diffusers/pr_14217/en/_app/immutable/nodes/150.3KEdTavA.js

import"../chunks/DsnmJJEf.js";import{i as B,h as V,C as R,H as i,a as s,D as t,E as Q,s as F}from"../chunks/BtE7mKSK.js";import{p as C,o as X,s as e,f as z,a as D,b as Y,c as a,d as G,n as o,r as n}from"../chunks/jDjavuwI.js";const S='{"title":"DreamLite","local":"dreamlite","sections":[{"title":"Text-to-image (Base)","local":"text-to-image-base","sections":[],"depth":2},{"title":"Image editing (Base)","local":"image-editing-base","sections":[],"depth":2},{"title":"Text-to-image (Mobile)","local":"text-to-image-mobile","sections":[],"depth":2},{"title":"Image editing (Mobile)","local":"image-editing-mobile","sections":[],"depth":2},{"title":"Notes and limitations","local":"notes-and-limitations","sections":[],"depth":2},{"title":"DreamLitePipeline","local":"diffusers.DreamLitePipeline","sections":[],"depth":2},{"title":"DreamLiteMobilePipeline","local":"diffusers.DreamLiteMobilePipeline","sections":[],"depth":2},{"title":"DreamLitePipelineOutput","local":"diffusers.DreamLitePipelineOutput","sections":[],"depth":2}],"depth":1}';var A=G('<meta name="hf:doc:metadata"/>'),O=G(`<p></p> <!> <!> <p>DreamLite is a text-to-image and image-editing model from ByteDance. It pairs a custom 2D U-Net
(<code>DreamLiteUNetModel</code>) with the <code>Qwen3-VL</code> multimodal encoder as its prompt / image-instruction encoder,
and uses an <code>AutoencoderTiny</code> (TAESD-style) VAE for fast latent encode/decode.</p> <p>Two pipelines are exposed:</p> <table><thead><tr><th>Pipeline</th><th>Modes</th><th>CFG</th><th>Use case</th></tr></thead><tbody><tr><td><a href="/docs/diffusers/pr_14217/en/api/pipelines/dreamlite#diffusers.DreamLitePipeline">DreamLitePipeline</a></td><td>text-to-image <strong>and</strong> image-editing (auto-selected by whether <code>image</code> is <code>None</code>)</td><td>3-branch dual CFG (<code>guidance_scale</code> on text branch, <code>image_guidance_scale</code> on image branch, à la InstructPix2Pix)</td><td>Highest quality</td></tr><tr><td><a href="/docs/diffusers/pr_14217/en/api/pipelines/dreamlite#diffusers.DreamLiteMobilePipeline">DreamLiteMobilePipeline</a></td><td>text-to-image <strong>and</strong> image-editing (auto-selected by whether <code>image</code> is <code>None</code>)</td><td>None — distilled, single UNet forward per step</td><td>On-device / low-latency</td></tr></tbody></table> <p>Official checkpoints:</p> <ul><li>Base model: <a href="https://huggingface.co/carlofkl/DreamLite-base" rel="nofollow">carlofkl/DreamLite-base</a></li> <li>Distilled mobile model: <a href="https://huggingface.co/carlofkl/DreamLite-mobile" rel="nofollow">carlofkl/DreamLite-mobile</a></li></ul> <blockquote class="tip"><p>Both pipelines auto-detect text-to-image vs. image-editing mode from whether the <code>image</code> argument is
provided. There is no separate <code>Img2Img</code> class.</p></blockquote> <blockquote class="tip"><p>When loading an input image for editing, prefer <code>diffusers.utils.load_image(...)</code> over raw <code>PIL.Image.open(...)</code>. <code>load_image</code> enforces an RGB conversion and applies EXIF orientation, both of which the pipeline assumes.
A plain <code>Image.open</code> of an RGBA / palette / EXIF-rotated source will silently produce a different latent
conditioning and degrade output quality.</p></blockquote> <!> <!> <!> <p>Pass an <code>image</code> to enter edit mode. Both <code>guidance_scale</code> (text branch) and <code>image_guidance_scale</code> (image branch) are active here.</p> <!> <!> <p>The mobile pipeline is distilled and skips CFG entirely — a single UNet forward per step. It accepts the
same <code>prompt</code> / <code>height</code> / <code>width</code> / <code>num_inference_steps</code> arguments, but <strong>ignores</strong> <code>guidance_scale</code> and <code>image_guidance_scale</code> if passed (a warning is logged).</p> <!> <!> <!> <!> <ul><li>Both pipelines force <code>batch_size = 1</code> internally; <code>num_images_per_prompt</code> controls how many samples
are drawn from the same prompt rather than parallel batching.</li> <li>The prompt encoder is <code>Qwen3-VL</code>, which is a multimodal model. Loading the full pipeline therefore
requires sufficient GPU memory for both the U-Net and the Qwen3-VL text encoder (~4 GB + ~0.7 GB
