Alibaba open-sources Wan2.1 AI video model that rivals Sora
Alibaba's Tongyi Lab just released Wan2.1, an open-source video generation suite that beats top models like Sora on key benchmarks while running 2.5x faster. It supports bilingual text rendering and video-to-audio, and includes a lightweight 1.3B model that runs locally on an RTX 4090 GPU.

Wan2.1 is Alibaba's newly open-sourced AI video generation suite, designed to turn text prompts and static images into fluid, dynamic video clips right on your own machine.
The suite recently took the number one spot on the VBench benchmark with an overall score of 86.22%, beating out top proprietary competitors like OpenAI's Sora. It is currently the only open-source model sitting in VBench's top five.
What it can do
Alibaba's Tongyi Lab released four dedicated models under the suite: T2V-14B and T2V-1.3B for text-to-video, alongside I2V-14B-720P and I2V-14B-480P for image-to-video generation.
Across the board, Wan2.1 generates video roughly 2.5x faster than comparable tools while virtually wiping out telltale AI artifacts and choppy motions. It handles complex physics, realistic lighting, and heavy motion dynamics with surprising consistency.
It is also the first video model capable of rendering legible text in both English and Chinese directly within the scenes. On top of that, it packs editing utilities like inpainting, outpainting, multi-image referencing, and reliable character consistency across shots.
Specs and running it locally
Instead of locking everything behind a cloud API, Alibaba built Wan2.1 to actually run on consumer hardware.
The lightweight 1.3B model lets you generate a 5-second 480P video clip in about 4 minutes on a single desktop RTX 4090 or a standard laptop. If you have enterprise GPUs, the full 14B parameter models provide top-tier production visual fidelity and take any standard image dimension as input.
All weights and inference code are freely available right now on Hugging Face and Alibaba's ModelScope platform.
Why it matters
While frontier labs keep their flagship video models locked behind paid subscriptions and strict access tiers, open source is catching up faster than expected.
Wan2.1 proves that high-end motion fidelity and coherent physics do not have to stay inside walled gardens. For indie creators, developers, and VFX studios looking to build custom pipelines without cloud fees, this release is easily the most compelling tool to test this week.
Sources (2)
- wanxai.com(opens in a new tab)Originalwanxai.com
- alibabagroup.com(opens in a new tab)Reportalibabagroup.com
Topic trackerAlibaba Wan2.1 video model series
- Alibaba open-sources 14B video model that bridges start and end frames
- Alibaba open-sources Wan2.1 AI video model that rivals SoraThis story