wan2.2 / index.html
chengzeyi's picture
Add UTM attribution to outbound wavespeed.ai links
5f89124
Raw
History Blame Contribute Delete
11.5 kB
<!doctype html>
<html lang="en">
<head>
<meta charset="utf-8" />
<meta name="viewport" content="width=device-width, initial-scale=1" />
<title>Wan 2.2 — open large-scale video generative models</title>
<meta name="description" content="Reference for Wan 2.2, the open video generation model family from Alibaba&#x27;s Tongyi Lab: architecture, released checkpoints, and hosted API usage." />
<link rel="canonical" href="https://wavespeed.ai/collections/wan-2-2?utm_source=huggingface&amp;utm_medium=space&amp;utm_campaign=wan2_2" />
<meta property="og:type" content="website" />
<meta property="og:title" content="Wan 2.2 — open large-scale video generative models" />
<meta property="og:description" content="Reference for Wan 2.2, the open video generation model family from Alibaba&#x27;s Tongyi Lab: architecture, released checkpoints, and hosted API usage." />
<meta property="og:url" content="https://wavespeed.ai/collections/wan-2-2" />
<meta name="twitter:card" content="summary_large_image" />
<link rel="stylesheet" href="style.css" />
</head>
<body>
<header class="site-header">
<div class="wrap">
<a class="brand" href="https://wavespeed.ai?utm_source=huggingface&amp;utm_medium=space&amp;utm_campaign=wan2_2" target="_blank" rel="noopener">
<span class="brand-mark" aria-hidden="true"></span>
<span>WaveSpeed AI</span>
</a>
<nav class="header-nav">
<a class="jump opt" href="#architecture">Architecture</a>
<a class="jump opt" href="#checkpoints">Checkpoints</a>
<a class="jump" href="#run">Run it</a>
<a class="jump" href="#resources">Resources</a>
<a href="https://wavespeed.ai/collections/wan-2-2?utm_source=huggingface&amp;utm_medium=space&amp;utm_campaign=wan2_2" target="_blank" rel="noopener">wavespeed.ai &#8599;</a>
</nav>
</div>
</header>
<main class="wrap">
<div class="hero">
<p class="eyebrow">Alibaba · Tongyi Lab</p>
<h1>Wan 2.2</h1>
<p class="lede">An open family of large-scale video generative models covering text-to-video, image-to-video and a compact hybrid text/image-to-video checkpoint. Weights are published on Hugging Face under Apache-2.0.</p>
<ul class="meta">
<li><b>Developer</b> Alibaba Tongyi Lab</li>
<li><b>Task</b> text-to-video · image-to-video</li>
<li><b>License</b> Apache-2.0</li>
<li><b>Released</b> July 2025</li>
</ul>
</div>
<section id="architecture">
<h2>What changed in 2.2</h2>
<p class="section-note">Wan 2.2 revises the 2.1 architecture in three places. The claims below are the authors' own, taken from the model cards and technical report.</p>
<div class="grid">
<div class="card">
<h3>Mixture-of-experts denoiser</h3>
<p>The denoising trajectory is split across specialised expert models rather than one monolithic network, which raises total parameter count without a matching rise in per-step inference cost.</p>
</div>
<div class="card">
<h3>Curated aesthetic supervision</h3>
<p>Training data carries explicit labels for lighting, composition, contrast and colour tone, so cinematographic attributes can be steered from the prompt instead of emerging by chance.</p>
</div>
<div class="card">
<h3>Larger motion corpus</h3>
<p>The authors report training on 65.6% more images and 83.2% more video than Wan 2.1, aimed primarily at motion fidelity and prompt adherence.</p>
</div>
<div class="card">
<h3>High-compression VAE</h3>
<p>The TI2V-5B checkpoint pairs with a Wan2.2-VAE at a 16×16×4 compression ratio, which is what makes 720p/24fps generation practical at that model size.</p>
</div>
</div>
</section>
<section id="checkpoints">
<h2>Released checkpoints</h2>