in bf16 for the base release).</li> <li>The VAE is <code>AutoencoderTiny</code> and exposes <code>encoder_block_out_channels</code>; <code>vae_scale_factor</code> is derived
from it at pipeline init time.</li></ul> <!> <div class="docstring border-l-2 border-t-2 pl-4 pt-3.5 border-gray-100 rounded-tl-xl mb-6 mt-8"><!> <p>DreamLite pipeline for text-to-image and instruction-based image editing.</p> <p>The same pipeline supports both modes; the operating mode is auto-detected from the inputs:</p> <ul><li><code>image is None</code> -> text-to-image (single CFG on text).</li> <li><code>image is not None</code> -> image-to-image / instruction edit (dual CFG: text + image).</li></ul> <p>Components:
text_encoder ([<em>~transformers.Qwen3VLForConditionalGeneration</em>]):
Multimodal text/vision encoder used to produce conditioning embeddings.
tokenizer ([<em>~transformers.AutoTokenizer</em>]):
Tokenizer for text-only (generate) mode.
processor ([<em>~transformers.Qwen3VLProcessor</em>]):
Multimodal processor for edit mode (text + image template).
vae ([<em>~diffusers.AutoencoderTiny</em>]):
Mobile-friendly tiny VAE for latent encode/decode.
unet ([<em>~diffusers.DreamLiteUNetModel</em>]):
DreamLite UNet (GQA + qk_norm + depthwise-separable convs).
scheduler ([<em>~diffusers.FlowMatchEulerDiscreteScheduler</em>]):
Flow-matching Euler scheduler with dynamic shift.</p> <p>Note: <code>batch_size</code> is currently forced to <code>1</code>; <code>num_images_per_prompt</code> is supported.</p> <div class="docstring border-l-2 border-t-2 pl-4 pt-3.5 border-gray-100 rounded-tl-xl mb-6 mt-8"><!> <p>Run the DreamLite pipeline.</p></div></div> <!> <div class="docstring border-l-2 border-t-2 pl-4 pt-3.5 border-gray-100 rounded-tl-xl mb-6 mt-8"><!> <p>DreamLite <strong>Mobile</strong> pipeline: a distilled, classifier-free-guidance-free variant of <a href="/docs/diffusers/pr_14217/en/api/pipelines/dreamlite#diffusers.DreamLitePipeline">DreamLitePipeline</a> for fast few-step inference (default 4 steps).</p> <p>The operating mode is auto-detected from inputs (same as the base pipeline):</p> <ul><li><code>image is None</code> -> text-to-image.</li> <li><code>image is not None</code> -> image-to-image / instruction edit.</li></ul> <p>Because classifier-free guidance is <strong>distilled away</strong>, <code>guidance_scale</code> and <code>image_guidance_scale</code> are
accepted for API parity with <a href="/docs/diffusers/pr_14217/en/api/pipelines/dreamlite#diffusers.DreamLitePipeline">DreamLitePipeline</a> but are ignored in the denoising loop. <code>negative_prompt</code> is intentionally absent.</p> <p>Components (identical to the base pipeline):
text_encoder ([<em>~transformers.Qwen3VLForConditionalGeneration</em>]):
Multimodal text/vision encoder.
tokenizer ([<em>~transformers.AutoTokenizer</em>]):
Tokenizer for text-only (generate) mode.
processor ([<em>~transformers.Qwen3VLProcessor</em>]):
Multimodal processor for edit mode.
vae ([<em>~diffusers.AutoencoderTiny</em>]):
Mobile-friendly tiny VAE.
unet ([<em>~diffusers.DreamLiteUNetModel</em>]):
DreamLite UNet.
scheduler ([<em>~diffusers.FlowMatchEulerDiscreteScheduler</em>]):
Flow-matching Euler scheduler with dynamic shift.</p> <p>Note: <code>batch_size</code> is currently forced to <code>1</code>; <code>num_images_per_prompt</code> is supported.</p> <div class="docstring border-l-2 border-t-2 pl-4 pt-3.5 border-gray-100 rounded-tl-xl mb-6 mt-8"><!> <p>Run the distilled DreamLite Mobile pipeline.</p></div></div> <!> <div class="docstring border-l-2 border-t-2 pl-4 pt-3.5 border-gray-100 rounded-tl-xl mb-6 mt-8"><!> <p>Output class for DreamLite pipelines.</p></div> <!> <p></p>`,1);function ee(W,P){C(P,!1),X(()=>{new URLSearchParams(window.location.search).get("fw")}),B();var d=O();V("13fsxjz",Z=>{var x=A();F(x,"content",S),D(Z,x)});var c=e(z(d),2);R(c,{containerStyle:"float: right; margin-left: 10px; display: inline-flex; position: relative; z-index: 10;"});var m=e(c,2);i(m,{title:"DreamLite",local:"dreamlite",headingTag:"h1"});var u=e(m,16);i(u,{title:"Text-to-image (Base)",local:"text-to-image-base",headingTag:"h2"});var g=e(u,2);s(g,{code:"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",highlighted:`<span class="hljs-keyword">import</span> torch
<span class="hljs-keyword">from</span> diffusers <span class="hljs-keyword">import</span> DreamLitePipeline

