Alibaba launches new open-source AI tools for video creators

Alibaba has launched Wan2.2, a new series of open-source video generation models using Mixture-of-Experts (MoE) architecture.

The lineup includes Wan2.2-T2V-A14B for text-to-video, Wan2.2-I2V-A14B for image-to-video, and Wan2.2-TI2V-5B, a hybrid model supporting both.

These models offer detailed control over elements like lighting, color tone, and camera angles.

The MoE architecture boosts efficiency by activating only 14 billion of 27 billion parameters at each step, cutting computational demands by up to 50%.

Wan2.2-TI2V-5B is optimized to generate 720p videos on standard consumer GPUs.

Trained on a larger dataset, Wan2.2 delivers better results in complex scenes and motions.

Developers can access the models via Hugging Face, GitHub, and Alibaba Cloud’s ModelScope.

Earlier models, Wan2.1 and Wan 2.1-VACE, have already seen over 5.4 million downloads.

🔗 Source: Alibaba

🧠 Food for thought

1️⃣ Open-source strategy drives massive adoption in competitive AI market

Alibaba’s consistent open-source approach is generating significant developer traction across their AI model portfolio.

Their Qwen 2.5 series alone attracted over 40 million downloads since launch, while the Wan video generation models specifically reached 5.4 million downloads 1. This represents substantial adoption compared to proprietary alternatives from competitors like OpenAI.

The company has systematically released over 100 open-source AI models in recent months, spanning everything from language models to video generation tools 2. This open-source strategy appears designed to build developer ecosystems and market share while competitors like OpenAI maintain closed, paid access models.

CEO Eddie Wu has explicitly stated this reflects “aggressive investment in AI R&D” to enhance competitive positioning against rivals like Microsoft and Baidu 1. The download numbers suggest this approach is successfully attracting developers who might otherwise use proprietary solutions.