Alibaba's Open-Source Wan Video Model Tops VBench: First-Hand Test Results!
Alibaba’s Wan video generation model sets a new benchmark with open-source capabilities, excelling in complex motion, instruction adherence, and physics modeling.
"AI Disruption" publication New Year 30% discount link.
Qwen+Wan, Alibaba Achieves Full-Modal Open Source
While many AI companies are still uncertain about whether to adopt open-source models, Alibaba's technical team has released a new model—Wan, a video generation model (including full inference code and weights, with the loosest open-source license).
Those familiar with video generation models know that many current models still face various challenges, such as difficulty in replicating complex character movements, struggling with physical changes during object interactions, or "selectively following" long-text instructions.
If a model performs well in all three areas, the model developers are usually reluctant to open-source it.