pipe = DreamLitePipeline.from_pretrained(<span class="hljs-string">&quot;carlofkl/DreamLite-base&quot;</span>, revision=<span class="hljs-string">&quot;diffusers&quot;</span>, torch_dtype=torch.bfloat16)
pipe = pipe.to(<span class="hljs-string">&quot;cuda&quot;</span>)

image = pipe(
    prompt=<span class="hljs-string">&quot;a dog running on the grass&quot;</span>,
    negative_prompt=<span class="hljs-string">&quot;&quot;</span>,
    height=<span class="hljs-number">1024</span>,
    width=<span class="hljs-number">1024</span>,
    num_inference_steps=<span class="hljs-number">28</span>,
    generator=torch.Generator(<span class="hljs-string">&quot;cpu&quot;</span>).manual_seed(<span class="hljs-number">42</span>),
).images[<span class="hljs-number">0</span>]
image.save(<span class="hljs-string">&quot;dreamlite_t2i.png&quot;</span>)`,lang:"python",wrap:!1});var h=e(g,2);i(h,{title:"Image editing (Base)",local:"image-editing-base",headingTag:"h2"});var f=e(h,4);s(f,{code:"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",highlighted:`<span class="hljs-keyword">import</span> torch
<span class="hljs-keyword">from</span> diffusers <span class="hljs-keyword">import</span> DreamLitePipeline
<span class="hljs-keyword">from</span> diffusers.utils <span class="hljs-keyword">import</span> load_image

pipe = DreamLitePipeline.from_pretrained(<span class="hljs-string">&quot;carlofkl/DreamLite-base&quot;</span>, revision=<span class="hljs-string">&quot;diffusers&quot;</span>, torch_dtype=torch.bfloat16)
pipe = pipe.to(<span class="hljs-string">&quot;cuda&quot;</span>)

source = load_image(<span class="hljs-string">&quot;https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/diffusers/cat.png&quot;</span>)

image = pipe(
    prompt=<span class="hljs-string">&quot;turn the cat into a corgi&quot;</span>,
    image=source,
    height=<span class="hljs-number">1024</span>,
    width=<span class="hljs-number">1024</span>,
    num_inference_steps=<span class="hljs-number">28</span>,
    generator=torch.Generator(<span class="hljs-string">&quot;cpu&quot;</span>).manual_seed(<span class="hljs-number">42</span>),
).images[<span class="hljs-number">0</span>]
image.save(<span class="hljs-string">&quot;dreamlite_edit.png&quot;</span>)`,lang:"python",wrap:!1});var _=e(f,2);i(_,{title:"Text-to-image (Mobile)",local:"text-to-image-mobile",headingTag:"h2"});var y=e(_,4);s(y,{code:"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",highlighted:`<span class="hljs-keyword">import</span> torch
<span class="hljs-keyword">from</span> diffusers <span class="hljs-keyword">import</span> DreamLiteMobilePipeline

