Wan 2.2 release date: July 28, 2025, and what the repo added

Wan 2.2's inference code and model weights were released on July 28, 2025, according to the Hugging Face model card. The repository then added more models: S2V-14B on Aug 26, 2025, Animate-14B on Sep 19, 2025, and a Diffusers integration for Animate-14B on Nov 13, 2025. The license in the README is Apache 2.0.

What is the timeline in the README?

What is in the release?

The model table lists T2V-A14B and I2V-A14B (480P and 720P), TI2V-5B (720P at 24 fps), S2V-14B and Animate-14B. The README's hardware lines are 80 GB minimum for the A14B models and 24 GB (an RTX 4090 is the example) for TI2V-5B. The card says the A14B design uses two experts for the high-noise and low-noise stages, and that training used 65.6 percent more images and 83.2 percent more videos than Wan2.1.

How old is that in model terms?

The weights are over a year old as of this post, and the README's last dated news item I read is from Nov 13, 2025. That does not make them unusable. Apache 2.0 weights that run on your own hardware stay available regardless of what the vendor releases next. It does mean newer hosted models may differ on quality, length and audio, so compare on your own prompts.

For a fixed pipeline where you control the version, an old open model has a real advantage: the output will not change under you. For a product that wants the latest quality, a hosted id changes when the provider changes it, so pin the id and re-test on each catalog update.

Does a newer Wan mean newer weights?

Not from these pages. They describe Wan 2.2 and nothing later, so do not assume later Wan names ship as downloads. On Sume, wan-3.0 is a hosted id that accepts 2 to 30 seconds; it is a different model from the Wan 2.2 weights, and the catalog code (read 2026-10-09) has no Wan 2.x row.

If you need a repeatable clip today without a GPU, the hosted ids are described in the video docs. If you want the weights, start from the README.