Buckets:

hf-doc-build
/

doc-dev

hf-doc-build/doc-dev / diffusers /pr_14045 /en /_app /immutable /nodes /145.034a3ddf.js

HuggingFaceDocBuilder's picture

HuggingFaceDocBuilder

12 days ago

107 kB

	import{s as Xt,o as zt,n as C}from"../chunks/scheduler.53228c21.js";import{S as Qt,i as Vt,e as h,s as i,c as d,h as Rt,a as b,d as o,b as c,f as F,g as M,j,k as Q,l as T,m as a,n as u,t as y,o as J,p as w}from"../chunks/index.cac5d66a.js";import{D as Js}from"../chunks/Docstring.6d051794.js";import{C as Z}from"../chunks/CodeBlock.606cbaf4.js";import{H as g,E as Et}from"../chunks/MermaidChart.svelte_svelte_type_style_lang.af4accee.js";import{H as os,a as _}from"../chunks/HfOption.6b51ddef.js";function Ht(f){let l,p;return l=new Z({props:{code:"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",highlighted:`<span class="hljs-keyword">import</span> json
	<span class="hljs-keyword">import</span> torch
	<span class="hljs-keyword">from</span> diffusers <span class="hljs-keyword">import</span> Cosmos3OmniPipeline
	<span class="hljs-keyword">from</span> diffusers.schedulers.scheduling_unipc_multistep <span class="hljs-keyword">import</span> UniPCMultistepScheduler
	<span class="hljs-keyword">from</span> diffusers.utils <span class="hljs-keyword">import</span> export_to_video

	<span class="hljs-comment"># JSON-upsampled positive and negative prompts (see "Prompt upsampling" above).</span>
	json_prompt = json.load(<span class="hljs-built_in">open</span>(<span class="hljs-string">"assets/example_t2v_prompt.json"</span>))
	negative_prompt = json.load(<span class="hljs-built_in">open</span>(<span class="hljs-string">"assets/negative_prompt.json"</span>))

	pipe = Cosmos3OmniPipeline.from_pretrained(
	<span class="hljs-string">"nvidia/Cosmos3-Nano"</span>, torch_dtype=torch.bfloat16, device_map=<span class="hljs-string">"cuda"</span>
	)
	pipe.scheduler = UniPCMultistepScheduler.from_config(
	pipe.scheduler.config, flow_shift=<span class="hljs-number">10.0</span>, use_karras_sigmas=<span class="hljs-literal">False</span>
	)

	result = pipe(
	prompt=json.dumps(json_prompt),
	negative_prompt=json.dumps(negative_prompt),
	num_frames=<span class="hljs-number">189</span>,
	height=<span class="hljs-number">720</span>,
	width=<span class="hljs-number">1280</span>,
	num_inference_steps=<span class="hljs-number">35</span>,
	guidance_scale=<span class="hljs-number">6.0</span>,
	fps=<span class="hljs-number">24.0</span>,
	)
	<span class="hljs-comment"># macro_block_size=1 allows arbitrary frame sizes (Cosmos3 outputs are not always divisible by 16).</span>
	export_to_video(result.video, <span class="hljs-string">"cosmos3_t2v.mp4"</span>, fps=<span class="hljs-number">24</span>, macro_block_size=<span class="hljs-number">1</span>)`,lang:"python",wrap:!1}}),{c(){d(l.$$.fragment)},l(s){M(l.$$.fragment,s)},m(s,r){u(l,s,r),p=!0},p:C,i(s){p\|\|(y(l.$$.fragment,s),p=!0)},o(s){J(l.$$.fragment,s),p=!1},d(s){w(l,s)}}}function St(f){let l,p;return l=new Z({props:{code:"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",highlighted:`<span class="hljs-keyword">import</span> json
	<span class="hljs-keyword">import</span> torch
	<span class="hljs-keyword">from</span> diffusers <span class="hljs-keyword">import</span> Cosmos3OmniPipeline
	<span class="hljs-keyword">from</span> diffusers.schedulers.scheduling_unipc_multistep <span class="hljs-keyword">import</span> UniPCMultistepScheduler
	<span class="hljs-keyword">from</span> diffusers.utils <span class="hljs-keyword">import</span> export_to_video

	<span class="hljs-comment"># JSON-upsampled positive and negative prompts (see "Prompt upsampling" above).</span>
	json_prompt = json.load(<span class="hljs-built_in">open</span>(<span class="hljs-string">"assets/example_t2v_prompt.json"</span>))
	negative_prompt = json.load(<span class="hljs-built_in">open</span>(<span class="hljs-string">"assets/negative_prompt.json"</span>))

	pipe = Cosmos3OmniPipeline.from_pretrained(
	<span class="hljs-string">"nvidia/Cosmos3-Super"</span>, torch_dtype=torch.bfloat16, device_map=<span class="hljs-string">"cuda"</span>
	)
	pipe.scheduler = UniPCMultistepScheduler.from_config(
	pipe.scheduler.config, flow_shift=<span class="hljs-number">10.0</span>, use_karras_sigmas=<span class="hljs-literal">False</span>
	)

	result = pipe(
	prompt=json.dumps(json_prompt),
	negative_prompt=json.dumps(negative_prompt),
	num_frames=<span class="hljs-number">189</span>,
	height=<span class="hljs-number">720</span>,
	width=<span class="hljs-number">1280</span>,
	num_inference_steps=<span class="hljs-number">35</span>,
	guidance_scale=<span class="hljs-number">6.0</span>,
	fps=<span class="hljs-number">24.0</span>,
	)
	<span class="hljs-comment"># macro_block_size=1 allows arbitrary frame sizes (Cosmos3 outputs are not always divisible by 16).</span>
	export_to_video(result.video, <span class="hljs-string">"cosmos3_t2v.mp4"</span>, fps=<span class="hljs-number">24</span>, macro_block_size=<span class="hljs-number">1</span>)`,lang:"python",wrap:!1}}),{c(){d(l.$$.fragment)},l(s){M(l.$$.fragment,s)},m(s,r){u(l,s,r),p=!0},p:C,i(s){p\|\|(y(l.$$.fragment,s),p=!0)},o(s){J(l.$$.fragment,s),p=!1},d(s){w(l,s)}}}function At(f){let l,p,s,r;return l=new _({props:{id:"model",option:"Nano",$$slots:{default:[Ht]},$$scope:{ctx:f}}}),s=new _({props:{id:"model",option:"Super",$$slots:{default:[St]},$$scope:{ctx:f}}}),{c(){d(l.$$.fragment),p=i(),d(s.$$.fragment)},l(n){M(l.$$.fragment,n),p=c(n),M(s.$$.fragment,n)},m(n,m){u(l,n,m),a(n,p,m),u(s,n,m),r=!0},p(n,m){const U={};m&2&&(U.$$scope={dirty:m,ctx:n}),l.$set(U);const v={};m&2&&(v.$$scope={dirty:m,ctx:n}),s.$set(v)},i(n){r\|\|(y(l.$$.fragment,n),y(s.$$.fragment,n),r=!0)},o(n){J(l.$$.fragment,n),J(s.$$.fragment,n),r=!1},d(n){n&&o(p),w(l,n),w(s,n)}}}function xt(f){let l,p;return l=new Z({props:{code:"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",highlighted:`<span class="hljs-keyword">import</span> json
	<span class="hljs-keyword">import</span> torch
	<span class="hljs-keyword">from</span> diffusers <span class="hljs-keyword">import</span> Cosmos3OmniPipeline

	<span class="hljs-comment"># JSON-upsampled prompt (see "Prompt upsampling" above).</span>
	json_prompt = json.load(<span class="hljs-built_in">open</span>(<span class="hljs-string">"assets/example_t2i_prompt.json"</span>))

	pipe = Cosmos3OmniPipeline.from_pretrained(
	<span class="hljs-string">"nvidia/Cosmos3-Nano"</span>, torch_dtype=torch.bfloat16, device_map=<span class="hljs-string">"cuda"</span>
	)

	result = pipe(prompt=json.dumps(json_prompt), num_frames=<span class="hljs-number">1</span>, height=<span class="hljs-number">720</span>, width=<span class="hljs-number">1280</span>)
	result.video[<span class="hljs-number">0</span>].save(<span class="hljs-string">"cosmos3_t2i.jpg"</span>, <span class="hljs-built_in">format</span>=<span class="hljs-string">"JPEG"</span>, quality=<span class="hljs-number">85</span>)`,lang:"python",wrap:!1}}),{c(){d(l.$$.fragment)},l(s){M(l.$$.fragment,s)},m(s,r){u(l,s,r),p=!0},p:C,i(s){p\|\|(y(l.$$.fragment,s),p=!0)},o(s){J(l.$$.fragment,s),p=!1},d(s){w(l,s)}}}function qt(f){let l,p;return l=new Z({props:{code:"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",highlighted:`<span class="hljs-keyword">import</span> json
	<span class="hljs-keyword">import</span> torch
	<span class="hljs-keyword">from</span> diffusers <span class="hljs-keyword">import</span> Cosmos3OmniPipeline

	<span class="hljs-comment"># JSON-upsampled prompt (see "Prompt upsampling" above).</span>
	json_prompt = json.load(<span class="hljs-built_in">open</span>(<span class="hljs-string">"assets/example_t2i_prompt.json"</span>))

	pipe = Cosmos3OmniPipeline.from_pretrained(
	<span class="hljs-string">"nvidia/Cosmos3-Super"</span>, torch_dtype=torch.bfloat16, device_map=<span class="hljs-string">"cuda"</span>
	)

	result = pipe(prompt=json.dumps(json_prompt), num_frames=<span class="hljs-number">1</span>, height=<span class="hljs-number">720</span>, width=<span class="hljs-number">1280</span>)
	result.video[<span class="hljs-number">0</span>].save(<span class="hljs-string">"cosmos3_t2i.jpg"</span>, <span class="hljs-built_in">format</span>=<span class="hljs-string">"JPEG"</span>, quality=<span class="hljs-number">85</span>)`,lang:"python",wrap:!1}}),{c(){d(l.$$.fragment)},l(s){M(l.$$.fragment,s)},m(s,r){u(l,s,r),p=!0},p:C,i(s){p\|\|(y(l.$$.fragment,s),p=!0)},o(s){J(l.$$.fragment,s),p=!1},d(s){w(l,s)}}}function Yt(f){let l,p,s,r;return l=new _({props:{id:"model",option:"Nano",$$slots:{default:[xt]},$$scope:{ctx:f}}}),s=new _({props:{id:"model",option:"Super",$$slots:{default:[qt]},$$scope:{ctx:f}}}),{c(){d(l.$$.fragment),p=i(),d(s.$$.fragment)},l(n){M(l.$$.fragment,n),p=c(n),M(s.$$.fragment,n)},m(n,m){u(l,n,m),a(n,p,m),u(s,n,m),r=!0},p(n,m){const U={};m&2&&(U.$$scope={dirty:m,ctx:n}),l.$set(U);const v={};m&2&&(v.$$scope={dirty:m,ctx:n}),s.$set(v)},i(n){r\|\|(y(l.$$.fragment,n),y(s.$$.fragment,n),r=!0)},o(n){J(l.$$.fragment,n),J(s.$$.fragment,n),r=!1},d(n){n&&o(p),w(l,n),w(s,n)}}}function Ft(f){let l,p;return l=new Z({props:{code:"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",highlighted:`<span class="hljs-keyword">import</span> json
	<span class="hljs-keyword">import</span> torch
	<span class="hljs-keyword">from</span> diffusers <span class="hljs-keyword">import</span> Cosmos3OmniPipeline
	<span class="hljs-keyword">from</span> diffusers.utils <span class="hljs-keyword">import</span> export_to_video, load_image

	<span class="hljs-comment"># JSON-upsampled positive and negative prompts (see "Prompt upsampling" above).</span>
	json_prompt = json.load(<span class="hljs-built_in">open</span>(<span class="hljs-string">"assets/example_i2v_prompt.json"</span>))
	negative_prompt = json.load(<span class="hljs-built_in">open</span>(<span class="hljs-string">"assets/negative_prompt_i2v.json"</span>))

	pipe = Cosmos3OmniPipeline.from_pretrained(
	<span class="hljs-string">"nvidia/Cosmos3-Nano"</span>, torch_dtype=torch.bfloat16, device_map=<span class="hljs-string">"cuda"</span>
	)

	image = load_image(
	<span class="hljs-string">"https://github.com/nvidia-cosmos/cosmos-dependencies/releases/download/assets/robot_153.jpg"</span>
	)

	result = pipe(
	prompt=json.dumps(json_prompt),
	negative_prompt=json.dumps(negative_prompt),
	image=image,
	num_frames=<span class="hljs-number">189</span>,
	height=<span class="hljs-number">720</span>,
	width=<span class="hljs-number">1280</span>,
	fps=<span class="hljs-number">24.0</span>,
	)
	<span class="hljs-comment"># macro_block_size=1 allows arbitrary frame sizes (Cosmos3 outputs are not always divisible by 16).</span>
	export_to_video(result.video, <span class="hljs-string">"cosmos3_i2v.mp4"</span>, fps=<span class="hljs-number">24</span>, macro_block_size=<span class="hljs-number">1</span>)`,lang:"python",wrap:!1}}),{c(){d(l.$$.fragment)},l(s){M(l.$$.fragment,s)},m(s,r){u(l,s,r),p=!0},p:C,i(s){p\|\|(y(l.$$.fragment,s),p=!0)},o(s){J(l.$$.fragment,s),p=!1},d(s){w(l,s)}}}function Lt(f){let l,p;return l=new Z({props:{code:"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",highlighted:`<span class="hljs-keyword">import</span> json
	<span class="hljs-keyword">import</span> torch
	<span class="hljs-keyword">from</span> diffusers <span class="hljs-keyword">import</span> Cosmos3OmniPipeline
	<span class="hljs-keyword">from</span> diffusers.utils <span class="hljs-keyword">import</span> export_to_video, load_image

