=f(e),C(T.$$.fragment,e),X=f(e),v=h(e,"P",{}),U(v).forEach(i),this.h()},h(){W(s,"name","hf:doc:metadata"),W(s,"content",ae)},m(e,t){Y(document.head,s),a(e,d,t),a(e,u,t),a(e,o,t),k(m,e,t),a(e,n,t),a(e,l,t),a(e,x,t),a(e,g,t),a(e,D,t),a(e,b,t),a(e,y,t),k($,e,t),a(e,S,t),a(e,_,t),a(e,M,t),k(c,e,t),a(e,P,t),k(T,e,t),a(e,X,t),a(e,v,t),H=!0},p(e,[t]){const O={};t&2&&(O.$$scope={dirty:t,ctx:e}),c.$set(O)},i(e){H||(z(m.$$.fragment,e),z($.$$.fragment,e),z(c.$$.fragment,e),z(T.$$.fragment,e),H=!0)},o(e){E(m.$$.fragment,e),E($.$$.fragment,e),E(c.$$.fragment,e),E(T.$$.fragment,e),H=!1},d(e){e&&(i(d),i(u),i(o),i(n),i(l),i(x),i(g),i(D),i(b),i(y),i(S),i(_),i(M),i(P),i(X),i(v)),i(s),q(m,e),q($,e),q(c,e),q(T,e)}}}const ae='{"title":"SDXL Turbo","local":"sdxl-turbo","sections":[{"title":"Tips","local":"tips","sections":[],"depth":2}],"depth":1}';function se(w){return F(()=>{new URLSearchParams(window.location.search).get("fw")}),[]}class fe extends K{constructor(s){super(),Q(this,s,se,ie,N,{})}}export{fe as component};