pipe = DreamLiteMobilePipeline.from_pretrained(<span class="hljs-string">&quot;carlofkl/DreamLite-mobile&quot;</span>, revision=<span class="hljs-string">&quot;diffusers&quot;</span>, torch_dtype=torch.bfloat16)
pipe = pipe.to(<span class="hljs-string">&quot;cuda&quot;</span>)

image = pipe(
    prompt=<span class="hljs-string">&quot;a dog running on the grass&quot;</span>,
    height=<span class="hljs-number">1024</span>,
    width=<span class="hljs-number">1024</span>,
    num_inference_steps=<span class="hljs-number">4</span>,
    generator=torch.Generator(<span class="hljs-string">&quot;cpu&quot;</span>).manual_seed(<span class="hljs-number">42</span>),
).images[<span class="hljs-number">0</span>]
image.save(<span class="hljs-string">&quot;dreamlite_mobile_t2i.png&quot;</span>)`,lang:"python",wrap:!1});var M=e(y,2);i(M,{title:"Image editing (Mobile)",local:"image-editing-mobile",headingTag:"h2"});var b=e(M,2);s(b,{code:"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",highlighted:`<span class="hljs-keyword">import</span> torch
<span class="hljs-keyword">from</span> diffusers <span class="hljs-keyword">import</span> DreamLiteMobilePipeline
<span class="hljs-keyword">from</span> diffusers.utils <span class="hljs-keyword">import</span> load_image

pipe = DreamLiteMobilePipeline.from_pretrained(<span class="hljs-string">&quot;carlofkl/DreamLite-mobile&quot;</span>, revision=<span class="hljs-string">&quot;diffusers&quot;</span>, torch_dtype=torch.bfloat16)
pipe = pipe.to(<span class="hljs-string">&quot;cuda&quot;</span>)

source = load_image(<span class="hljs-string">&quot;https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/diffusers/cat.png&quot;</span>)

