Spaces:
Running
Running
File size: 11,520 Bytes
b842027 0d93ed2 b842027 5f89124 b842027 0d93ed2 b842027 5f89124 b842027 5f89124 b842027 5f89124 b842027 5f89124 b842027 5f89124 b842027 5f89124 b842027 0d93ed2 4887ea5 | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 | <!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's Tongyi Lab: architecture, released checkpoints, and hosted API usage." />
<link rel="canonical" href="https://wavespeed.ai/collections/wan-2-2?utm_source=huggingface&utm_medium=space&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'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&utm_medium=space&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&utm_medium=space&utm_campaign=wan2_2" target="_blank" rel="noopener">wavespeed.ai ↗</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 "https://api.wavespeed.ai/api/v3/wavespeed-ai/wan-2.2/t2v-480p" \
-H "Authorization: Bearer $WAVESPEED_API_KEY" \
-H "Content-Type: application/json" \
-d '{
"prompt": "A paper boat drifting down a rain-slicked gutter at dusk, shallow depth of field, warm street lights",
"duration": 5,
"enable_sync_mode": false
}'
# -> {"code": 200, "data": {"id": "<request-id>", "status": "created", ...}}
# 2. poll until status is "completed"
curl "https://api.wavespeed.ai/api/v3/predictions/<request-id>/result" \
-H "Authorization: Bearer $WAVESPEED_API_KEY"
# -> {"code": 200, "data": {"status": "completed", "outputs": ["https://..."]}}</code></pre>
<pre id="run-1" role="tabpanel" hidden><code>import os, time, requests
API = "https://api.wavespeed.ai/api/v3"
KEY = os.environ["WAVESPEED_API_KEY"]
HEADERS = {"Authorization": f"Bearer {KEY}"}
# submit
res = requests.post(
f"{API}/wavespeed-ai/wan-2.2/t2v-480p",
headers={**HEADERS, "Content-Type": "application/json"},
json={
"prompt": "A paper boat drifting down a rain-slicked gutter at dusk, shallow depth of field, warm street lights",
"duration": 5,
"enable_sync_mode": false
},
timeout=30,
)
res.raise_for_status()
request_id = res.json()["data"]["id"]
# poll
while True:
data = requests.get(
f"{API}/predictions/{request_id}/result",
headers=HEADERS,
timeout=30,
).json()["data"]
if data["status"] == "completed":
print(data["outputs"][0])
break
if data["status"] == "failed":
raise RuntimeError(data.get("error", "generation failed"))
time.sleep(1.5)</code></pre>
<pre id="run-2" role="tabpanel" hidden><code>const API = "https://api.wavespeed.ai/api/v3";
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: "POST",
headers: { ...headers, "Content-Type": "application/json" },
body: JSON.stringify({
"prompt": "A paper boat drifting down a rain-slicked gutter at dusk, shallow depth of field, warm street lights",
"duration": 5,
"enable_sync_mode": 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 === "completed") {
console.log(data.outputs[0]);
break;
}
if (data.status === "failed") throw new Error(data.error ?? "generation failed");
await new Promise((r) => 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/<id>/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&utm_medium=space&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&utm_medium=space&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&utm_medium=space&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&utm_medium=space&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&utm_medium=space&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&utm_medium=space&utm_campaign=wan2_2" target="_blank" rel="noopener">WaveSpeed AI</a>
<a href="https://wavespeed.ai/docs?utm_source=huggingface&utm_medium=space&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>
|