	<span class="hljs-comment"># JSON-upsampled positive and negative prompts (see "Prompt upsampling" above).</span>
	json_prompt = json.load(<span class="hljs-built_in">open</span>(<span class="hljs-string">"assets/example_i2v_prompt.json"</span>))
	negative_prompt = json.load(<span class="hljs-built_in">open</span>(<span class="hljs-string">"assets/negative_prompt_i2v.json"</span>))

	pipe = Cosmos3OmniPipeline.from_pretrained(
	<span class="hljs-string">"nvidia/Cosmos3-Super"</span>, torch_dtype=torch.bfloat16, device_map=<span class="hljs-string">"cuda"</span>
	)

	image = load_image(
	<span class="hljs-string">"https://github.com/nvidia-cosmos/cosmos-dependencies/releases/download/assets/robot_153.jpg"</span>
	)

	result = pipe(
	prompt=json.dumps(json_prompt),
	negative_prompt=json.dumps(negative_prompt),
	image=image,
	num_frames=<span class="hljs-number">189</span>,
	height=<span class="hljs-number">720</span>,
	width=<span class="hljs-number">1280</span>,
	fps=<span class="hljs-number">24.0</span>,
	)
	<span class="hljs-comment"># macro_block_size=1 allows arbitrary frame sizes (Cosmos3 outputs are not always divisible by 16).</span>
	export_to_video(result.video, <span class="hljs-string">"cosmos3_i2v.mp4"</span>, fps=<span class="hljs-number">24</span>, macro_block_size=<span class="hljs-number">1</span>)`,lang:"python",wrap:!1}}),{c(){d(l.$$.fragment)},l(s){M(l.$$.fragment,s)},m(s,r){u(l,s,r),p=!0},p:C,i(s){p\|\|(y(l.$$.fragment,s),p=!0)},o(s){J(l.$$.fragment,s),p=!1},d(s){w(l,s)}}}function Pt(f){let l,p,s,r;return l=new _({props:{id:"model",option:"Nano",$$slots:{default:[Ft]},$$scope:{ctx:f}}}),s=new _({props:{id:"model",option:"Super",$$slots:{default:[Lt]},$$scope:{ctx:f}}}),{c(){d(l.$$.fragment),p=i(),d(s.$$.fragment)},l(n){M(l.$$.fragment,n),p=c(n),M(s.$$.fragment,n)},m(n,m){u(l,n,m),a(n,p,m),u(s,n,m),r=!0},p(n,m){const U={};m&2&&(U.$$scope={dirty:m,ctx:n}),l.$set(U);const v={};m&2&&(v.$$scope={dirty:m,ctx:n}),s.$set(v)},i(n){r\|\|(y(l.$$.fragment,n),y(s.$$.fragment,n),r=!0)},o(n){J(l.$$.fragment,n),J(s.$$.fragment,n),r=!1},d(n){n&&o(p),w(l,n),w(s,n)}}}function Dt(f){let l,p;return l=new Z({props:{code:"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",highlighted:`<span class="hljs-keyword">import</span> json
	<span class="hljs-keyword">import</span> torch
	<span class="hljs-keyword">from</span> diffusers <span class="hljs-keyword">import</span> Cosmos3OmniPipeline
	<span class="hljs-keyword">from</span> diffusers.schedulers.scheduling_unipc_multistep <span class="hljs-keyword">import</span> UniPCMultistepScheduler
	<span class="hljs-keyword">from</span> diffusers.utils <span class="hljs-keyword">import</span> export_to_video, load_video

	<span class="hljs-comment"># JSON-upsampled positive and negative prompts (see "Prompt upsampling" above).</span>
	json_prompt = json.load(<span class="hljs-built_in">open</span>(<span class="hljs-string">"assets/example_v2v_prompt.json"</span>))
	negative_prompt = json.load(<span class="hljs-built_in">open</span>(<span class="hljs-string">"assets/negative_prompt_i2v.json"</span>))

	pipe = Cosmos3OmniPipeline.from_pretrained(
	<span class="hljs-string">"nvidia/Cosmos3-Nano"</span>, torch_dtype=torch.bfloat16, device_map=<span class="hljs-string">"cuda"</span>
	)
	pipe.scheduler = UniPCMultistepScheduler.from_config(
	pipe.scheduler.config, flow_shift=<span class="hljs-number">10.0</span>, use_karras_sigmas=<span class="hljs-literal">False</span>
	)

	video = load_video(
	<span class="hljs-string">"https://github.com/nvidia-cosmos/cosmos-dependencies/raw/refs/heads/assets/cosmos3/inputs/vision/robot_pouring.mp4"</span>
	)

	result = pipe(
	prompt=json.dumps(json_prompt),
	negative_prompt=json.dumps(negative_prompt),
	video=video,
	condition_frame_indexes_vision=[<span class="hljs-number">0</span>, <span class="hljs-number">1</span>],
	condition_video_keep=<span class="hljs-string">"first"</span>,
	num_frames=<span class="hljs-number">189</span>,
	height=<span class="hljs-number">720</span>,
	width=<span class="hljs-number">1280</span>,
	num_inference_steps=<span class="hljs-number">35</span>,
	guidance_scale=<span class="hljs-number">6.0</span>,
	fps=<span class="hljs-number">24.0</span>,
	)
	<span class="hljs-comment"># macro_block_size=1 allows arbitrary frame sizes (Cosmos3 outputs are not always divisible by 16).</span>
	export_to_video(result.video, <span class="hljs-string">"cosmos3_v2v.mp4"</span>, fps=<span class="hljs-number">24</span>, macro_block_size=<span class="hljs-number">1</span>)`,lang:"python",wrap:!1}}),{c(){d(l.$$.fragment)},l(s){M(l.$$.fragment,s)},m(s,r){u(l,s,r),p=!0},p:C,i(s){p\|\|(y(l.$$.fragment,s),p=!0)},o(s){J(l.$$.fragment,s),p=!1},d(s){w(l,s)}}}function Ot(f){let l,p;return l=new Z({props:{code:"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",highlighted:`<span class="hljs-keyword">import</span> json
	<span class="hljs-keyword">import</span> torch
	<span class="hljs-keyword">from</span> diffusers <span class="hljs-keyword">import</span> Cosmos3OmniPipeline
	<span class="hljs-keyword">from</span> diffusers.schedulers.scheduling_unipc_multistep <span class="hljs-keyword">import</span> UniPCMultistepScheduler
	<span class="hljs-keyword">from</span> diffusers.utils <span class="hljs-keyword">import</span> export_to_video, load_video

	<span class="hljs-comment"># JSON-upsampled positive and negative prompts (see "Prompt upsampling" above).</span>
	json_prompt = json.load(<span class="hljs-built_in">open</span>(<span class="hljs-string">"assets/example_v2v_prompt.json"</span>))
	negative_prompt = json.load(<span class="hljs-built_in">open</span>(<span class="hljs-string">"assets/negative_prompt_i2v.json"</span>))

	pipe = Cosmos3OmniPipeline.from_pretrained(
	<span class="hljs-string">"nvidia/Cosmos3-Super"</span>, torch_dtype=torch.bfloat16, device_map=<span class="hljs-string">"cuda"</span>
	)
	pipe.scheduler = UniPCMultistepScheduler.from_config(
	pipe.scheduler.config, flow_shift=<span class="hljs-number">10.0</span>, use_karras_sigmas=<span class="hljs-literal">False</span>
	)

	video = load_video(
	<span class="hljs-string">"https://github.com/nvidia-cosmos/cosmos-dependencies/raw/refs/heads/assets/cosmos3/inputs/vision/robot_pouring.mp4"</span>
	)

	result = pipe(
	prompt=json.dumps(json_prompt),
	negative_prompt=json.dumps(negative_prompt),
	video=video,
	condition_frame_indexes_vision=[<span class="hljs-number">0</span>, <span class="hljs-number">1</span>],
	condition_video_keep=<span class="hljs-string">"first"</span>,
	num_frames=<span class="hljs-number">189</span>,
	height=<span class="hljs-number">720</span>,
	width=<span class="hljs-number">1280</span>,
	num_inference_steps=<span class="hljs-number">35</span>,
	guidance_scale=<span class="hljs-number">6.0</span>,
	fps=<span class="hljs-number">24.0</span>,
	)
	<span class="hljs-comment"># macro_block_size=1 allows arbitrary frame sizes (Cosmos3 outputs are not always divisible by 16).</span>
	export_to_video(result.video, <span class="hljs-string">"cosmos3_v2v.mp4"</span>, fps=<span class="hljs-number">24</span>, macro_block_size=<span class="hljs-number">1</span>)`,lang:"python",wrap:!1}}),{c(){d(l.$$.fragment)},l(s){M(l.$$.fragment,s)},m(s,r){u(l,s,r),p=!0},p:C,i(s){p\|\|(y(l.$$.fragment,s),p=!0)},o(s){J(l.$$.fragment,s),p=!1},d(s){w(l,s)}}}function Kt(f){let l,p,s,r;return l=new _({props:{id:"model",option:"Nano",$$slots:{default:[Dt]},$$scope:{ctx:f}}}),s=new _({props:{id:"model",option:"Super",$$slots:{default:[Ot]},$$scope:{ctx:f}}}),{c(){d(l.$$.fragment),p=i(),d(s.$$.fragment)},l(n){M(l.$$.fragment,n),p=c(n),M(s.$$.fragment,n)},m(n,m){u(l,n,m),a(n,p,m),u(s,n,m),r=!0},p(n,m){const U={};m&2&&(U.$$scope={dirty:m,ctx:n}),l.$set(U);const v={};m&2&&(v.$$scope={dirty:m,ctx:n}),s.$set(v)},i(n){r\|\|(y(l.$$.fragment,n),y(s.$$.fragment,n),r=!0)},o(n){J(l.$$.fragment,n),J(s.$$.fragment,n),r=!1},d(n){n&&o(p),w(l,n),w(s,n)}}}function en(f){let l,p;return l=new Z({props:{code:"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",highlighted:`<span class="hljs-keyword">import</span> json
	<span class="hljs-keyword">import</span> torch
	<span class="hljs-keyword">from</span> diffusers <span class="hljs-keyword">import</span> Cosmos3OmniPipeline
	<span class="hljs-keyword">from</span> diffusers.schedulers.scheduling_unipc_multistep <span class="hljs-keyword">import</span> UniPCMultistepScheduler
	<span class="hljs-keyword">from</span> diffusers.utils <span class="hljs-keyword">import</span> encode_video, load_video

	<span class="hljs-comment"># JSON-upsampled positive and negative prompts (see "Prompt upsampling" above).</span>
	json_prompt = json.load(<span class="hljs-built_in">open</span>(<span class="hljs-string">"assets/example_v2v_prompt.json"</span>))
	negative_prompt = json.load(<span class="hljs-built_in">open</span>(<span class="hljs-string">"assets/negative_prompt_i2v.json"</span>))

	pipe = Cosmos3OmniPipeline.from_pretrained(
	<span class="hljs-string">"nvidia/Cosmos3-Nano"</span>, torch_dtype=torch.bfloat16, device_map=<span class="hljs-string">"cuda"</span>
	)
	pipe.scheduler = UniPCMultistepScheduler.from_config(
	pipe.scheduler.config, flow_shift=<span class="hljs-number">10.0</span>, use_karras_sigmas=<span class="hljs-literal">False</span>
	)

	video = load_video(
	<span class="hljs-string">"https://github.com/nvidia-cosmos/cosmos-dependencies/raw/refs/heads/assets/cosmos3/inputs/vision/robot_pouring.mp4"</span>
	)

	result = pipe(
	prompt=json.dumps(json_prompt),
	negative_prompt=json.dumps(negative_prompt),
	video=video,
	condition_frame_indexes_vision=[<span class="hljs-number">0</span>, <span class="hljs-number">1</span>],
	condition_video_keep=<span class="hljs-string">"first"</span>,
	num_frames=<span class="hljs-number">189</span>,
	height=<span class="hljs-number">720</span>,
	width=<span class="hljs-number">1280</span>,
	fps=<span class="hljs-number">24.0</span>,
	enable_sound=<span class="hljs-literal">True</span>,
	)

	encode_video(
	result.video,
	fps=<span class="hljs-number">24</span>,
	audio=result.sound,
	audio_sample_rate=pipe.sound_tokenizer.config.sampling_rate,
	output_path=<span class="hljs-string">"cosmos3_v2v_with_sound.mp4"</span>,
	)`,lang:"python",wrap:!1}}),{c(){d(l.$$.fragment)},l(s){M(l.$$.fragment,s)},m(s,r){u(l,s,r),p=!0},p:C,i(s){p\|\|(y(l.$$.fragment,s),p=!0)},o(s){J(l.$$.fragment,s),p=!1},d(s){w(l,s)}}}function sn(f){let l,p;return l=new Z({props:{code:"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",highlighted:`<span class="hljs-keyword">import</span> json
	<span class="hljs-keyword">import</span> torch
	<span class="hljs-keyword">from</span> diffusers <span class="hljs-keyword">import</span> Cosmos3OmniPipeline
	<span class="hljs-keyword">from</span> diffusers.schedulers.scheduling_unipc_multistep <span class="hljs-keyword">import</span> UniPCMultistepScheduler
	<span class="hljs-keyword">from</span> diffusers.utils <span class="hljs-keyword">import</span> encode_video, load_video