<p class="section-note">All weights are on the Hugging Face Hub under the Wan-AI organisation.</p>
<div class="table-scroll">
<table>
<thead><tr><th>Checkpoint</th><th>Task</th><th>Params</th><th>Weights</th></tr></thead>
<tbody>
<tr><td><code>Wan2.2-T2V-A14B</code></td><td>Text-to-video</td><td>14B (MoE)</td><td><a href="https://huggingface.co/Wan-AI/Wan2.2-T2V-A14B" target="_blank" rel="noopener">base</a> · <a href="https://huggingface.co/Wan-AI/Wan2.2-T2V-A14B-Diffusers" target="_blank" rel="noopener">diffusers</a></td></tr>
<tr><td><code>Wan2.2-I2V-A14B</code></td><td>Image-to-video</td><td>14B (MoE)</td><td><a href="https://huggingface.co/Wan-AI/Wan2.2-I2V-A14B" target="_blank" rel="noopener">base</a> · <a href="https://huggingface.co/Wan-AI/Wan2.2-I2V-A14B-Diffusers" target="_blank" rel="noopener">diffusers</a></td></tr>
<tr><td><code>Wan2.2-TI2V-5B</code></td><td>Text + image-to-video, 720p/24fps</td><td>5B</td><td><a href="https://huggingface.co/Wan-AI/Wan2.2-TI2V-5B" target="_blank" rel="noopener">base</a> · <a href="https://huggingface.co/Wan-AI/Wan2.2-TI2V-5B-Diffusers" target="_blank" rel="noopener">diffusers</a></td></tr>
</tbody>
</table>
</div>
</section>
<section id="run">
<h2>Run it</h2>
<p class="section-note">If you would rather not provision GPUs, the same checkpoints are served as a hosted endpoint. Available variants: <code>wan-2.2/t2v-480p</code>, <code>t2v-720p</code>, <code>i2v-480p</code> and <code>i2v-720p</code>.</p>
<div class="code">
<div class="code-tabs" role="tablist">
<button type="button" role="tab" aria-selected="true" data-panel="run-0">cURL</button>
<button type="button" role="tab" aria-selected="false" data-panel="run-1">Python</button>
<button type="button" role="tab" aria-selected="false" data-panel="run-2">JavaScript</button>
</div>
<pre id="run-0" role="tabpanel"><code># 1. submit the job
curl -X POST &quot;https://api.wavespeed.ai/api/v3/wavespeed-ai/wan-2.2/t2v-480p&quot; \
-H &quot;Authorization: Bearer $WAVESPEED_API_KEY&quot; \
-H &quot;Content-Type: application/json&quot; \
-d &#x27;{
&quot;prompt&quot;: &quot;A paper boat drifting down a rain-slicked gutter at dusk, shallow depth of field, warm street lights&quot;,
&quot;duration&quot;: 5,
&quot;enable_sync_mode&quot;: false
}&#x27;
# -&gt; {&quot;code&quot;: 200, &quot;data&quot;: {&quot;id&quot;: &quot;&lt;request-id&gt;&quot;, &quot;status&quot;: &quot;created&quot;, ...}}
# 2. poll until status is &quot;completed&quot;
curl &quot;https://api.wavespeed.ai/api/v3/predictions/&lt;request-id&gt;/result&quot; \
-H &quot;Authorization: Bearer $WAVESPEED_API_KEY&quot;
# -&gt; {&quot;code&quot;: 200, &quot;data&quot;: {&quot;status&quot;: &quot;completed&quot;, &quot;outputs&quot;: [&quot;https://...&quot;]}}</code></pre>
<pre id="run-1" role="tabpanel" hidden><code>import os, time, requests
API = &quot;https://api.wavespeed.ai/api/v3&quot;
KEY = os.environ[&quot;WAVESPEED_API_KEY&quot;]
HEADERS = {&quot;Authorization&quot;: f&quot;Bearer {KEY}&quot;}
# submit
res = requests.post(
f&quot;{API}/wavespeed-ai/wan-2.2/t2v-480p&quot;,
headers={**HEADERS, &quot;Content-Type&quot;: &quot;application/json&quot;},
json={
&quot;prompt&quot;: &quot;A paper boat drifting down a rain-slicked gutter at dusk, shallow depth of field, warm street lights&quot;,
&quot;duration&quot;: 5,
&quot;enable_sync_mode&quot;: false
},
timeout=30,
)
res.raise_for_status()
request_id = res.json()[&quot;data&quot;][&quot;id&quot;]
# poll
while True:
data = requests.get(
f&quot;{API}/predictions/{request_id}/result&quot;,
headers=HEADERS,
timeout=30,
).json()[&quot;data&quot;]
if data[&quot;status&quot;] == &quot;completed&quot;:
print(data[&quot;outputs&quot;][0])
break
