Shiwei Zhang

Wan2.1 Algorithm Engineer 🎥🔧 | Passionate about AI and data-driven innovation!

[Wan @Alibaba_Wan]

What a fantastic milestone! Thanks for the shoutout, @arena . The entire @Alibaba_Wan team is absolutely thrilled.
A huge amount of hard work and dedication went into the Wan 2.5 Preview, and seeing this kind of reception from the community is incredibly rewarding for all our researchers and engineers.
This is just the preview, and we're more excited than ever to continue pushing forward. Thanks again to the entire community for the incredible support!

[Shiwei Zhang @StevenZhang66]

Welcome! Look out for our more powerful models.🚀🚀🚀

[Arena.ai @arena]

🎬 🖼️ Video & Image Leaderboard Update

Wan 2.5 preview models i2v and t2i have landed in the Top 5 on the Image-to-Video and Text-to-Image leaderboards.

🔹Wan 2.5-i2v-preview ranks #3 on Image-to-Video
🔹Wan 2.5-t2i-preview ranks #5 on Text-to-Image

Congrats to the @Alibaba_Wan 2.5 team for this incredible milestone! 👏

[852話(hakoniwa) @8co28]

Wan2.5リリース記念の配信にお付き合いいただきありがとうございました。配信時点でカクカクで流せなかった動画を改めていくつか貼ります。
・パルクール（接地がポイント）
・複数人アクション（コレまでサービス向きだった）
・時間経過と天候変化（変化が残り続ける）
・複数ショット（一貫性）

[Ring Hyacinth @ring_hyacinth]

Wan2.5 实测！可以用音频驱动画面生成~！

用《夜上海》的音频驱动，做了一个博斯风格的上海小动画。我肯定是加班加多了...

目前夸克的「造点 AI」上可以 7 天免费玩 WAN 2.5~!

[WaytoAGI｜通往AGI之路 @WaytoAGI]

9月24日阿里夸克发布了AI创作平台“造点”，支持AI 生图和AI生视频两大能力，并且率先接入通义万相Wan2.5，甚至直接开放了 7 天免费体验！

这是国内首个支持音画同步视频生成的平台，生图方面，造点AI还引入了图像生成模型Midjourney V7。一个夸克「造点AI」= 生图 + 生视频 + 加配乐 + 对口型 +P 图编辑，太丝滑了！

[Ramón Teleco @ramonteleco]

Teniendo en cuenta que esto está generado con Wan2.5 y está al nivel de Veo3, a Open AI se le puede complicar la película como Sora2 no sea algo espectacular

[梦影Erislia @withheldglow]

Wan2.5 test

[Shiwei Zhang @StevenZhang66]

Wan2.5

[Shiwei Zhang @StevenZhang66]

By Wan2.5

[Ofir Brown @OfirBrown]

@Alibaba_Wan Wan 2.5 is impressive🤩

[Shiwei Zhang @StevenZhang66]

You can always count on Wan! (pic from a nice netizen)

[TTPlanet @ttplanet]

Wan2.2 The quality is significantly improved, tested some human body performance. Amazing
@Alibaba_Wan #wan2.2

[ComfyUI @ComfyUI]

Wan2.2 is now natively supported in ComfyUI on Day 0!

🔹 A next-gen video model with MoE (Mixture of Experts) architecture with dual noise experts, under Apache 2.0 license!
- Cinematic-level Aesthetic Control
- Large-scale Complex Motion
- Precise Semantic Compliance

📚 Versions available:
- Wan2.2-TI2V-5B: FP16
- Wan2.2-I2V-14B: FP16/FP8
- Wan2.2-T2V-14B: FP16/FP8

💻 Down to 8GB VRAM requirement for the 5B version with ComfyUI auto-offloading.

[Wan @Alibaba_Wan]

Ready to be amazed? 🚀
Our new concept film is out! See how Wan2.2 makes life bloom in every frame.
See it for yourself! 👇

[Wan @Alibaba_Wan]

🚨 https://t.co/6Ceb2mGcUF just leveled up. Starting April 23, we’re entering the commercial era—with a twist: Still free. Now faster. Built for creators.
🎥 All users can have access to Free Unlimited Generations：
Relax mode = totally free + unlimited free generations
Credit mode = totally free + priority processing
⚡ Go Pro for Speed & Power: Unlock priority processing, watermark-free downloads, more concurrent tasks, access to all advanced features, including image style palettes, Start/End Frames and more coming soon!
🔥 Flash Perks: Daily login grants 10 Credits; new Pro/Premium users get additional gift credits+ exclusive tools during launch.
Join us: no limits, no waiting—just pure creative freedom.
👉 https://t.co/GDM97nlMMM

[ComfyUI @ComfyUI]

ComfyUI now natively supports @Alibaba_wan Wan2.1 FLF2V, a First-Last-Frame video generation model with 720P generation capability

We provide the fp16 repackaged model for users to run on local GPU. Checkout the workflow and some great examples 👇

[Shiwei Zhang @StevenZhang66]

I'm delighted to share that our Wan2.1 GitHub repo has received over 1w stars, and the model has been downloaded more than 2 million times on HuggingFace and Modelscope. This is a great motivation for us to keep striving. We'll continue to release more downstream models and powerful foundational models next. @Alibaba_Wan

#videogeneration #opensource

[Shiwei Zhang @StevenZhang66]

@Neverthelesser @Alibaba_Wan We will take a deep look at the corruption issue. Could you please provide your environment details (such as torch and flash attention versions, GPU information, inference commands, etc.), the prompt, and the seed so we can reproduce this issue?

[Shiwei Zhang @StevenZhang66]

Our Wan2.1 video generation model will keep updating regularly. Let me know what you're most excited about – we're committed to listening to your expectations! @Alibaba_Wan
#VideoGeneration #OpenSource

[Shiwei Zhang @StevenZhang66]

We have retrained the UniAnimate model based on Wan2.1-14B-I2V recently. And we've released the report and open-sourced the training, inference code, and the models. You're welcome to try it out.

Github: https://t.co/vpc2TSPW5s
Report: https://t.co/7cNk2wsvqA