	<span class="hljs-comment"># JSON-upsampled positive and negative prompts (see "Prompt upsampling" above).</span>
	json_prompt = json.load(<span class="hljs-built_in">open</span>(<span class="hljs-string">"assets/example_v2v_prompt.json"</span>))
	negative_prompt = json.load(<span class="hljs-built_in">open</span>(<span class="hljs-string">"assets/negative_prompt_i2v.json"</span>))

	pipe = Cosmos3OmniPipeline.from_pretrained(
	<span class="hljs-string">"nvidia/Cosmos3-Super"</span>, torch_dtype=torch.bfloat16, device_map=<span class="hljs-string">"cuda"</span>
	)
	pipe.scheduler = UniPCMultistepScheduler.from_config(
	pipe.scheduler.config, flow_shift=<span class="hljs-number">10.0</span>, use_karras_sigmas=<span class="hljs-literal">False</span>
	)

	video = load_video(
	<span class="hljs-string">"https://github.com/nvidia-cosmos/cosmos-dependencies/raw/refs/heads/assets/cosmos3/inputs/vision/robot_pouring.mp4"</span>
	)

	result = pipe(
	prompt=json.dumps(json_prompt),
	negative_prompt=json.dumps(negative_prompt),
	video=video,
	condition_frame_indexes_vision=[<span class="hljs-number">0</span>, <span class="hljs-number">1</span>],
	condition_video_keep=<span class="hljs-string">"first"</span>,
	num_frames=<span class="hljs-number">189</span>,
	height=<span class="hljs-number">720</span>,
	width=<span class="hljs-number">1280</span>,
	fps=<span class="hljs-number">24.0</span>,
	enable_sound=<span class="hljs-literal">True</span>,
	)

	encode_video(
	result.video,
	fps=<span class="hljs-number">24</span>,
	audio=result.sound,
	audio_sample_rate=pipe.sound_tokenizer.config.sampling_rate,
	output_path=<span class="hljs-string">"cosmos3_v2v_with_sound.mp4"</span>,
	)`,lang:"python",wrap:!1}}),{c(){d(l.$$.fragment)},l(s){M(l.$$.fragment,s)},m(s,r){u(l,s,r),p=!0},p:C,i(s){p\|\|(y(l.$$.fragment,s),p=!0)},o(s){J(l.$$.fragment,s),p=!1},d(s){w(l,s)}}}function ln(f){let l,p,s,r;return l=new _({props:{id:"model",option:"Nano",$$slots:{default:[en]},$$scope:{ctx:f}}}),s=new _({props:{id:"model",option:"Super",$$slots:{default:[sn]},$$scope:{ctx:f}}}),{c(){d(l.$$.fragment),p=i(),d(s.$$.fragment)},l(n){M(l.$$.fragment,n),p=c(n),M(s.$$.fragment,n)},m(n,m){u(l,n,m),a(n,p,m),u(s,n,m),r=!0},p(n,m){const U={};m&2&&(U.$$scope={dirty:m,ctx:n}),l.$set(U);const v={};m&2&&(v.$$scope={dirty:m,ctx:n}),s.$set(v)},i(n){r\|\|(y(l.$$.fragment,n),y(s.$$.fragment,n),r=!0)},o(n){J(l.$$.fragment,n),J(s.$$.fragment,n),r=!1},d(n){n&&o(p),w(l,n),w(s,n)}}}function tn(f){let l,p;return l=new Z({props:{code:"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",highlighted:`<span class="hljs-keyword">import</span> json
	<span class="hljs-keyword">import</span> torch
	<span class="hljs-keyword">from</span> diffusers <span class="hljs-keyword">import</span> Cosmos3OmniPipeline
	<span class="hljs-keyword">from</span> diffusers.utils <span class="hljs-keyword">import</span> encode_video

	<span class="hljs-comment"># JSON-upsampled positive and negative prompts (see "Prompt upsampling" above).</span>
	json_prompt = json.load(<span class="hljs-built_in">open</span>(<span class="hljs-string">"assets/example_t2v_sound_prompt.json"</span>))
	negative_prompt = json.load(<span class="hljs-built_in">open</span>(<span class="hljs-string">"assets/negative_prompt.json"</span>))

	pipe = Cosmos3OmniPipeline.from_pretrained(
	<span class="hljs-string">"nvidia/Cosmos3-Nano"</span>, torch_dtype=torch.bfloat16, device_map=<span class="hljs-string">"cuda"</span>
	)

	result = pipe(
	prompt=json.dumps(json_prompt),
	negative_prompt=json.dumps(negative_prompt),
	num_frames=<span class="hljs-number">189</span>,
	height=<span class="hljs-number">720</span>,
	width=<span class="hljs-number">1280</span>,
	fps=<span class="hljs-number">24.0</span>,
	enable_sound=<span class="hljs-literal">True</span>,
	)

	encode_video(
	result.video,
	fps=<span class="hljs-number">24</span>,
	audio=result.sound,
	audio_sample_rate=pipe.sound_tokenizer.config.sampling_rate,
	output_path=<span class="hljs-string">"cosmos3_with_sound.mp4"</span>,
	)`,lang:"python",wrap:!1}}),{c(){d(l.$$.fragment)},l(s){M(l.$$.fragment,s)},m(s,r){u(l,s,r),p=!0},p:C,i(s){p\|\|(y(l.$$.fragment,s),p=!0)},o(s){J(l.$$.fragment,s),p=!1},d(s){w(l,s)}}}function nn(f){let l,p;return l=new Z({props:{code:"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",highlighted:`<span class="hljs-keyword">import</span> json
	<span class="hljs-keyword">import</span> torch
	<span class="hljs-keyword">from</span> diffusers <span class="hljs-keyword">import</span> Cosmos3OmniPipeline
	<span class="hljs-keyword">from</span> diffusers.utils <span class="hljs-keyword">import</span> encode_video

	<span class="hljs-comment"># JSON-upsampled positive and negative prompts (see "Prompt upsampling" above).</span>
	json_prompt = json.load(<span class="hljs-built_in">open</span>(<span class="hljs-string">"assets/example_t2v_sound_prompt.json"</span>))
	negative_prompt = json.load(<span class="hljs-built_in">open</span>(<span class="hljs-string">"assets/negative_prompt.json"</span>))

	pipe = Cosmos3OmniPipeline.from_pretrained(
	<span class="hljs-string">"nvidia/Cosmos3-Super"</span>, torch_dtype=torch.bfloat16, device_map=<span class="hljs-string">"cuda"</span>
	)

	result = pipe(
	prompt=json.dumps(json_prompt),
	negative_prompt=json.dumps(negative_prompt),
	num_frames=<span class="hljs-number">189</span>,
	height=<span class="hljs-number">720</span>,
	width=<span class="hljs-number">1280</span>,
	fps=<span class="hljs-number">24.0</span>,
	enable_sound=<span class="hljs-literal">True</span>,
	)

	encode_video(
	result.video,
	fps=<span class="hljs-number">24</span>,
	audio=result.sound,
	audio_sample_rate=pipe.sound_tokenizer.config.sampling_rate,
	output_path=<span class="hljs-string">"cosmos3_with_sound.mp4"</span>,
	)`,lang:"python",wrap:!1}}),{c(){d(l.$$.fragment)},l(s){M(l.$$.fragment,s)},m(s,r){u(l,s,r),p=!0},p:C,i(s){p\|\|(y(l.$$.fragment,s),p=!0)},o(s){J(l.$$.fragment,s),p=!1},d(s){w(l,s)}}}function on(f){let l,p,s,r;return l=new _({props:{id:"model",option:"Nano",$$slots:{default:[tn]},$$scope:{ctx:f}}}),s=new _({props:{id:"model",option:"Super",$$slots:{default:[nn]},$$scope:{ctx:f}}}),{c(){d(l.$$.fragment),p=i(),d(s.$$.fragment)},l(n){M(l.$$.fragment,n),p=c(n),M(s.$$.fragment,n)},m(n,m){u(l,n,m),a(n,p,m),u(s,n,m),r=!0},p(n,m){const U={};m&2&&(U.$$scope={dirty:m,ctx:n}),l.$set(U);const v={};m&2&&(v.$$scope={dirty:m,ctx:n}),s.$set(v)},i(n){r\|\|(y(l.$$.fragment,n),y(s.$$.fragment,n),r=!0)},o(n){J(l.$$.fragment,n),J(s.$$.fragment,n),r=!1},d(n){n&&o(p),w(l,n),w(s,n)}}}function an(f){let l,p;return l=new Z({props:{code:"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",highlighted:`<span class="hljs-keyword">import</span> json

	<span class="hljs-keyword">import</span> torch
	<span class="hljs-keyword">from</span> diffusers <span class="hljs-keyword">import</span> Cosmos3OmniPipeline, CosmosActionCondition
	<span class="hljs-keyword">from</span> diffusers.schedulers.scheduling_unipc_multistep <span class="hljs-keyword">import</span> UniPCMultistepScheduler
	<span class="hljs-keyword">from</span> diffusers.utils <span class="hljs-keyword">import</span> export_to_video, load_video

	pipe = Cosmos3OmniPipeline.from_pretrained(
	<span class="hljs-string">"nvidia/Cosmos3-Nano"</span>, torch_dtype=torch.bfloat16, device_map=<span class="hljs-string">"cuda"</span>
	)
	pipe.scheduler = UniPCMultistepScheduler.from_config(
	pipe.scheduler.config, flow_shift=<span class="hljs-number">10.0</span>, use_karras_sigmas=<span class="hljs-literal">False</span>
	)

	prompt = <span class="hljs-string">"Put the pot to the left of the purple item."</span>
	video = load_video(
	<span class="hljs-string">"https://github.com/nvidia-cosmos/cosmos-dependencies/raw/refs/heads/assets/cosmos3/inputs/action/bridge_20260501_0.mp4"</span>
	)

	result = pipe(
	prompt=prompt,
	action=CosmosActionCondition(
	mode=<span class="hljs-string">"policy"</span>,
	chunk_size=<span class="hljs-number">16</span>,
	domain_name=<span class="hljs-string">"bridge_orig_lerobot"</span>,
	resolution_tier=<span class="hljs-number">480</span>,
	video=video,
	view_point=<span class="hljs-string">"ego_view"</span>,
	),
	fps=<span class="hljs-number">5</span>,
	num_inference_steps=<span class="hljs-number">30</span>,
	guidance_scale=<span class="hljs-number">1.0</span>,
	use_system_prompt=<span class="hljs-literal">False</span>,
	)

	<span class="hljs-comment"># macro_block_size=1 allows arbitrary frame sizes (Cosmos3 outputs are not always divisible by 16).</span>
	export_to_video(result.video, <span class="hljs-string">"sample.mp4"</span>, fps=<span class="hljs-number">5</span>, macro_block_size=<span class="hljs-number">1</span>)

	<span class="hljs-keyword">if</span> result.action <span class="hljs-keyword">is</span> <span class="hljs-keyword">not</span> <span class="hljs-literal">None</span>:
	<span class="hljs-keyword">with</span> <span class="hljs-built_in">open</span>(<span class="hljs-string">"sample_action.json"</span>, <span class="hljs-string">"w"</span>) <span class="hljs-keyword">as</span> f:
	json.dump(result.action[<span class="hljs-number">0</span>].tolist(), f)`,lang:"python",wrap:!1}}),{c(){d(l.$$.fragment)},l(s){M(l.$$.fragment,s)},m(s,r){u(l,s,r),p=!0},p:C,i(s){p\|\|(y(l.$$.fragment,s),p=!0)},o(s){J(l.$$.fragment,s),p=!1},d(s){w(l,s)}}}function pn(f){let l,p;return l=new Z({props:{code:"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",highlighted:`<span class="hljs-keyword">import</span> json

