Wan2.2-Lightning / README.md
FX-FeiHou's picture
Update README.md
6cd92d0 verified
|
raw
history blame
5.7 kB
<!-- Telegram and Civitai Buttons -->
<p align="center" style="margin-top: 20px; margin-bottom: 20px; display:flex; justify-content:center; gap:20px; flex-wrap:wrap;">
<span style="text-align:center;">
<a href="https://t.me/wan22remix" target="_blank" rel="noopener">
<img src="https://img.shields.io/badge/Telegram-Group-1CA9E0?logo=telegram&logoColor=white&style=for-the-badge&rounded=true"
alt="Join Telegram Group"
style="width: 220px; height: 40px; border-radius:10px;"/>
</a>
<br/>
<span style="font-size: 16px; font-weight: bold; display: inline-block; margin-top: 8px;">
👉 Join our Telegram group
</span>
</span>
<span style="text-align:center;">
<a href="https://civitai.com/models/2003153" target="_blank" rel="noopener">
<img src="https://img.shields.io/badge/Civitai-View-0055FF?logo=googlescholar&logoColor=white&style=for-the-badge&rounded=true"
alt="View on Civitai"
style="width: 220px; height: 40px; border-radius:10px;"/>
</a>
<br/>
<span style="font-size: 16px; font-weight: bold; display: inline-block; margin-top: 8px;">
👉 View on Civitai
</span>
</span>
</p>
<section>
<h2>Model Overview — Wan2.2-Remix</h2>
<p>
<strong>Wan2.2-Remix</strong> is a <strong>research and creative model</strong> built to generate <strong>short, cinematic video clips</strong> directly from text prompts.
It focuses on <strong>human dynamics, realistic motion,</strong> and <strong>scene consistency</strong>, producing lifelike movement and smooth transitions without requiring LoRA setup.
</p>
<p>
This model comes in <strong>NSFW</strong> and <strong>SFW</strong> versions:<br/>
The <strong>NSFW version</strong> can generate high-quality adult content without loading LoRA, but it is <strong>not recommended for normal video generation</strong>;<br/>
For generating regular content, please use the <strong>SFW version</strong>.
</p>
<ul>
<li><strong>High-noise model T2V:</strong> <a href="https://huggingface.co/lightx2v/Wan2.2-Lightning/tree/main/Wan2.2-T2V-A14B-4steps-250928-dyno" target="_blank" rel="noopener">Lightx2v Wan2.2 Lightning dyno version</a> (quantized to <strong>fp8</strong>)</li>
<li><strong>High-noise model I2V:</strong> <a href="https://huggingface.co/Comfy-Org/Wan_2.2_ComfyUI_Repackaged/tree/main/split_files/diffusion_models" target="_blank" rel="noopener">Comfy-Org Wan 2.2 ComfyUI Repackaged</a> (fp16, quantized to <strong>fp8</strong>)</li>
<li><strong>Low-noise model:</strong> <a href="https://huggingface.co/Comfy-Org/Wan_2.2_ComfyUI_Repackaged/tree/main/split_files/diffusion_models" target="_blank" rel="noopener">Comfy-Org Wan 2.2 ComfyUI Repackaged</a> (quantized to <strong>fp8</strong>)</li>
<li><strong>Enhancements:</strong> Blended with open-source motion LoRA data and refined pose training for enhanced realism, anatomical precision, and gesture diversity.</li>
<li><strong>Integration:</strong> Fully compatible with <a href="https://huggingface.co/lightx2v/Wan2.2-Lightning" target="_blank" rel="noopener"><strong>lightx2v</strong></a>, no extra setup required.</li>
<li><strong>Recommended CLIP:</strong> <a href="https://huggingface.co/NSFW-API/NSFW-Wan-UMT5-XXL/tree/main" target="_blank" rel="noopener">NSFW-Wan-UMT5-XXL</a></li>
</ul>
<p>This model is currently in <strong>beta</strong>. It’s functional and improving—feedback is welcome!</p>
</section>
<hr/>
<section>
<h2>Changelog</h2>
<h4>I2V 0.8</h4>
<ul>
<li>Updated <a href="https://huggingface.co/lightx2v/Wan2.2-Lightning" target="_blank" rel="noopener"><strong>lighting acceleration</strong> module to <strong>Wan2.1 version</strong></a>, fixing slow-motion character issues</li>
<li>Integrated over <strong>20 open-source LoRAs</strong> and <strong>10 custom-trained LoRAs</strong> to enhance motion dynamics</li>
<li>Removed several <strong>low-quality LoRAs</strong> to improve overall visual fidelity</li>
</ul>
<hr style="margin:20px 0; border:1px dashed #aaa;"/>
<h4>I2V 0.6</h4>
<ul>
<li>Enhanced <strong>walking, dancing, and running</strong> movements</li>
<li>Improved <strong>overall motion coherence</strong></li>
<li>Introduced <strong>gesture-rich motion set</strong></li>
</ul>
<hr style="margin:20px 0; border:1px dashed #aaa;"/>
<h4>T2V 1.0 SFW</h4>
<ul>
<li>Optimized <strong>character motion naturalness</strong> for safe content generation</li>
<li>Improved motion fluidity and gesture diversity</li>
<li>Enhanced visual clarity and interaction scene quality</li>
</ul>
<hr style="margin:20px 0; border:1px dashed #aaa;"/>
<h4>T2V 1.0</h4>
<ul>
<li>Reduced <strong>dyno weighting</strong> to improve stability across different random seeds</li>
<li>Added more <strong>Asian facial datasets</strong> for greater diversity</li>
<li>Enhanced <strong>character motion dynamics</strong> for fluid animation</li>
<li>Improved <strong>motion amplitude, gesture diversity</strong>, and overall visual quality</li>
</ul>
<hr style="margin:20px 0; border:1px dashed #aaa;"/>
<h4>T2V 0.9</h4>
<ul>
<li>High-noise T2V model: <strong>Lightx2v Wan2.2 Lightning dyno version</strong> (fp8)</li>
<li>High-noise I2V model: <strong>Comfy-Org Wan 2.2 ComfyUI Repackaged</strong> (fp16, fp8 quantized)</li>
<li>Low-noise model: <strong>Comfy-Org Wan 2.2 ComfyUI Repackaged</strong></li>
<li>Refined <strong>character motion blending</strong> for smoother transitions</li>
<li>Enhanced <strong>anatomical articulation and natural body flow</strong></li>
</ul>
</section>