image = pipe(
    prompt=<span class="hljs-string">&quot;turn the cat into a corgi&quot;</span>,
    image=source,
    height=<span class="hljs-number">1024</span>,
    width=<span class="hljs-number">1024</span>,
    num_inference_steps=<span class="hljs-number">4</span>,
    generator=torch.Generator(<span class="hljs-string">&quot;cpu&quot;</span>).manual_seed(<span class="hljs-number">42</span>),
).images[<span class="hljs-number">0</span>]
image.save(<span class="hljs-string">&quot;dreamlite_mobile_edit.png&quot;</span>)`,lang:"python",wrap:!1});var J=e(b,2);i(J,{title:"Notes and limitations",local:"notes-and-limitations",headingTag:"h2"});var w=e(J,4);i(w,{title:"DreamLitePipeline",local:"diffusers.DreamLitePipeline",headingTag:"h2"});var l=e(w,2),T=a(l);t(T,{name:"class diffusers.DreamLitePipeline",anchor:"diffusers.DreamLitePipeline",source:"https://github.com/huggingface/diffusers/blob/vr_14217/src/diffusers/pipelines/dreamlite/pipeline_dreamlite.py#L155",parameters:[{name:"text_encoder",val:": Qwen3VLForConditionalGeneration"},{name:"tokenizer",val:": AutoTokenizer"},{name:"processor",val:": Qwen3VLProcessor"},{name:"vae",val:": AutoencoderTiny"},{name:"unet",val:": DreamLiteUNetModel"},{name:"scheduler",val:": FlowMatchEulerDiscreteScheduler"}]});var U=e(T,12),N=a(U);t(N,{name:"__call__",anchor:"diffusers.DreamLitePipeline.__call__",source:"https://github.com/huggingface/diffusers/blob/vr_14217/src/diffusers/pipelines/dreamlite/pipeline_dreamlite.py#L388",parameters:[{name:"prompt",val:": typing.Optional[str] = None"},{name:"negative_prompt",val:": typing.Optional[str] = None"},{name:"image",val:": typing.Optional[PIL.Image.Image] = None"},{name:"height",val:": typing.Optional[int] = None"},{name:"width",val:": typing.Optional[int] = None"},{name:"guidance_scale",val:": float = 3.5"},{name:"image_guidance_scale",val:": float = 1.5"},{name:"num_inference_steps",val:": int = 30"},{name:"sigmas",val:": typing.Optional[typing.List[float]] = None"},{name:"num_images_per_prompt",val:": typing.Optional[int] = 1"},{name:"generator",val:": typing.Union[torch.Generator, typing.List[torch.Generator], NoneType] = None"},{name:"output_type",val:": typing.Optional[str] = 'pil'"},{name:"return_dict",val:": bool = True"},{name:"max_sequence_length",val:": int = 200"},{name:"text_pad_embedding",val:": typing.Optional[torch.Tensor] = None"}],parametersDescription:[{anchor:"diffusers.DreamLitePipeline.__call__.prompt",description:"<strong>prompt</strong> &#x2014; Text prompt.",name:"prompt"},{anchor:"diffusers.DreamLitePipeline.__call__.negative_prompt",description:"<strong>negative_prompt</strong> &#x2014; Negative text prompt (defaults to empty string).",name:"negative_prompt"},{anchor:"diffusers.DreamLitePipeline.__call__.image",description:`<strong>image</strong> &#x2014; Optional input image. If provided, the pipeline runs in <strong>edit / image-to-image</strong> mode
with dual classifier-free guidance; otherwise it runs in <strong>text-to-image</strong> mode.`,name:"image"},{anchor:"diffusers.DreamLitePipeline.__call__.height",description:`<strong>height</strong> &#x2014; Output resolution (height). Defaults to <code>default_sample_size * vae_scale_factor</code> (1024).
The same default applies in both T2I and I2I; pass an explicit value to override.`,name:"height"},{anchor:"diffusers.DreamLitePipeline.__call__.width",description:`<strong>width</strong> &#x2014; Output resolution (width). Defaults to <code>default_sample_size * vae_scale_factor</code> (1024).
The same default applies in both T2I and I2I; pass an explicit value to override.`,name:"width"},{anchor:"diffusers.DreamLitePipeline.__call__.guidance_scale",description:"<strong>guidance_scale</strong> &#x2014; CFG scale on the text branch (both modes).",name:"guidance_scale"},{anchor:"diffusers.DreamLitePipeline.__call__.image_guidance_scale",description:"<strong>image_guidance_scale</strong> &#x2014; Additional CFG scale on the image branch (edit mode only).",name:"image_guidance_scale"},{anchor:"diffusers.DreamLitePipeline.__call__.num_inference_steps",description:"<strong>num_inference_steps</strong> &#x2014; Number of denoising steps.",name:"num_inference_steps"},{anchor:"diffusers.DreamLitePipeline.__call__.sigmas",description:"<strong>sigmas</strong> &#x2014; Optional explicit FlowMatch sigmas; defaults to a uniform linspace.",name:"sigmas"},{anchor:"diffusers.DreamLitePipeline.__call__.num_images_per_prompt",description:"<strong>num_images_per_prompt</strong> &#x2014; Output images per prompt (note: <code>batch_size</code> is forced to 1).",name:"num_images_per_prompt"},{anchor:"diffusers.DreamLitePipeline.__call__.generator",description:"<strong>generator</strong> &#x2014; Random generator(s).",name:"generator"},{anchor:"diffusers.DreamLitePipeline.__call__.output_type",description:"<strong>output_type</strong> &#x2014; <code>&quot;pil&quot;</code>, <code>&quot;np&quot;</code>, <code>&quot;pt&quot;</code> or <code>&quot;latent&quot;</code>.",name:"output_type"},{anchor:"diffusers.DreamLitePipeline.__call__.return_dict",description:'<strong>return_dict</strong> &#x2014; If True, returns a <a href="/docs/diffusers/pr_14217/en/api/pipelines/dreamlite#diffusers.DreamLitePipelineOutput">DreamLitePipelineOutput</a>; else a tuple <code>(images,)</code>.',name:"return_dict"},{anchor:"diffusers.DreamLitePipeline.__call__.max_sequence_length",description:`<strong>max_sequence_length</strong> &#x2014; Maximum number of user-prompt tokens kept after dropping the chat-template
prefix. Only applies to <code>generate</code> mode (the <code>edit</code> mode uses the multimodal processor&#x2019;s native
padding).`,name:"max_sequence_length"},{anchor:"diffusers.DreamLitePipeline.__call__.text_pad_embedding",description:"<strong>text_pad_embedding</strong> &#x2014; Optional learned pad embedding for masked positions.",name:"text_pad_embedding"}],returnDescription:`<script context="module">export const metadata = 'undefined';<\/script>


