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&#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>