	<span class="hljs-keyword">import</span> torch
	<span class="hljs-keyword">from</span> diffusers <span class="hljs-keyword">import</span> Cosmos3OmniPipeline, CosmosActionCondition
	<span class="hljs-keyword">from</span> diffusers.schedulers.scheduling_unipc_multistep <span class="hljs-keyword">import</span> UniPCMultistepScheduler
	<span class="hljs-keyword">from</span> diffusers.utils <span class="hljs-keyword">import</span> export_to_video, load_video

	pipe = Cosmos3OmniPipeline.from_pretrained(
	<span class="hljs-string">"nvidia/Cosmos3-Super"</span>, torch_dtype=torch.bfloat16, device_map=<span class="hljs-string">"cuda"</span>
	)
	pipe.scheduler = UniPCMultistepScheduler.from_config(
	pipe.scheduler.config, flow_shift=<span class="hljs-number">10.0</span>, use_karras_sigmas=<span class="hljs-literal">False</span>
	)

	prompt = <span class="hljs-string">"Put the pot to the left of the purple item."</span>
	video = load_video(
	<span class="hljs-string">"https://github.com/nvidia-cosmos/cosmos-dependencies/raw/refs/heads/assets/cosmos3/inputs/action/bridge_20260501_0.mp4"</span>
	)

	result = pipe(
	prompt=prompt,
	action=CosmosActionCondition(
	mode=<span class="hljs-string">"policy"</span>,
	chunk_size=<span class="hljs-number">16</span>,
	domain_name=<span class="hljs-string">"bridge_orig_lerobot"</span>,
	resolution_tier=<span class="hljs-number">480</span>,
	video=video,
	view_point=<span class="hljs-string">"ego_view"</span>,
	),
	fps=<span class="hljs-number">5</span>,
	num_inference_steps=<span class="hljs-number">30</span>,
	guidance_scale=<span class="hljs-number">1.0</span>,
	use_system_prompt=<span class="hljs-literal">False</span>,
	)

	<span class="hljs-comment"># macro_block_size=1 allows arbitrary frame sizes (Cosmos3 outputs are not always divisible by 16).</span>
	export_to_video(result.video, <span class="hljs-string">"sample.mp4"</span>, fps=<span class="hljs-number">5</span>, macro_block_size=<span class="hljs-number">1</span>)

	<span class="hljs-keyword">if</span> result.action <span class="hljs-keyword">is</span> <span class="hljs-keyword">not</span> <span class="hljs-literal">None</span>:
	<span class="hljs-keyword">with</span> <span class="hljs-built_in">open</span>(<span class="hljs-string">"sample_action.json"</span>, <span class="hljs-string">"w"</span>) <span class="hljs-keyword">as</span> f:
	json.dump(result.action[<span class="hljs-number">0</span>].tolist(), f)`,lang:"python",wrap:!1}}),{c(){d(l.$$.fragment)},l(s){M(l.$$.fragment,s)},m(s,r){u(l,s,r),p=!0},p:C,i(s){p\|\|(y(l.$$.fragment,s),p=!0)},o(s){J(l.$$.fragment,s),p=!1},d(s){w(l,s)}}}function cn(f){let l,p,s,r;return l=new _({props:{id:"model",option:"Nano",$$slots:{default:[an]},$$scope:{ctx:f}}}),s=new _({props:{id:"model",option:"Super",$$slots:{default:[pn]},$$scope:{ctx:f}}}),{c(){d(l.$$.fragment),p=i(),d(s.$$.fragment)},l(n){M(l.$$.fragment,n),p=c(n),M(s.$$.fragment,n)},m(n,m){u(l,n,m),a(n,p,m),u(s,n,m),r=!0},p(n,m){const U={};m&2&&(U.$$scope={dirty:m,ctx:n}),l.$set(U);const v={};m&2&&(v.$$scope={dirty:m,ctx:n}),s.$set(v)},i(n){r\|\|(y(l.$$.fragment,n),y(s.$$.fragment,n),r=!0)},o(n){J(l.$$.fragment,n),J(s.$$.fragment,n),r=!1},d(n){n&&o(p),w(l,n),w(s,n)}}}function rn(f){let l,p,s,r,n,m,U,v="NVIDIA Cosmos 3 is a unified world foundation model (WFM) for Physical AI — a single omni-model that combines world generation, physical reasoning, and action generation. It replaces the separate Predict, Reason, and Transfer models from earlier Cosmos releases: whether you’re building for robotics, autonomous vehicles, or smart spaces, Cosmos 3 gives you one foundation to simulate and understand the physical world.",hs,L,ql="What’s shipping with this release:",bs,P,Yl="<li>Models on the Hugging Face Hub with model cards and licensing</li> <li>Cosmos 3 Diffusers integration for generation pipelines (this page)</li> <li>Post-training scripts for fine-tuning Cosmos 3 on your own data</li> <li>Open synthetic data generation (SDG) datasets for Physical AI</li>",fs,D,js,O,Fl="The biggest change from previous Cosmos releases is that Cosmos 3 is an <em>omni-model</em>, built on a Mixture-of-Transformers (MoT) architecture. Previously, developers worked with separate models for world generation (Predict), controlled generation (Transfer), scene understanding (Reason), and action-policy generation. Cosmos 3 unifies all of these in one model that reasons and generates across modalities in a single forward pass.",Us,K,Ll="From one model you can:",Ts,ee,Pl="<li>Generate physically plausible video worlds from text, images, or action inputs (image-to-video, text-to-video, action-conditioned video generation).</li> <li>Reason about physical properties like motion, causality, and spatial relationships.</li> <li>Predict future video and action sequences from the current state.</li> <li>Transfer scenes across viewpoints and conditions with structural control <em>(coming soon)</em>.</li>",vs,se,Dl='Under the hood, a single <code>Cosmos3OmniTransformer</code> runs a Qwen-style language model in parallel with a diffusion generation pathway: text tokens flow through a causal “understanding” stream while video and sound latents flow through a bi-directionally-attended “generation” stream, joined by a 3D multimodal RoPE. See the <a href="https://huggingface.co/papers/2501.03575" rel="nofollow">Cosmos World Foundation Model Platform paper</a> for the architectural background.',Zs,le,gs,te,Ol='Two checkpoints are released on the Hub — <a href="https://huggingface.co/nvidia/Cosmos3-Nano" rel="nofollow"><code>nvidia/Cosmos3-Nano</code></a> (smaller, faster) and <a href="https://huggingface.co/nvidia/Cosmos3-Super" rel="nofollow"><code>nvidia/Cosmos3-Super</code></a> (larger, higher quality). The same pipeline class supports text-to-image, text-to-video, image-to-video, and (with a sound-capable checkpoint) text+image-to-video-with-sound — pick a repo and use the per-model tab in each workflow below.',Bs,V,Kl='<p>Make sure to check out the Schedulers <a href="../../using-diffusers/schedulers">guide</a> to learn how to explore the tradeoff between scheduler speed and quality, and see the <a href="../../using-diffusers/loading#reuse-a-pipeline">reuse components across pipelines</a> section to learn how to efficiently load the same components into multiple pipelines.</p>',_s,ne,Cs,oe,et='Cosmos 3 was trained on long, highly descriptive captions. For optimal quality, short text prompts should be <strong>upsampled into a specific JSON structure</strong> before they are passed to the pipeline. The upsampler lives in the <a href="https://github.com/NVIDIA/cosmos-framework" rel="nofollow">cosmos-framework</a> package.',Ns,ae,st="Start from a short, plain-text prompt and save it to <code>assets/prompt.txt</code>. For the text-to-video example below, the original prompt is <em>“A robotic arm is cleaning a plate in a kitchen”</em>:",Is,pe,Gs,ie,lt="Then install the framework and run the upsampler. The example below upsamples for text-to-video using Opus-4.6:",$s,ce,ks,re,tt="Switch <code>--mode</code> to match the workflow you are targeting (<code>text2image</code>, <code>text2video</code>, <code>image2video</code>). The command writes the upsampled prompt(s) to the <code>--output</code> file as a JSON array (one object per non-empty line in <code>--input</code>); pass a <code>.jsonl</code> path instead to get one JSON object per line. For <code>image2video</code>, you must also supply the conditioning image via <code>--image-url</code> (a URL or local path) or <code>--image-list</code> (one image per prompt).",Ws,me,nt="A pre-upsampled positive prompt (<code>assets/example_t2v_prompt.json</code>) and negative prompt (<code>assets/negative_prompt.json</code>) are provided for convenience, and are used by the generation examples below. The examples load these JSON files and pass them to the pipeline as JSON strings via <code>json.dumps(...)</code>.",Xs,de,zs,Me,ot='Multi-frame generation conditioned on text alone. Pick <code>num_frames</code> based on the target duration — the default <code>num_frames=189</code> produces ≈ 7.9 s at 24 FPS. The prompt and negative prompt are read from the JSON-upsampled files described in <a href="#prompt-upsampling">Prompt upsampling</a>.',Qs,R,Vs,ue,Rs,ye,at="Single-frame generation. The model is conditioned only on the text prompt; pass <code>num_frames=1</code>. Upsample with <code>--mode text2image</code> to produce the JSON prompt.",Es,E,Hs,Je,Ss,we,pt="Pass a conditioning image via <code>image=</code>. The pipeline anchors frame 0 to the supplied image and denoises the rest. Upsample with <code>--mode image2video</code> to produce the JSON prompt.",As,H,xs,he,qs,be,it='Pass a conditioning clip via <code>video=</code> (e.g. from <code>load_video</code>). The pipeline anchors the leading latent frames given by <code>condition_frame_indexes_vision</code> (default <code>[0, 1]</code>) to the clip and denoises the rest. Use <code>condition_video_keep</code> (<code>"first"</code> or <code>"last"</code>) to choose which end of a longer source clip the conditioning frames are taken from. As with the other modes, the prompt should follow the descriptive JSON structure described in <a href="#prompt-upsampling">Prompt upsampling</a>.',Ys,S,Fs,fe,Ls,je,ct='When the checkpoint carries a <code>sound_tokenizer</code>, add <code>enable_sound=True</code> to the video-to-video call to jointly generate a synchronized audio track. The waveform is returned alongside the video and can be muxed into the MP4 with <a href="/docs/diffusers/pr_14045/en/api/utilities#diffusers.utils.encode_video">encode_video()</a>.',Ps,A,Ds,Ue,Os,Te,rt='When the checkpoint carries a <code>sound_tokenizer</code>, pass <code>enable_sound=True</code> to jointly generate a synchronized audio track. The waveform is returned alongside the video and can be muxed into the MP4 with <a href="/docs/diffusers/pr_14045/en/api/utilities#diffusers.utils.encode_video">encode_video()</a>.',Ks,ve,mt="This is the same call as the text-to-video example above with <code>enable_sound=True</code> added:",el,x,sl,Ze,ll,ge,dt='Action runs group every action-specific input into a <a href="/docs/diffusers/pr_14045/en/api/pipelines/cosmos3#diffusers.CosmosActionCondition">CosmosActionCondition</a> passed via the <code>action</code> argument instead of the top-level <code>image</code> / <code>video</code> / <code>height</code> / <code>width</code> arguments. Set <code>resolution_tier</code> (<code>256</code>/<code>480</code>/<code>704</code>/<code>720</code>) close to the input video’s native resolution; it selects the conditioning canvas. Cosmos 3 supports three action modes — <code>policy</code>, <code>forward_dynamics</code>, and <code>inverse_dynamics</code>. <code>policy</code> and <code>forward_dynamics</code> condition only on the first frame (so an <code>image</code> or a <code>video</code> both work), while <code>inverse_dynamics</code> requires a <code>video</code>. The conditioning video for an action run is set on <code>action.video</code> (or <code>action.image</code>), not on the pipeline’s top-level <code>video</code> argument.',tl,Be,Mt="Pass a plain task description as <code>prompt</code> and pick the camera with <code>action.view_point</code> (default <code>"ego_view"</code>; also <code>"third_person_view"</code>, <code>"wrist_view"</code>, <code>"concat_view"</code>). The pipeline turns these into the structured JSON caption the model was trained on, so action prompts should not be LLM-upsampled.",nl,_e,ol,Ce,ut="Action policy generation predicts future video and action tokens from the first observation frame, text prompt, and action domain metadata. The example below uses the Bridge robot domain and writes the predicted action chunk to JSON in model-normalized action space.",al,q,pl,Ne,il,Ie,yt="<code>tokenize_prompt</code> appends short metadata sentences inside the user message so the LLM sees the conditioning the model was trained with. The positive prompt gets sentences like <em>“The video is 7.9 seconds long and is of 24 FPS.”</em> and <em>“This video is of 720x1280 resolution.”</em>; the negative prompt gets the inverse (<em>”… is not …”</em>).",cl,Ge,Jt="Both are on by default. Disable either pair through <code>__call__</code>:",rl,$e,ml,ke,wt="<code>add_duration_template</code> has no effect when <code>num_frames == 1</code> (image mode); only the resolution sentence is appended in that case.",dl,We,Ml,Xe,ht='Cosmos3 wires up the <a href="https://pypi.org/project/cosmos-guardrail/" rel="nofollow"><code>cosmos_guardrail</code></a> <code>CosmosSafetyChecker</code> and runs it <strong>by default</strong>. The text guardrail rejects unsafe prompts before generation (<code>ValueError</code>); the video guardrail runs on the decoded frames and either pixelates detected faces or rejects the output. Audio output is not guardrailed.',ul,ze,bt="Install the optional dependency to enable the default checker:",yl,Qe,Jl,Ve,ft="The checker is mandatory under the NVIDIA Open Model License Agreement. The two flags below exist for tests and development workflows where the guardrail would be redundant (e.g., the input has already been cleared, or you are intentionally exercising the pipeline on edge inputs).",wl,Re,jt="<strong>Disable at construction</strong> (no checker is instantiated, so no guardrail models are downloaded or loaded into memory):",hl,Ee,bl,He,Ut="<strong>Disable for a single call</strong> (checker stays loaded — useful for one-off bypass while keeping the default on for subsequent calls):",fl,Se,jl,Ae,Tt="To supply a custom checker (e.g., a no-op subclass for fast tests), pass it as <code>safety_checker=</code>:",Ul,xe,Tl,qe,vl,N,Ye,$l,k,Fe,kl,as,vt="Decode a sound latent <code>[C, T]</code> to a waveform <code>[audio_ch, N]</code>.",Wl,ps,Zt="Adds/removes the batch dimension expected by the sound tokenizer decoder.",Xl,Y,Le,zl,is,gt="Build conditioning + initial noise for a single sample.",Ql,B,Pe,Vl,cs,Bt="Apply prompt-augmentation templates and tokenize cond/uncond prompts via the Qwen2 chat template.",Rl,rs,_t=`This pipeline does not run a separate text encoder: the joint Cosmos3 transformer consumes raw Qwen2 token IDs
	alongside vision (and optionally sound) tokens.`,El,ms,Ct=`When <code>negative_prompt</code> is <code>None</code>, an empty string is used; the Cosmos3 docs page documents recommended
	quality-control negative prompts to pass explicitly for text2video / image2video. The duration and resolution
	templates are appended to the prompt, and inverse templates are appended to the negative prompt, when enabled.`,Hl,ds,Nt=`When <code>action_mode</code> is set, the prompt is instead converted to the structured action JSON caption the model
	was trained on (see <code>_build_action_json_prompt</code>), using <code>action_view_point</code> for the framing field; the
	flat metadata templates are skipped because the JSON already carries duration/fps/resolution/aspect_ratio.`,Zl,De,It="<li>all</li> <li><strong>call</strong></li>",gl,Oe,Bl,G,Ke,Sl,Ms,Gt="Groups every input required for a Cosmos 3 action-conditioned generation task.",Al,us,$t=`Pass this to <code>Cosmos3OmniPipeline.__call__()</code> via the <code>action</code> argument instead of the top-level <code>image</code> / <code>height</code>
	/ <code>width</code> arguments, which are reserved for t2v, i2v runs.`,_l,es,Cl,W,ss,xl,ys,kt='Output dataclass for <a href="/docs/diffusers/pr_14045/en/api/pipelines/cosmos3#diffusers.Cosmos3OmniPipeline">Cosmos3OmniPipeline</a>.',Nl,ls,Il,ws,Gl;return n=new g({props:{title:"Cosmos 3",local:"cosmos-3",headingTag:"h1"}}),D=new g({props:{title:"What’s new in Cosmos 3",local:"whats-new-in-cosmos-3",headingTag:"h2"}}),le=new g({props:{title:"Available checkpoints",local:"available-checkpoints",headingTag:"h2"}}),ne=new g({props:{title:"Prompt upsampling",local:"prompt-upsampling",headingTag:"h2"}}),pe=new Z({props:{code:"bWtkaXIlMjAtcCUyMGFzc2V0cyUwQWVjaG8lMjAlMjJBJTIwcm9ib3RpYyUyMGFybSUyMGlzJTIwY2xlYW5pbmclMjBhJTIwcGxhdGUlMjBpbiUyMGElMjBraXRjaGVuJTIyJTIwJTNFJTIwYXNzZXRzJTJGcHJvbXB0LnR4dA==",highlighted:`<span class="hljs-built_in">mkdir</span> -p assets
	<span class="hljs-built_in">echo</span> <span class="hljs-string">"A robotic arm is cleaning a plate in a kitchen"</span> > assets/prompt.txt`,lang:"bash",wrap:!1}}),ce=new Z({props:{code:"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",highlighted:`git <span class="hljs-built_in">clone</span> https://github.com/NVIDIA/cosmos-framework.git packages/cosmos-framework
	pip install -e packages/cosmos-framework