<p><a
  href="/docs/diffusers/pr_14217/en/api/pipelines/dreamlite#diffusers.DreamLitePipelineOutput"
>DreamLitePipelineOutput</a> or <code>tuple</code>.</p>
`}),o(2),n(U),n(l);var v=e(l,2);i(v,{title:"DreamLiteMobilePipeline",local:"diffusers.DreamLiteMobilePipeline",headingTag:"h2"});var r=e(v,2),I=a(r);t(I,{name:"class diffusers.DreamLiteMobilePipeline",anchor:"diffusers.DreamLiteMobilePipeline",source:"https://github.com/huggingface/diffusers/blob/vr_14217/src/diffusers/pipelines/dreamlite/pipeline_dreamlite_mobile.py#L156",parameters:[{name:"text_encoder",val:": Qwen3VLForConditionalGeneration"},{name:"tokenizer",val:": AutoTokenizer"},{name:"processor",val:": Qwen3VLProcessor"},{name:"vae",val:": AutoencoderTiny"},{name:"unet",val:": DreamLiteUNetModel"},{name:"scheduler",val:": FlowMatchEulerDiscreteScheduler"}]});var L=e(I,14),k=a(L);t(k,{name:"__call__",anchor:"diffusers.DreamLiteMobilePipeline.__call__",source:"https://github.com/huggingface/diffusers/blob/vr_14217/src/diffusers/pipelines/dreamlite/pipeline_dreamlite_mobile.py#L384",parameters:[{name:"prompt",val:": typing.Union[str, typing.List[str]] = None"},{name:"image",val:": typing.Optional[PIL.Image.Image] = None"},{name:"height",val:": typing.Optional[int] = None"},{name:"width",val:": typing.Optional[int] = None"},{name:"num_inference_steps",val:": int = 4"},{name:"guidance_scale",val:": typing.Optional[float] = None"},{name:"image_guidance_scale",val:": typing.Optional[float] = None"},{name:"sigmas",val:": typing.Optional[typing.List[float]] = None"},{name:"num_images_per_prompt",val:": typing.Optional[int] = 1"},{name:"generator",val:": typing.Union[torch.Generator, typing.List[torch.Generator], NoneType] = None"},{name:"output_type",val:": typing.Optional[str] = 'pil'"},{name:"return_dict",val:": bool = True"},{name:"max_sequence_length",val:": int = 200"},{name:"text_pad_embedding",val:": typing.Optional[torch.Tensor] = None"}],parametersDescription:[{anchor:"diffusers.DreamLiteMobilePipeline.__call__.prompt",description:"<strong>prompt</strong> &#x2014; Text prompt.",name:"prompt"},{anchor:"diffusers.DreamLiteMobilePipeline.__call__.image",description:`<strong>image</strong> &#x2014; Optional input image. If provided, runs in <strong>edit / image-to-image</strong> mode;
otherwise runs in <strong>text-to-image</strong> mode.`,name:"image"},{anchor:"diffusers.DreamLiteMobilePipeline.__call__.height",description:"<strong>height</strong> &#x2014; Output resolution (height). Defaults to <code>default_sample_size * vae_scale_factor</code> (1024).",name:"height"},{anchor:"diffusers.DreamLiteMobilePipeline.__call__.width",description:"<strong>width</strong> &#x2014; Output resolution (width). Defaults to <code>default_sample_size * vae_scale_factor</code> (1024).",name:"width"},{anchor:"diffusers.DreamLiteMobilePipeline.__call__.num_inference_steps",description:"<strong>num_inference_steps</strong> &#x2014; Number of denoising steps. Defaults to <strong>4</strong> (distilled).",name:"num_inference_steps"},{anchor:"diffusers.DreamLiteMobilePipeline.__call__.guidance_scale",description:`<strong>guidance_scale</strong> &#x2014; Accepted for API parity with <a href="/docs/diffusers/pr_14217/en/api/pipelines/dreamlite#diffusers.DreamLitePipeline">DreamLitePipeline</a>; <strong>ignored</strong>
because CFG was distilled away.`,name:"guidance_scale"},{anchor:"diffusers.DreamLiteMobilePipeline.__call__.image_guidance_scale",description:`<strong>image_guidance_scale</strong> &#x2014; Accepted for API parity with <a href="/docs/diffusers/pr_14217/en/api/pipelines/dreamlite#diffusers.DreamLitePipeline">DreamLitePipeline</a>; <strong>ignored</strong>
because CFG was distilled away.`,name:"image_guidance_scale"},{anchor:"diffusers.DreamLiteMobilePipeline.__call__.sigmas",description:"<strong>sigmas</strong> &#x2014; Optional explicit FlowMatch sigmas; defaults to a uniform linspace.",name:"sigmas"},{anchor:"diffusers.DreamLiteMobilePipeline.__call__.num_images_per_prompt",description:"<strong>num_images_per_prompt</strong> &#x2014; Output images per prompt (note: <code>batch_size</code> is forced to 1).",name:"num_images_per_prompt"},{anchor:"diffusers.DreamLiteMobilePipeline.__call__.generator",description:"<strong>generator</strong> &#x2014; Random generator(s).",name:"generator"},{anchor:"diffusers.DreamLiteMobilePipeline.__call__.output_type",description:"<strong>output_type</strong> &#x2014; <code>&quot;pil&quot;</code>, <code>&quot;np&quot;</code>, <code>&quot;pt&quot;</code> or <code>&quot;latent&quot;</code>.",name:"output_type"},{anchor:"diffusers.DreamLiteMobilePipeline.__call__.return_dict",description:'<strong>return_dict</strong> &#x2014; If True, returns a <a href="/docs/diffusers/pr_14217/en/api/pipelines/dreamlite#diffusers.DreamLitePipelineOutput">DreamLitePipelineOutput</a>; else <code>(images,)</code>.',name:"return_dict"},{anchor:"diffusers.DreamLiteMobilePipeline.__call__.max_sequence_length",description:`<strong>max_sequence_length</strong> &#x2014; Maximum number of user-prompt tokens kept after dropping the chat-template
prefix. Only applies to <code>generate</code> mode (the <code>edit</code> mode uses the multimodal processor&#x2019;s native
padding).`,name:"max_sequence_length"},{anchor:"diffusers.DreamLiteMobilePipeline.__call__.text_pad_embedding",description:"<strong>text_pad_embedding</strong> &#x2014; Optional learned pad embedding for masked positions.",name:"text_pad_embedding"}],returnDescription:`<script context="module">export const metadata = 'undefined';<\/script>


