Wan: Open and Advanced Large-Scale Video Generative Models

#ai #deeplearning #computerscience #machinelearning

Wan — Open, Large-Scale Video Generator for Everyone

Meet Wan, a simple to use set of models that turns pictures, text or edits into short videos.
The team made two sizes so you can pick speed or power, a small fast 1.
3B and a big, strong 14B model trained on billions of clips.
Together they push what video generation can do, giving smoother motion and cleaner edits than many free tools and some paid ones.
The smaller model runs on typical home cards, it only needs 8.
19 GB so many people can try it on consumer GPUs.
Wan shares all code and models, it's fully open-source, meaning creators, students and hobbyists can download and play.
The team also built simple checks to judge quality and used huge collections of clips to teach the models, though you don't need to know that to start.
Imagine editing a clip from a single photo or following a short instruction to change a scene — this widens real creative possibilities for everyone.
Try it, tinker, make something new.