Wan 3.0 🧠 Model Open source

What it is

Wan is Alibaba's video generation family. It has two tracks. Earlier versions such as Wan 2.1 and 2.2 were published as open weights under Apache-2.0, and they became the default open video models in ComfyUI. The newest version, Wan 3.0, entered beta on Alibaba Cloud Model Studio in August 2026 and is offered as an API. We found no open weights for Wan 3.0.

Key features

• Wan 3.0: up to 30 seconds per clip, double the 15 seconds of Wan 2.7
• Accepts text, images, video, audio and even documents (PDF, PPT) as input
• Multilingual voices with synchronised facial expressions
• Instruction-based editing of existing clips
• Wan 2.2 (open): 480p/720p text-to-video, image-to-video, speech-to-video and character animation

Where you can use it

Wan 3.0 is available on Alibaba Cloud Model Studio and Qwen Cloud, and in ComfyUI partner nodes (Wan 3.0 and 2.7). Wan 2.6/2.7 is documented in Higgsfield, Krea, Leonardo.Ai and invideo AI. The open Wan 2.x weights run locally in ComfyUI.

Pricing and rights

Wan 2.1/2.2 weights are free to use under Apache-2.0, including commercial use. You need your own GPU. Alibaba had not published Wan 3.0 API pricing at the beta announcement. In partner apps, Wan uses that app's credits.

Who it is for

Tinkerers and studios that want a capable open model to run and fine-tune locally. Also developers who want a long-clip API from a major cloud.

Verdict

Wan is the most practical open video family, and the Apache licence is creator-friendly. The best new features (Wan 3.0, editing, multi-reference) are API-only. Details like resolution and pricing for 3.0 were still thin in beta.