<p><a
  href="/docs/diffusers/pr_14217/en/api/pipelines/dreamlite#diffusers.DreamLitePipelineOutput"
>DreamLitePipelineOutput</a> or <code>tuple</code>.</p>
`}),o(2),n(L),n(r);var j=e(r,2);i(j,{title:"DreamLitePipelineOutput",local:"diffusers.DreamLitePipelineOutput",headingTag:"h2"});var p=e(j,2),E=a(p);t(E,{name:"class diffusers.DreamLitePipelineOutput",anchor:"diffusers.DreamLitePipelineOutput",source:"https://github.com/huggingface/diffusers/blob/vr_14217/src/diffusers/pipelines/dreamlite/pipeline_output.py#L25",parameters:[{name:"images",val:": typing.Union[typing.List[PIL.Image.Image], numpy.ndarray]"}],parametersDescription:[{anchor:"diffusers.DreamLitePipelineOutput.images",description:`<strong>images</strong> (<code>List[PIL.Image.Image]</code> or <code>np.ndarray</code>) &#x2014;
List of denoised PIL images of length <code>batch_size</code> or NumPy array of shape <code>(batch_size, height, width, num_channels)</code>. PIL images or NumPy array present the denoised images of the diffusion pipeline.`,name:"images"}]}),o(2),n(p);var q=e(p,2);Q(q,{source:"https://github.com/huggingface/diffusers/blob/main/docs/source/en/api/pipelines/dreamlite.md"}),o(2),D(W,d),Y()}export{ee as component};