Alibaba releases part of its image and video generation AI 'Wan 2.1' as open source



Four models from version 2.1 of Wan , an advanced and powerful visual generative model developed by Alibaba Group 's Tongyi Lab, have been open-sourced.

Alibaba makes AI video generation model free to use globally
https://www.cnbc.com/2025/02/26/alibaba-makes-ai-video-generation-model-free-to-use-globally.html

Alibaba offers free access to its AI model that can generate realistic video and images
https://www.engadget.com/ai/alibaba-offers-free-access-to-its-ai-model-that-can-generate-realistic-video-and-images-133045633.html

Alibaba makes AI model for video, image generation publicly available | Reuters
https://www.reuters.com/technology/artificial-intelligence/alibaba-release-open-source-version-video-generating-ai-model-2025-02-25/

On February 26, 2025 local time, Alibaba open-sourced four of the Wan 2.1 models that it had announced it would open source. Initially, Alibaba called this model 'WanX 2.1', but later shortened the name to 'Wan 2.1'. This time, Alibaba open-sourced four models of Wan 2.1: 'T2V-1.3B', 'T2V-14B', 'I2V-14B-720P', and 'I2V-14B-480P'. The '14B' in the model name indicates that the number of parameters is 14 billion.




Wan 2.1 achieved top results in VBench , a comprehensive benchmark of video generation models, and appears to be particularly strong in features such as multi-object interaction.

The open source models are available through Alibaba Cloud's ModelScope and Hugging Face, a known large repository of AI models.

Wan-AI (Wan-AI)
https://huggingface.co/Wan-AI



Wan 2.1 has been open-sourced, so the source code is available on GitHub.

GitHub - Wan-Video/Wan2.1: Wan: Open and Advanced Large-Scale Video
Generative Models

https://github.com/Wan-Video/Wan2.1



In January 2025, China's DeepSeek announced the R1 inference model, which attracted attention because its training cost was only 3% of that of OpenAI's o1 inference model. Since then, open source AI models have been attracting attention from around the world.

How did DeepSeek surpass O1 at 3% the cost of OpenAI? - GIGAZINE



The open source model doesn't generate direct revenue for the company like OpenAI's closed model does, but open sourcing technology can serve a variety of purposes, including driving innovation and building a community around the product.

In particular, Chinese companies are promoting the open source model, and the Alibaba and DeepSeek models are said to be the most widely used models in the world . Alibaba released its first open source model in August 2023.

Outside of China, Meta is promoting open source AI models.

Meta releases AI model 'Llama 3.3', 70B model demonstrates performance comparable to Llama 3.1's 405B model - GIGAZINE



Related Posts:

in Software, Posted by logu_ii