if data[&quot;status&quot;] == &quot;failed&quot;:
raise RuntimeError(data.get(&quot;error&quot;, &quot;generation failed&quot;))
time.sleep(1.5)</code></pre>
<pre id="run-2" role="tabpanel" hidden><code>const API = &quot;https://api.wavespeed.ai/api/v3&quot;;
const KEY = process.env.WAVESPEED_API_KEY;
const headers = { Authorization: `Bearer ${KEY}` };
// submit
const submit = await fetch(`${API}/wavespeed-ai/wan-2.2/t2v-480p`, {
method: &quot;POST&quot;,
headers: { ...headers, &quot;Content-Type&quot;: &quot;application/json&quot; },
body: JSON.stringify({
&quot;prompt&quot;: &quot;A paper boat drifting down a rain-slicked gutter at dusk, shallow depth of field, warm street lights&quot;,
&quot;duration&quot;: 5,
&quot;enable_sync_mode&quot;: false
}),
});
const { data: { id } } = await submit.json();
// poll
for (;;) {
const res = await fetch(`${API}/predictions/${id}/result`, { headers });
const { data } = await res.json();
if (data.status === &quot;completed&quot;) {
console.log(data.outputs[0]);
break;
}
if (data.status === &quot;failed&quot;) throw new Error(data.error ?? &quot;generation failed&quot;);
await new Promise((r) =&gt; setTimeout(r, 1500));
}</code></pre>
</div>
<div class="callout"><p>Requests are asynchronous: <code>POST</code> returns a request id, then you poll <code>/predictions/&lt;id&gt;/result</code> until <code>status</code> is <code>completed</code>. Set <code>enable_sync_mode: true</code> to have the call block and return outputs directly.</p><p>API keys are created in the <a href="https://wavespeed.ai/dashboard?utm_source=huggingface&amp;utm_medium=space&amp;utm_campaign=wan2_2" target="_blank" rel="noopener">WaveSpeed dashboard</a>.</p></div>
<div class="btn-row">
<a class="btn" href="https://wavespeed.ai/collections/wan-2-2?utm_source=huggingface&amp;utm_medium=space&amp;utm_campaign=wan2_2" target="_blank" rel="noopener">Open Wan 2.2 on WaveSpeed</a>
<a class="btn secondary" href="https://wavespeed.ai/docs?utm_source=huggingface&amp;utm_medium=space&amp;utm_campaign=wan2_2" target="_blank" rel="noopener">API reference</a>
</div>
</section>
<section id="resources">
<h2>Resources</h2>
<ul class="links">
<li><a href="https://huggingface.co/Wan-AI" target="_blank" rel="noopener"><span>Wan-AI on Hugging Face</span><span class="host">huggingface.co</span></a></li>
<li><a href="https://github.com/Wan-Video/Wan2.2" target="_blank" rel="noopener"><span>Wan2.2 on GitHub</span><span class="host">github.com</span></a></li>
<li><a href="https://wavespeed.ai/collections/wan-2-2?utm_source=huggingface&amp;utm_medium=space&amp;utm_campaign=wan2_2" target="_blank" rel="noopener"><span>Hosted endpoints</span><span class="host">wavespeed.ai</span></a></li>
<li><a href="https://wavespeed.ai/docs?utm_source=huggingface&amp;utm_medium=space&amp;utm_campaign=wan2_2" target="_blank" rel="noopener"><span>WaveSpeed API docs</span><span class="host">wavespeed.ai</span></a></li>
</ul>
</section>
</main>
<footer class="site-footer">
<div class="wrap">
<p>This page is a model reference maintained by WaveSpeed AI. The model itself is developed and released by its respective authors; trademarks belong to them. WaveSpeed AI provides hosted inference for it.</p>
<div class="footer-links">
<a href="https://wavespeed.ai?utm_source=huggingface&amp;utm_medium=space&amp;utm_campaign=wan2_2" target="_blank" rel="noopener">WaveSpeed AI</a>
<a href="https://wavespeed.ai/docs?utm_source=huggingface&amp;utm_medium=space&amp;utm_campaign=wan2_2" target="_blank" rel="noopener">Docs</a>
<a href="https://huggingface.co/wavespeed" target="_blank" rel="noopener">Hugging Face</a>
</div>
</div>
</footer>
<script src="tabs.js"></script>
</body>
</html>