	<span class="hljs-built_in">export</span> PROMPT_UPSAMPLER_ENDPOINT_URL=<span class="hljs-string">"https://api.anthropic.com/v1/"</span>
	<span class="hljs-built_in">export</span> PROMPT_UPSAMPLER_MODEL_NAME=<span class="hljs-string">"claude-opus-4-6"</span>
	<span class="hljs-built_in">export</span> PROMPT_UPSAMPLER_API_TOKEN=<span class="hljs-string">"<your_token>"</span>

	python -m cosmos_framework.inference.prompt_upsampling \\
	--input assets/prompt.txt \\
	--output assets/example_t2v_prompt.json \\
	--mode text2video \\
	--endpoint-url <span class="hljs-string">"<span class="hljs-variable">\${PROMPT_UPSAMPLER_ENDPOINT_URL}</span>"</span> \\
	--model <span class="hljs-string">"<span class="hljs-variable">\${PROMPT_UPSAMPLER_MODEL_NAME}</span>"</span> \\
	--api-token <span class="hljs-string">"<span class="hljs-variable">\${PROMPT_UPSAMPLER_API_TOKEN}</span>"</span> \\
	--resolution 720 \\
	--aspect-ratio <span class="hljs-string">"16,9"</span>`,lang:"bash",wrap:!1}}),de=new g({props:{title:"Text-to-video",local:"text-to-video",headingTag:"h2"}}),R=new os({props:{id:"model",options:["Nano","Super"],$$slots:{default:[At]},$$scope:{ctx:f}}}),ue=new g({props:{title:"Text-to-image",local:"text-to-image",headingTag:"h2"}}),E=new os({props:{id:"model",options:["Nano","Super"],$$slots:{default:[Yt]},$$scope:{ctx:f}}}),Je=new g({props:{title:"Image-to-video",local:"image-to-video",headingTag:"h2"}}),H=new os({props:{id:"model",options:["Nano","Super"],$$slots:{default:[Pt]},$$scope:{ctx:f}}}),he=new g({props:{title:"Video-to-video",local:"video-to-video",headingTag:"h2"}}),S=new os({props:{id:"model",options:["Nano","Super"],$$slots:{default:[Kt]},$$scope:{ctx:f}}}),fe=new g({props:{title:"Video-to-video with sound",local:"video-to-video-with-sound",headingTag:"h2"}}),A=new os({props:{id:"model",options:["Nano","Super"],$$slots:{default:[ln]},$$scope:{ctx:f}}}),Ue=new g({props:{title:"Text-to-video with sound",local:"text-to-video-with-sound",headingTag:"h2"}}),x=new os({props:{id:"model",options:["Nano","Super"],$$slots:{default:[on]},$$scope:{ctx:f}}}),Ze=new g({props:{title:"Action-conditioned generation",local:"action-conditioned-generation",headingTag:"h2"}}),_e=new g({props:{title:"Action policy",local:"action-policy",headingTag:"h3"}}),q=new os({props:{id:"model",options:["Nano","Super"],$$slots:{default:[cn]},$$scope:{ctx:f}}}),Ne=new g({props:{title:"Metadata templates",local:"metadata-templates",headingTag:"h2"}}),$e=new Z({props:{code:"cmVzdWx0JTIwJTNEJTIwcGlwZSglMEElMjAlMjAlMjAlMjBwcm9tcHQlM0Rwcm9tcHQlMkMlMEElMjAlMjAlMjAlMjBuZWdhdGl2ZV9wcm9tcHQlM0RuZWdhdGl2ZV9wcm9tcHQlMkMlMEElMjAlMjAlMjAlMjBudW1fZnJhbWVzJTNEMTg5JTJDJTBBJTIwJTIwJTIwJTIwaGVpZ2h0JTNENzIwJTJDJTBBJTIwJTIwJTIwJTIwd2lkdGglM0QxMjgwJTJDJTBBJTIwJTIwJTIwJTIwZnBzJTNEMjQuMCUyQyUwQSUyMCUyMCUyMCUyMGFkZF9kdXJhdGlvbl90ZW1wbGF0ZSUzREZhbHNlJTJDJTIwJTIwJTIwJTIwJTIzJTIwc2tpcCUyMHRoZSUyMGR1cmF0aW9uJTIwc2VudGVuY2UlMjBvbiUyMGJvdGglMjBwcm9tcHRzJTBBJTIwJTIwJTIwJTIwYWRkX3Jlc29sdXRpb25fdGVtcGxhdGUlM0RGYWxzZSUyQyUyMCUyMCUyMyUyMHNraXAlMjB0aGUlMjByZXNvbHV0aW9uJTIwc2VudGVuY2UlMjBvbiUyMGJvdGglMjBwcm9tcHRzJTBBKQ==",highlighted:`result = pipe(
	prompt=prompt,
	negative_prompt=negative_prompt,
	num_frames=<span class="hljs-number">189</span>,
	height=<span class="hljs-number">720</span>,
	width=<span class="hljs-number">1280</span>,
	fps=<span class="hljs-number">24.0</span>,
	add_duration_template=<span class="hljs-literal">False</span>, <span class="hljs-comment"># skip the duration sentence on both prompts</span>
	add_resolution_template=<span class="hljs-literal">False</span>, <span class="hljs-comment"># skip the resolution sentence on both prompts</span>
	)`,lang:"python",wrap:!1}}),We=new g({props:{title:"Safety checker",local:"safety-checker",headingTag:"h2"}}),Qe=new Z({props:{code:"cGlwJTIwaW5zdGFsbCUyMGNvc21vc19ndWFyZHJhaWw=",highlighted:'pip <span class="hljs-keyword">install</span> cosmos_guardrail',lang:"",wrap:!1}}),Ee=new Z({props:{code:"aW1wb3J0JTIwdG9yY2glMEFmcm9tJTIwZGlmZnVzZXJzJTIwaW1wb3J0JTIwQ29zbW9zM09tbmlQaXBlbGluZSUwQSUwQXBpcGUlMjAlM0QlMjBDb3Ntb3MzT21uaVBpcGVsaW5lLmZyb21fcHJldHJhaW5lZCglMEElMjAlMjAlMjAlMjAlMjJudmlkaWElMkZDb3Ntb3MzLU5hbm8lMjIlMkMlMEElMjAlMjAlMjAlMjB0b3JjaF9kdHlwZSUzRHRvcmNoLmJmbG9hdDE2JTJDJTBBJTIwJTIwJTIwJTIwZGV2aWNlX21hcCUzRCUyMmN1ZGElMjIlMkMlMEElMjAlMjAlMjAlMjBlbmFibGVfc2FmZXR5X2NoZWNrZXIlM0RGYWxzZSUyQyUwQSk=",highlighted:`<span class="hljs-keyword">import</span> torch
	<span class="hljs-keyword">from</span> diffusers <span class="hljs-keyword">import</span> Cosmos3OmniPipeline

