Wan: video generator — what it does and what you need to run it

Wan is an open-source video generator from Alibaba that tackles the text-to-video task. The model converts a text description into a video sequence and, according to the developers, is capable of running on consumer graphics cards.

What it does

The authors describe Wan as an open and advanced large-scale model for video generation. The repository description states that the T2V-1.3B model generates a five-second video at 480P on an RTX 4090 in about four minutes without optimization techniques such as quantization. The developers note that the model's performance is comparable to some closed-source counterparts.

What you need to run it

• Platform and language: Python, the diffusers library.
• VRAM: as described by the author, the T2V-1.3B model requires 8.19 GB of VRAM.
• Model weights files: the largest weights file takes up 10.6 GB. The authors do not specify the total size of all weights files.
• License: the code and weights are distributed under Apache-2.0.
• Repository: github.com/Wan-Video/Wan2.1, created on February 25, 2025. The latest code change was on March 5, 2026.
• Model: huggingface.co/Wan-AI/Wan2.1-T2V-14B.
• Website: wan.video.

Who it suits

The tool may appeal to those looking for an open-source model for text-to-video generation that can run on local hardware. The stated VRAM consumption of 8.19 GB allows it to run on many consumer GPUs. The model is not suitable for those who need high-resolution video generation in a short time without additional optimization — the authors provide data only for 480P, and rendering time on a top-tier card is about four minutes per five seconds of video.

Wan is distributed under the permissive Apache-2.0 license for both code and weights, which removes many restrictions on use and modification. However, it is worth noting that all performance estimates are provided by the developers themselves, not obtained through independent testing.