	pipe = Cosmos3OmniPipeline.from_pretrained(
	<span class="hljs-string">"nvidia/Cosmos3-Nano"</span>,
	torch_dtype=torch.bfloat16,
	device_map=<span class="hljs-string">"cuda"</span>,
	enable_safety_checker=<span class="hljs-literal">False</span>,
	)`,lang:"python",wrap:!1}}),Se=new Z({props:{code:"cmVzdWx0JTIwJTNEJTIwcGlwZSglMEElMjAlMjAlMjAlMjBwcm9tcHQlM0Rwcm9tcHQlMkMlMEElMjAlMjAlMjAlMjBudW1fZnJhbWVzJTNEMTg5JTJDJTBBJTIwJTIwJTIwJTIwaGVpZ2h0JTNENzIwJTJDJTBBJTIwJTIwJTIwJTIwd2lkdGglM0QxMjgwJTJDJTBBJTIwJTIwJTIwJTIwZnBzJTNEMjQuMCUyQyUwQSUyMCUyMCUyMCUyMGVuYWJsZV9zYWZldHlfY2hlY2slM0RGYWxzZSUyQyUwQSk=",highlighted:`result = pipe(
	prompt=prompt,
	num_frames=<span class="hljs-number">189</span>,
	height=<span class="hljs-number">720</span>,
	width=<span class="hljs-number">1280</span>,
	fps=<span class="hljs-number">24.0</span>,
	enable_safety_check=<span class="hljs-literal">False</span>,
	)`,lang:"python",wrap:!1}}),xe=new Z({props:{code:"cGlwZSUyMCUzRCUyMENvc21vczNPbW5pUGlwZWxpbmUuZnJvbV9wcmV0cmFpbmVkKCUwQSUyMCUyMCUyMCUyMCUyMm52aWRpYSUyRkNvc21vczMtTmFubyUyMiUyQyUwQSUyMCUyMCUyMCUyMHRvcmNoX2R0eXBlJTNEdG9yY2guYmZsb2F0MTYlMkMlMEElMjAlMjAlMjAlMjBkZXZpY2VfbWFwJTNEJTIyY3VkYSUyMiUyQyUwQSUyMCUyMCUyMCUyMHNhZmV0eV9jaGVja2VyJTNETXlDdXN0b21TYWZldHlDaGVja2VyKCklMkMlMEEp",highlighted:`pipe = Cosmos3OmniPipeline.from_pretrained(
	<span class="hljs-string">"nvidia/Cosmos3-Nano"</span>,
	torch_dtype=torch.bfloat16,
	device_map=<span class="hljs-string">"cuda"</span>,
	safety_checker=MyCustomSafetyChecker(),
	)`,lang:"python",wrap:!1}}),qe=new g({props:{title:"Cosmos3OmniPipeline",local:"diffusers.Cosmos3OmniPipeline",headingTag:"h2"}}),Ye=new Js({props:{name:"class diffusers.Cosmos3OmniPipeline",anchor:"diffusers.Cosmos3OmniPipeline",parameters:[{name:"transformer",val:": Cosmos3OmniTransformer"},{name:"text_tokenizer",val:": AutoTokenizer"},{name:"vae",val:": AutoencoderKLWan"},{name:"scheduler",val:": UniPCMultistepScheduler"},{name:"sound_tokenizer",val:": diffusers.models.autoencoders.autoencoder_cosmos3_audio.Cosmos3AVAEAudioTokenizer \| None = None"},{name:"safety_checker",val:": diffusers.pipelines.cosmos.pipeline_cosmos3_omni.CosmosSafetyChecker \| None = None"},{name:"enable_safety_checker",val:": bool = True"}],source:"https://github.com/huggingface/diffusers/blob/vr_14045/src/diffusers/pipelines/cosmos/pipeline_cosmos3_omni.py#L365"}}),Fe=new Js({props:{name:"decode_sound",anchor:"diffusers.Cosmos3OmniPipeline.decode_sound",parameters:[{name:"latent",val:": Tensor"}],source:"https://github.com/huggingface/diffusers/blob/vr_14045/src/diffusers/pipelines/cosmos/pipeline_cosmos3_omni.py#L459"}}),Le=new Js({props:{name:"prepare_latents",anchor:"diffusers.Cosmos3OmniPipeline.prepare_latents",parameters:[{name:"image",val:": torch.Tensor \| None = None"},{name:"video",val:": list[PIL.Image.Image] \| torch.Tensor \| numpy.ndarray \| None = None"},{name:"condition_frame_indexes_vision",val:": Iterable = (0, 1)"},{name:"condition_video_keep",val:": typing.Literal['first', 'last'] = 'first'"},{name:"num_frames",val:": int \| None = None"},{name:"height",val:": int \| None = None"},{name:"width",val:": int \| None = None"},{name:"fps",val:": float = 24.0"},{name:"latents",val:": torch.Tensor \| None = None"},{name:"sound_latents",val:": torch.Tensor \| None = None"},{name:"action_latents",val:": torch.Tensor \| None = None"},{name:"generator",val:": torch._C.Generator \| None = None"},{name:"device",val:": str = 'cuda'"},{name:"dtype",val:": dtype = torch.bfloat16"},{name:"enable_sound",val:": bool = False"},{name:"action",val:": CosmosActionCondition \| None = None"}],source:"https://github.com/huggingface/diffusers/blob/vr_14045/src/diffusers/pipelines/cosmos/pipeline_cosmos3_omni.py#L705",returnDescription:`<script context="module">export const metadata = 'undefined';<\/script>


	<p>Initial noisy tensors plus condition masks/metadata for vision, sound, and optional action modalities.</p>
	`}}),Pe=new Js({props:{name:"tokenize_prompt",anchor:"diffusers.Cosmos3OmniPipeline.tokenize_prompt",parameters:[{name:"prompt",val:": str"},{name:"negative_prompt",val:": str \| None = None"},{name:"num_frames",val:": int = 189"},{name:"height",val:": int = 720"},{name:"width",val:": int = 1280"},{name:"fps",val:": float = 24.0"},{name:"use_system_prompt",val:": bool = True"},{name:"add_resolution_template",val:": bool = True"},{name:"add_duration_template",val:": bool = True"},{name:"action_mode",val:": str \| None = None"},{name:"action_view_point",val:": str \| None = None"}],source:"https://github.com/huggingface/diffusers/blob/vr_14045/src/diffusers/pipelines/cosmos/pipeline_cosmos3_omni.py#L1075",returnDescription:`<script context="module">export const metadata = 'undefined';<\/script>


	<p><code>(cond_input_ids, uncond_input_ids)</code> — token-id lists for this sample.</p>
	`}}),Oe=new g({props:{title:"CosmosActionCondition",local:"diffusers.CosmosActionCondition",headingTag:"h2"}}),Ke=new Js({props:{name:"class diffusers.CosmosActionCondition",anchor:"diffusers.CosmosActionCondition",parameters:[{name:"mode",val:": typing.Literal['policy', 'forward_dynamics', 'inverse_dynamics']"},{name:"chunk_size",val:": int"},{name:"domain_name",val:": str"},{name:"resolution_tier",val:": int = 480"},{name:"raw_actions",val:": torch.Tensor \| None = None"},{name:"image",val:": PIL.Image.Image \| numpy.ndarray \| torch.Tensor \| None = None"},{name:"video",val:": list \| numpy.ndarray \| torch.Tensor \| None = None"},{name:"view_point",val:": str = 'ego_view'"}],parametersDescription:[{anchor:"diffusers.CosmosActionCondition.mode",description:`<strong>mode</strong> (<code>str</code>) —
	The action task. One of <code>"forward_dynamics"</code> (roll out future video from a first frame and a given
	<code>raw_actions</code> sequence), <code>"inverse_dynamics"</code> (infer the actions connecting the conditioning frames), or
	<code>"policy"</code> (jointly roll out future video and actions from the first frame).`,name:"mode"},{anchor:"diffusers.CosmosActionCondition.chunk_size",description:`<strong>chunk_size</strong> (<code>int</code>) —
	Number of action transition steps in the chunk. The paired conditioning video spans <code>chunk_size + 1</code>
	frames.`,name:"chunk_size"},{anchor:"diffusers.CosmosActionCondition.domain_name",description:`<strong>domain_name</strong> (<code>str</code>) —
	Embodiment domain selecting the domain-aware action projection weights. Must be one of the registered
	Cosmos 3 embodiment domains. It also fixes the unpadded action width used to slice predicted actions,
	resolved internally from this name (see <code>_EMBODIMENT_TO_RAW_ACTION_DIM</code>).`,name:"domain_name"},{anchor:"diffusers.CosmosActionCondition.resolution_tier",description:`<strong>resolution_tier</strong> (<code>int</code>, defaults to <code>480</code>) —
	Action conditioning resolution <em>tier</em> (one of <code>256</code>, <code>480</code>, <code>704</code>, <code>720</code>). The tier picks a predefined
	canvas whose aspect ratio is closest to the input; the input is downscaled (never upscaled) and padded into
	it for conditioning. This is not the output frame size, which tracks the input content. Match the tier to
	the input’s native resolution: a lower tier discards detail, while a higher tier adds no resolution (no
	upscaling), wastes compute on padding, and is a train/inference mismatch that can hurt quality.`,name:"resolution_tier"},{anchor:"diffusers.CosmosActionCondition.raw_actions",description:`<strong>raw_actions</strong> (<code>torch.Tensor</code>, <em>optional</em>) —
	Raw domain action vectors of shape <code>[T, raw_action_dim]</code> driving <code>"forward_dynamics"</code>. Sequences shorter
	than <code>chunk_size</code> repeat the last action; longer ones are truncated. Channels beyond the model’s
	<code>action_dim</code> are rejected, and narrower inputs are zero-padded up to <code>action_dim</code>.`,name:"raw_actions"},{anchor:"diffusers.CosmosActionCondition.image",description:`<strong>image</strong> (<code>PIL.Image.Image</code>, <code>np.ndarray</code>, or <code>torch.Tensor</code>, <em>optional</em>) —
	Conditioning frame for <code>"policy"</code> / <code>"forward_dynamics"</code>. Mutually exclusive with <code>video</code>.`,name:"image"},{anchor:"diffusers.CosmosActionCondition.video",description:`<strong>video</strong> (<code>list</code>, <code>np.ndarray</code>, or <code>torch.Tensor</code>, <em>optional</em>) —
	Conditioning video, required for <code>"inverse_dynamics"</code>. For <code>"policy"</code> / <code>"forward_dynamics"</code> only its first
	frame is used. Mutually exclusive with <code>image</code>.`,name:"video"},{anchor:"diffusers.CosmosActionCondition.view_point",description:`<strong>view_point</strong> (<code>str</code>, defaults to <code>"ego_view"</code>) —
	Camera perspective label used to populate the action caption’s <code>cinematography.framing</code> field. One of
	<code>"ego_view"</code>, <code>"third_person_view"</code>, <code>"wrist_view"</code>, or <code>"concat_view"</code>. The action model was trained on
	structured JSON captions that carry this viewpoint sentence; an unrecognized label drops the framing field
	(with a warning).`,name:"view_point"}],source:"https://github.com/huggingface/diffusers/blob/vr_14045/src/diffusers/pipelines/cosmos/pipeline_cosmos3_omni.py#L254"}}),es=new g({props:{title:"Cosmos3OmniPipelineOutput",local:"diffusers.pipelines.cosmos.pipeline_cosmos3_omni.Cosmos3OmniPipelineOutput",headingTag:"h2"}}),ss=new Js({props:{name:"class diffusers.pipelines.cosmos.pipeline_cosmos3_omni.Cosmos3OmniPipelineOutput",anchor:"diffusers.pipelines.cosmos.pipeline_cosmos3_omni.Cosmos3OmniPipelineOutput",parameters:[{name:"video",val:": typing.Any"},{name:"sound",val:": torch.Tensor \| None = None"},{name:"action",val:": list[torch.Tensor] \| None = None"}],parametersDescription:[{anchor:"diffusers.pipelines.cosmos.pipeline_cosmos3_omni.Cosmos3OmniPipelineOutput.video",description:`<strong>video</strong> — The generated video. The exact type depends on <code>output_type</code>
	passed to the pipeline: a list of PIL frames for <code>"pil"</code> (default), an <code>np.ndarray</code> of shape <code>[T, H, W, C]</code> for <code>"np"</code>, a <code>torch.Tensor</code> of shape <code>[T, C, H, W]</code> for <code>"pt"</code>, or a raw latent tensor
	when <code>output_type="latent"</code>.`,name:"video"},{anchor:"diffusers.pipelines.cosmos.pipeline_cosmos3_omni.Cosmos3OmniPipelineOutput.sound",description:`<strong>sound</strong> — Decoded audio waveform of shape <code>[C, N]</code>. <code>None</code> when
	<code>enable_sound=False</code>.`,name:"sound"},{anchor:"diffusers.pipelines.cosmos.pipeline_cosmos3_omni.Cosmos3OmniPipelineOutput.action",description:"<strong>action</strong> — Predicted action tokens. <code>None</code> unless an action mode predicts actions.",name:"action"}],source:"https://github.com/huggingface/diffusers/blob/vr_14045/src/diffusers/pipelines/cosmos/pipeline_cosmos3_omni.py#L235"}}),ls=new Et({props:{source:"https://github.com/huggingface/diffusers/blob/main/docs/source/en/api/pipelines/cosmos3.md"}}),{c(){l=h("meta"),p=i(),s=h("p"),r=i(),d(n.$$.fragment),m=i(),U=h("p"),U.textContent=v,hs=i(),L=h("p"),L.textContent=ql,bs=i(),P=h("ul"),P.innerHTML=Yl,fs=i(),d(D.$$.fragment),js=i(),O=h("p"),O.innerHTML=Fl,Us=i(),K=h("p"),K.textContent=Ll,Ts=i(),ee=h("ul"),ee.innerHTML=Pl,vs=i(),se=h("p"),se.innerHTML=Dl,Zs=i(),d(le.$$.fragment),gs=i(),te=h("p"),te.innerHTML=Ol,Bs=i(),V=h("blockquote"),V.innerHTML=Kl,_s=i(),d(ne.$$.fragment),Cs=i(),oe=h("p"),oe.innerHTML=et,Ns=i(),ae=h("p"),ae.innerHTML=st,Is=i(),d(pe.$$.fragment),Gs=i(),ie=h("p"),ie.textContent=lt,$s=i(),d(ce.$$.fragment),ks=i(),re=h("p"),re.innerHTML=tt,Ws=i(),me=h("p"),me.innerHTML=nt,Xs=i(),d(de.$$.fragment),zs=i(),Me=h("p"),Me.innerHTML=ot,Qs=i(),d(R.$$.fragment),Vs=i(),d(ue.$$.fragment),Rs=i(),ye=h("p"),ye.innerHTML=at,Es=i(),d(E.$$.fragment),Hs=i(),d(Je.$$.fragment),Ss=i(),we=h("p"),we.innerHTML=pt,As=i(),d(H.$$.fragment),xs=i(),d(he.$$.fragment),qs=i(),be=h("p"),be.innerHTML=it,Ys=i(),d(S.$$.fragment),Fs=i(),d(fe.$$.fragment),Ls=i(),je=h("p"),je.innerHTML=ct,Ps=i(),d(A.$$.fragment),Ds=i(),d(Ue.$$.fragment),Os=i(),Te=h("p"),Te.innerHTML=rt,Ks=i(),ve=h("p"),ve.innerHTML=mt,el=i(),d(x.$$.fragment),sl=i(),d(Ze.$$.fragment),ll=i(),ge=h("p"),ge.innerHTML=dt,tl=i(),Be=h("p"),Be.innerHTML=Mt,nl=i(),d(_e.$$.fragment),ol=i(),Ce=h("p"),Ce.textContent=ut,al=i(),d(q.$$.fragment),pl=i(),d(Ne.$$.fragment),il=i(),Ie=h("p"),Ie.innerHTML=yt,cl=i(),Ge=h("p"),Ge.innerHTML=Jt,rl=i(),d($e.$$.fragment),ml=i(),ke=h("p"),ke.innerHTML=wt,dl=i(),d(We.$$.fragment),Ml=i(),Xe=h("p"),Xe.innerHTML=ht,ul=i(),ze=h("p"),ze.textContent=bt,yl=i(),d(Qe.$$.fragment),Jl=i(),Ve=h("p"),Ve.textContent=ft,wl=i(),Re=h("p"),Re.innerHTML=jt,hl=i(),d(Ee.$$.fragment),bl=i(),He=h("p"),He.innerHTML=Ut,fl=i(),d(Se.$$.fragment),jl=i(),Ae=h("p"),Ae.innerHTML=Tt,Ul=i(),d(xe.$$.fragment),Tl=i(),d(qe.$$.fragment),vl=i(),N=h("div"),d(Ye.$$.fragment),$l=i(),k=h("div"),d(Fe.$$.fragment),kl=i(),as=h("p"),as.innerHTML=vt,Wl=i(),ps=h("p"),ps.textContent=Zt,Xl=i(),Y=h("div"),d(Le.$$.fragment),zl=i(),is=h("p"),is.textContent=gt,Ql=i(),B=h("div"),d(Pe.$$.fragment),Vl=i(),cs=h("p"),cs.textContent=Bt,Rl=i(),rs=h("p"),rs.textContent=_t,El=i(),ms=h("p"),ms.innerHTML=Ct,Hl=i(),ds=h("p"),ds.innerHTML=Nt,Zl=i(),De=h("ul"),De.innerHTML=It,gl=i(),d(Oe.$$.fragment),Bl=i(),G=h("div"),d(Ke.$$.fragment),Sl=i(),Ms=h("p"),Ms.textContent=Gt,Al=i(),us=h("p"),us.innerHTML=$t,_l=i(),d(es.$$.fragment),Cl=i(),W=h("div"),d(ss.$$.fragment),xl=i(),ys=h("p"),ys.innerHTML=kt,Nl=i(),d(ls.$$.fragment),Il=i(),ws=h("p"),this.h()},l(e){const t=Rt("svelte-u9bgzb",document.head);l=b(t,"META",{name:!0,content:!0}),t.forEach(o),p=c(e),s=b(e,"P",{}),F(s).forEach(o),r=c(e),M(n.$$.fragment,e),m=c(e),U=b(e,"P",{"data-svelte-h":!0}),j(U)!=="svelte-1bfg7u9"&&(U.textContent=v),hs=c(e),L=b(e,"P",{"data-svelte-h":!0}),j(L)!=="svelte-16acpvv"&&(L.textContent=ql),bs=c(e),P=b(e,"UL",{"data-svelte-h":!0}),j(P)!=="svelte-l7xz5s"&&(P.innerHTML=Yl),fs=c(e),M(D.$$.fragment,e),js=c(e),O=b(e,"P",{"data-svelte-h":!0}),j(O)!=="svelte-t3k2lx"&&(O.innerHTML=Fl),Us=c(e),K=b(e,"P",{"data-svelte-h":!0}),j(K)!=="svelte-1qyj0gu"&&(K.textContent=Ll),Ts=c(e),ee=b(e,"UL",{"data-svelte-h":!0}),j(ee)!=="svelte-v2tptm"&&(ee.innerHTML=Pl),vs=c(e),se=b(e,"P",{"data-svelte-h":!0}),j(se)!=="svelte-1xm4gj6"&&(se.innerHTML=Dl),Zs=c(e),M(le.$$.fragment,e),gs=c(e),te=b(e,"P",{"data-svelte-h":!0}),j(te)!=="svelte-1jbbrtc"&&(te.innerHTML=Ol),Bs=c(e),V=b(e,"BLOCKQUOTE",{class:!0,"data-svelte-h":!0}),j(V)!=="svelte-r1jcqf"&&(V.innerHTML=Kl),_s=c(e),M(ne.$$.fragment,e),Cs=c(e),oe=b(e,"P",{"data-svelte-h":!0}),j(oe)!=="svelte-1ti3vkt"&&(oe.innerHTML=et),Ns=c(e),ae=b(e,"P",{"data-svelte-h":!0}),j(ae)!=="svelte-1wm6erl"&&(ae.innerHTML=st),Is=c(e),M(pe.$$.fragment,e),Gs=c(e),ie=b(e,"P",{"data-svelte-h":!0}),j(ie)!=="svelte-rijkrp"&&(ie.textContent=lt),$s=c(e),M(ce.$$.fragment,e),ks=c(e),re=b(e,"P",{"data-svelte-h":!0}),j(re)!=="svelte-11hzy0w"&&(re.innerHTML=tt),Ws=c(e),me=b(e,"P",{"data-svelte-h":!0}),j(me)!=="svelte-13bvyeh"&&(me.innerHTML=nt),Xs=c(e),M(de.$$.fragment,e),zs=c(e),Me=b(e,"P",{"data-svelte-h":!0}),j(Me)!=="svelte-1n3tsas"&&(Me.innerHTML=ot),Qs=c(e),M(R.$$.fragment,e),Vs=c(e),M(ue.$$.fragment,e),Rs=c(e),ye=b(e,"P",{"data-svelte-h":!0}),j(ye)!=="svelte-1qs318i"&&(ye.innerHTML=at),Es=c(e),M(E.$$.fragment,e),Hs=c(e),M(Je.$$.fragment,e),Ss=c(e),we=b(e,"P",{"data-svelte-h":!0}),j(we)!=="svelte-1r5qgcf"&&(we.innerHTML=pt),As=c(e),M(H.$$.fragment,e),xs=c(e),M(he.$$.fragment,e),qs=c(e),be=b(e,"P",{"data-svelte-h":!0}),j(be)!=="svelte-owjwrf"&&(be.innerHTML=it),Ys=c(e),M(S.$$.fragment,e),Fs=c(e),M(fe.$$.fragment,e),Ls=c(e),je=b(e,"P",{"data-svelte-h":!0}),j(je)!=="svelte-1jkf232"&&(je.innerHTML=ct),Ps=c(e),M(A.$$.fragment,e),Ds=c(e),M(Ue.$$.fragment,e),Os=c(e),Te=b(e,"P",{"data-svelte-h":!0}),j(Te)!=="svelte-78nlel"&&(Te.innerHTML=rt),Ks=c(e),ve=b(e,"P",{"data-svelte-h":!0}),j(ve)!=="svelte-4uszwp"&&(ve.innerHTML=mt),el=c(e),M(x.$$.fragment,e),sl=c(e),M(Ze.$$.fragment,e),ll=c(e),ge=b(e,"P",{"data-svelte-h":!0}),j(ge)!=="svelte-1ggwg7d"&&(ge.innerHTML=dt),tl=c(e),Be=b(e,"P",{"data-svelte-h":!0}),j(Be)!=="svelte-f96g6"&&(Be.innerHTML=Mt),nl=c(e),M(_e.$$.fragment,e),ol=c(e),Ce=b(e,"P",{"data-svelte-h":!0}),j(Ce)!=="svelte-bdkgaq"&&(Ce.textContent=ut),al=c(e),M(q.$$.fragment,e),pl=c(e),M(Ne.$$.fragment,e),il=c(e),Ie=b(e,"P",{"data-svelte-h":!0}),j(Ie)!=="svelte-13gle33"&&(Ie.innerHTML=yt),cl=c(e),Ge=b(e,"P",{"data-svelte-h":!0}),j(Ge)!=="svelte-2xtold"&&(Ge.innerHTML=Jt),rl=c(e),M($e.$$.fragment,e),ml=c(e),ke=b(e,"P",{"data-svelte-h":!0}),j(ke)!=="svelte-1326i2u"&&(ke.innerHTML=wt),dl=c(e),M(We.$$.fragment,e),Ml=c(e),Xe=b(e,"P",{"data-svelte-h":!0}),j(Xe)!=="svelte-2cm2bd"&&(Xe.innerHTML=ht),ul=c(e),ze=b(e,"P",{"data-svelte-h":!0}),j(ze)!=="svelte-10ubvxy"&&(ze.textContent=bt),yl=c(e),M(Qe.$$.fragment,e),Jl=c(e),Ve=b(e,"P",{"data-svelte-h":!0}),j(Ve)!=="svelte-fkq18s"&&(Ve.textContent=ft),wl=c(e),Re=b(e,"P",{"data-svelte-h":!0}),j(Re)!=="svelte-jq9awm"&&(Re.innerHTML=jt),hl=c(e),M(Ee.$$.fragment,e),bl=c(e),He=b(e,"P",{"data-svelte-h":!0}),j(He)!=="svelte-ze8d61"&&(He.innerHTML=Ut),fl=c(e),M(Se.$$.fragment,e),jl=c(e),Ae=b(e,"P",{"data-svelte-h":!0}),j(Ae)!=="svelte-10arwsi"&&(Ae.innerHTML=Tt),Ul=c(e),M(xe.$$.fragment,e),Tl=c(e),M(qe.$$.fragment,e),vl=c(e),N=b(e,"DIV",{class:!0});var $=F(N);M(Ye.$$.fragment,$),$l=c($),k=b($,"DIV",{class:!0});var X=F(k);M(Fe.$$.fragment,X),kl=c(X),as=b(X,"P",{"data-svelte-h":!0}),j(as)!=="svelte-1bfskfo"&&(as.innerHTML=vt),Wl=c(X),ps=b(X,"P",{"data-svelte-h":!0}),j(ps)!=="svelte-13hldjp"&&(ps.textContent=Zt),X.forEach(o),Xl=c($),Y=b($,"DIV",{class:!0});var ts=F(Y);M(Le.$$.fragment,ts),zl=c(ts),is=b(ts,"P",{"data-svelte-h":!0}),j(is)!=="svelte-1r3tkns"&&(is.textContent=gt),ts.forEach(o),Ql=c($),B=b($,"DIV",{class:!0});var I=F(B);M(Pe.$$.fragment,I),Vl=c(I),cs=b(I,"P",{"data-svelte-h":!0}),j(cs)!=="svelte-1g5f9fd"&&(cs.textContent=Bt),Rl=c(I),rs=b(I,"P",{"data-svelte-h":!0}),j(rs)!=="svelte-1nulle2"&&(rs.textContent=_t),El=c(I),ms=b(I,"P",{"data-svelte-h":!0}),j(ms)!=="svelte-1p5xsfp"&&(ms.innerHTML=Ct),Hl=c(I),ds=b(I,"P",{"data-svelte-h":!0}),j(ds)!=="svelte-41qm80"&&(ds.innerHTML=Nt),I.forEach(o),$.forEach(o),Zl=c(e),De=b(e,"UL",{"data-svelte-h":!0}),j(De)!=="svelte-1p6h59i"&&(De.innerHTML=It),gl=c(e),M(Oe.$$.fragment,e),Bl=c(e),G=b(e,"DIV",{class:!0});var z=F(G);M(Ke.$$.fragment,z),Sl=c(z),Ms=b(z,"P",{"data-svelte-h":!0}),j(Ms)!=="svelte-1vmlzw9"&&(Ms.textContent=Gt),Al=c(z),us=b(z,"P",{"data-svelte-h":!0}),j(us)!=="svelte-8770m5"&&(us.innerHTML=$t),z.forEach(o),_l=c(e),M(es.$$.fragment,e),Cl=c(e),W=b(e,"DIV",{class:!0});var ns=F(W);M(ss.$$.fragment,ns),xl=c(ns),ys=b(ns,"P",{"data-svelte-h":!0}),j(ys)!=="svelte-1i23v48"&&(ys.innerHTML=kt),ns.forEach(o),Nl=c(e),M(ls.$$.fragment,e),Il=c(e),ws=b(e,"P",{}),F(ws).forEach(o),this.h()},h(){Q(l,"name","hf:doc:metadata"),Q(l,"content",mn),Q(V,"class","tip"),Q(k,"class","docstring border-l-2 border-t-2 pl-4 pt-3.5 border-gray-100 rounded-tl-xl mb-6 mt-8"),Q(Y,"class","docstring border-l-2 border-t-2 pl-4 pt-3.5 border-gray-100 rounded-tl-xl mb-6 mt-8"),Q(B,"class","docstring border-l-2 border-t-2 pl-4 pt-3.5 border-gray-100 rounded-tl-xl mb-6 mt-8"),Q(N,"class","docstring border-l-2 border-t-2 pl-4 pt-3.5 border-gray-100 rounded-tl-xl mb-6 mt-8"),Q(G,"class","docstring border-l-2 border-t-2 pl-4 pt-3.5 border-gray-100 rounded-tl-xl mb-6 mt-8"),Q(W,"class","docstring border-l-2 border-t-2 pl-4 pt-3.5 border-gray-100 rounded-tl-xl mb-6 mt-8")},m(e,t){T(document.head,l),a(e,p,t),a(e,s,t),a(e,r,t),u(n,e,t),a(e,m,t),a(e,U,t),a(e,hs,t),a(e,L,t),a(e,bs,t),a(e,P,t),a(e,fs,t),u(D,e,t),a(e,js,t),a(e,O,t),a(e,Us,t),a(e,K,t),a(e,Ts,t),a(e,ee,t),a(e,vs,t),a(e,se,t),a(e,Zs,t),u(le,e,t),a(e,gs,t),a(e,te,t),a(e,Bs,t),a(e,V,t),a(e,_s,t),u(ne,e,t),a(e,Cs,t),a(e,oe,t),a(e,Ns,t),a(e,ae,t),a(e,Is,t),u(pe,e,t),a(e,Gs,t),a(e,ie,t),a(e,$s,t),u(ce,e,t),a(e,ks,t),a(e,re,t),a(e,Ws,t),a(e,me,t),a(e,Xs,t),u(de,e,t),a(e,zs,t),a(e,Me,t),a(e,Qs,t),u(R,e,t),a(e,Vs,t),u(ue,e,t),a(e,Rs,t),a(e,ye,t),a(e,Es,t),u(E,e,t),a(e,Hs,t),u(Je,e,t),a(e,Ss,t),a(e,we,t),a(e,As,t),u(H,e,t),a(e,xs,t),u(he,e,t),a(e,qs,t),a(e,be,t),a(e,Ys,t),u(S,e,t),a(e,Fs,t),u(fe,e,t),a(e,Ls,t),a(e,je,t),a(e,Ps,t),u(A,e,t),a(e,Ds,t),u(Ue,e,t),a(e,Os,t),a(e,Te,t),a(e,Ks,t),a(e,ve,t),a(e,el,t),u(x,e,t),a(e,sl,t),u(Ze,e,t),a(e,ll,t),a(e,ge,t),a(e,tl,t),a(e,Be,t),a(e,nl,t),u(_e,e,t),a(e,ol,t),a(e,Ce,t),a(e,al,t),u(q,e,t),a(e,pl,t),u(Ne,e,t),a(e,il,t),a(e,Ie,t),a(e,cl,t),a(e,Ge,t),a(e,rl,t),u($e,e,t),a(e,ml,t),a(e,ke,t),a(e,dl,t),u(We,e,t),a(e,Ml,t),a(e,Xe,t),a(e,ul,t),a(e,ze,t),a(e,yl,t),u(Qe,e,t),a(e,Jl,t),a(e,Ve,t),a(e,wl,t),a(e,Re,t),a(e,hl,t),u(Ee,e,t),a(e,bl,t),a(e,He,t),a(e,fl,t),u(Se,e,t),a(e,jl,t),a(e,Ae,t),a(e,Ul,t),u(xe,e,t),a(e,Tl,t),u(qe,e,t),a(e,vl,t),a(e,N,t),u(Ye,N,null),T(N,$l),T(N,k),u(Fe,k,null),T(k,kl),T(k,as),T(k,Wl),T(k,ps),T(N,Xl),T(N,Y),u(Le,Y,null),T(Y,zl),T(Y,is),T(N,Ql),T(N,B),u(Pe,B,null),T(B,Vl),T(B,cs),T(B,Rl),T(B,rs),T(B,El),T(B,ms),T(B,Hl),T(B,ds),a(e,Zl,t),a(e,De,t),a(e,gl,t),u(Oe,e,t),a(e,Bl,t),a(e,G,t),u(Ke,G,null),T(G,Sl),T(G,Ms),T(G,Al),T(G,us),a(e,_l,t),u(es,e,t),a(e,Cl,t),a(e,W,t),u(ss,W,null),T(W,xl),T(W,ys),a(e,Nl,t),u(ls,e,t),a(e,Il,t),a(e,ws,t),Gl=!0},p(e,[t]){const $={};t&2&&($.$$scope={dirty:t,ctx:e}),R.$set($);const X={};t&2&&(X.$$scope={dirty:t,ctx:e}),E.$set(X);const ts={};t&2&&(ts.$$scope={dirty:t,ctx:e}),H.$set(ts);const I={};t&2&&(I.$$scope={dirty:t,ctx:e}),S.$set(I);const z={};t&2&&(z.$$scope={dirty:t,ctx:e}),A.$set(z);const ns={};t&2&&(ns.$$scope={dirty:t,ctx:e}),x.$set(ns);const Wt={};t&2&&(Wt.$$scope={dirty:t,ctx:e}),q.$set(Wt)},i(e){Gl\|\|(y(n.$$.fragment,e),y(D.$$.fragment,e),y(le.$$.fragment,e),y(ne.$$.fragment,e),y(pe.$$.fragment,e),y(ce.$$.fragment,e),y(de.$$.fragment,e),y(R.$$.fragment,e),y(ue.$$.fragment,e),y(E.$$.fragment,e),y(Je.$$.fragment,e),y(H.$$.fragment,e),y(he.$$.fragment,e),y(S.$$.fragment,e),y(fe.$$.fragment,e),y(A.$$.fragment,e),y(Ue.$$.fragment,e),y(x.$$.fragment,e),y(Ze.$$.fragment,e),y(_e.$$.fragment,e),y(q.$$.fragment,e),y(Ne.$$.fragment,e),y($e.$$.fragment,e),y(We.$$.fragment,e),y(Qe.$$.fragment,e),y(Ee.$$.fragment,e),y(Se.$$.fragment,e),y(xe.$$.fragment,e),y(qe.$$.fragment,e),y(Ye.$$.fragment,e),y(Fe.$$.fragment,e),y(Le.$$.fragment,e),y(Pe.$$.fragment,e),y(Oe.$$.fragment,e),y(Ke.$$.fragment,e),y(es.$$.fragment,e),y(ss.$$.fragment,e),y(ls.$$.fragment,e),Gl=!0)},o(e){J(n.$$.fragment,e),J(D.$$.fragment,e),J(le.$$.fragment,e),J(ne.$$.fragment,e),J(pe.$$.fragment,e),J(ce.$$.fragment,e),J(de.$$.fragment,e),J(R.$$.fragment,e),J(ue.$$.fragment,e),J(E.$$.fragment,e),J(Je.$$.fragment,e),J(H.$$.fragment,e),J(he.$$.fragment,e),J(S.$$.fragment,e),J(fe.$$.fragment,e),J(A.$$.fragment,e),J(Ue.$$.fragment,e),J(x.$$.fragment,e),J(Ze.$$.fragment,e),J(_e.$$.fragment,e),J(q.$$.fragment,e),J(Ne.$$.fragment,e),J($e.$$.fragment,e),J(We.$$.fragment,e),J(Qe.$$.fragment,e),J(Ee.$$.fragment,e),J(Se.$$.fragment,e),J(xe.$$.fragment,e),J(qe.$$.fragment,e),J(Ye.$$.fragment,e),J(Fe.$$.fragment,e),J(Le.$$.fragment,e),J(Pe.$$.fragment,e),J(Oe.$$.fragment,e),J(Ke.$$.fragment,e),J(es.$$.fragment,e),J(ss.$$.fragment,e),J(ls.$$.fragment,e),Gl=!1},d(e){e&&(o(p),o(s),o(r),o(m),o(U),o(hs),o(L),o(bs),o(P),o(fs),o(js),o(O),o(Us),o(K),o(Ts),o(ee),o(vs),o(se),o(Zs),o(gs),o(te),o(Bs),o(V),o(_s),o(Cs),o(oe),o(Ns),o(ae),o(Is),o(Gs),o(ie),o($s),o(ks),o(re),o(Ws),o(me),o(Xs),o(zs),o(Me),o(Qs),o(Vs),o(Rs),o(ye),o(Es),o(Hs),o(Ss),o(we),o(As),o(xs),o(qs),o(be),o(Ys),o(Fs),o(Ls),o(je),o(Ps),o(Ds),o(Os),o(Te),o(Ks),o(ve),o(el),o(sl),o(ll),o(ge),o(tl),o(Be),o(nl),o(ol),o(Ce),o(al),o(pl),o(il),o(Ie),o(cl),o(Ge),o(rl),o(ml),o(ke),o(dl),o(Ml),o(Xe),o(ul),o(ze),o(yl),o(Jl),o(Ve),o(wl),o(Re),o(hl),o(bl),o(He),o(fl),o(jl),o(Ae),o(Ul),o(Tl),o(vl),o(N),o(Zl),o(De),o(gl),o(Bl),o(G),o(_l),o(Cl),o(W),o(Nl),o(Il),o(ws)),o(l),w(n,e),w(D,e),w(le,e),w(ne,e),w(pe,e),w(ce,e),w(de,e),w(R,e),w(ue,e),w(E,e),w(Je,e),w(H,e),w(he,e),w(S,e),w(fe,e),w(A,e),w(Ue,e),w(x,e),w(Ze,e),w(_e,e),w(q,e),w(Ne,e),w($e,e),w(We,e),w(Qe,e),w(Ee,e),w(Se,e),w(xe,e),w(qe,e),w(Ye),w(Fe),w(Le),w(Pe),w(Oe,e),w(Ke),w(es,e),w(ss),w(ls,e)}}}const mn='{"title":"Cosmos 3","local":"cosmos-3","sections":[{"title":"What’s new in Cosmos 3","local":"whats-new-in-cosmos-3","sections":[],"depth":2},{"title":"Available checkpoints","local":"available-checkpoints","sections":[],"depth":2},{"title":"Prompt upsampling","local":"prompt-upsampling","sections":[],"depth":2},{"title":"Text-to-video","local":"text-to-video","sections":[],"depth":2},{"title":"Text-to-image","local":"text-to-image","sections":[],"depth":2},{"title":"Image-to-video","local":"image-to-video","sections":[],"depth":2},{"title":"Video-to-video","local":"video-to-video","sections":[],"depth":2},{"title":"Video-to-video with sound","local":"video-to-video-with-sound","sections":[],"depth":2},{"title":"Text-to-video with sound","local":"text-to-video-with-sound","sections":[],"depth":2},{"title":"Action-conditioned generation","local":"action-conditioned-generation","sections":[{"title":"Action policy","local":"action-policy","sections":[],"depth":3}],"depth":2},{"title":"Metadata templates","local":"metadata-templates","sections":[],"depth":2},{"title":"Safety checker","local":"safety-checker","sections":[],"depth":2},{"title":"Cosmos3OmniPipeline","local":"diffusers.Cosmos3OmniPipeline","sections":[],"depth":2},{"title":"CosmosActionCondition","local":"diffusers.CosmosActionCondition","sections":[],"depth":2},{"title":"Cosmos3OmniPipelineOutput","local":"diffusers.pipelines.cosmos.pipeline_cosmos3_omni.Cosmos3OmniPipelineOutput","sections":[],"depth":2}],"depth":1}';function dn(f){return zt(()=>{new URLSearchParams(window.location.search).get("fw")}),[]}class bn extends Qt{constructor(l){super(),Vt(this,l,dn,rn,Xt,{})}}export{bn as component};

Xet Storage Details

Size:: 107 kB
Xet hash:: 04090b1f8099b668a36171f4fd0470abdbc94c9ce601ab34e7c5c20bd7761ce5

Xet efficiently stores files, intelligently splitting them into unique chunks and accelerating uploads and downloads. More info.