Wan 2.1 and Qwen by Alibaba: video and text AI for marketers

Searches for "Alibaba One 2.1" most likely refer to Wan 2.1, the video generation model that Alibaba Cloud released as open source in February 2025 ("Wan" sounds like "one"). Alibaba has no AI model officially called "One 2.1". Wan sits next to Qwen, Alibaba's family of large language models, formerly known as Tongyi Qianwen. This guide, updated in October 2026, explains what these two families do, how they have evolved, and how a marketing team can use them without losing control of its brand.

Wan 2.1: Alibaba's open-source video model

On February 25, 2025, Alibaba published the inference code and weights of Wan 2.1 on GitHub. According to Alibaba Group's announcement, four models were released, in 14-billion and 1.3-billion parameter versions: two for text-to-video (T2V-14B, T2V-1.3B) and two for image-to-video (I2V-14B-720P, I2V-14B-480P). They are available on ModelScope and Hugging Face.

Alibaba highlights two points for creative teams:

• Wan 2.1 is presented as the first video generation model able to render text effects in both Chinese and English.
• The smallest model (T2V-1.3B) can generate a 5-second 480p video on a standard personal laptop in "as little as around 4 minutes", according to the company.

The models are released under the Apache 2.0 license. The GitHub repository states that Alibaba claims no rights over the content users generate, and reminds users that they remain fully accountable for how they use the models.

From Wan 2.2 to Wan 3.0

Alibaba has updated Wan several times since then:

• Wan 2.2 (July 2025): also open source under Apache 2.0, with a mixture-of-experts architecture. Its smallest version can generate 720p video on a single RTX 4090 GPU (source: The Decoder).
• Wan 2.7: generation window of 2 to 15 seconds and the first video editing features.
• Wan 3.0 (August 24, 2026): general availability on Alibaba Cloud Model Studio, with clips of up to 30 seconds in 480p, 720p or 1080p, and reference inputs that include images, audio, video, web pages and documents such as PDF or PPT. API pricing starts at $0.05 per generated second in 480p and reaches $0.20 in 1080p (source: Alibaba Cloud). According to Dataconomy, Alibaba did not release the weights of Wan 3.0, unlike earlier versions.

Wan competes with other video models such as Google Veo (see our article on Google Veo 2 and its successors) and Kling AI 2.0.

Qwen: Alibaba's language models

Alibaba launched its language model in beta in April 2023 under the name Tongyi Qianwen, later renamed Qwen (source: Wikipedia). Some Qwen models are published with open weights, often under Apache 2.0, while the most powerful versions (Max, Plus) are offered through Alibaba Cloud. For example, the Qwen3 family released on April 29, 2025 included eight models under Apache 2.0 (source: Simon Willison). In August 2026, Alibaba unveiled Qwen3.8, a 2.4-trillion-parameter model focused on coding and office work (source: TechNode).

For a marketer, Qwen covers the same text tasks as other large language models: drafting, translation, summaries and copy variations. To compare Chinese and American models, see our DeepSeek vs OpenAI comparison.

Creative use cases for marketing teams

• Animating existing visuals: Wan's image-to-video mode can turn a validated key visual into a short motion version. Our guide on how to animate an image with AI covers the method.
• Prototyping: testing a video idea or storyboard before a shoot.
• Bilingual text in video: Alibaba highlights Chinese and English text rendering, which can help for campaigns in both languages. Every output still needs checking.
• Self-hosting: because the Wan 2.1 and 2.2 weights are open, a technical team can run them on its own infrastructure, which keeps source files in-house.

Limits and rights to check

• Quality claims: scores such as Wan 2.1's VBench result come from Alibaba itself. Test the models on your own briefs.
• Licenses: Apache 2.0 allows commercial use of the open models, but closed versions such as Wan 3.0 or Qwen Max follow Alibaba Cloud's terms of service. Read them before production.
• Copyright: in the United States, the Copyright Office concluded on January 29, 2025 that prompts alone do not make a person the author of an AI output (source: U.S. Copyright Office).
• Labelling in the EU: Article 50 of the AI Act requires deployers to disclose deepfakes from August 2, 2026 (source).
• Data: when you use a cloud API, check where prompts and uploaded files are processed and stored, in line with your data policy.

Brandeploy: keeping AI output on brand

Models like Wan or Qwen produce raw material. The brand still has to decide what is published, in which formats and for which markets. Brandeploy is a creative automation platform: teams centralize their assets, build templates with locked brand elements and send content through approval workflows before release. An AI-generated clip or text can be reviewed and adapted there like any other asset. Learn more about Brandeploy's creative AI automation, or book a demo to see the platform with your own use cases.

FAQ

What is "Alibaba One 2.1"?

Alibaba has no AI model officially called "One 2.1". The search most likely refers to Wan 2.1, the video generation model Alibaba released as open source in February 2025, or more broadly to Qwen, Alibaba's family of language models.

What can Wan 2.1 do?

Wan 2.1 generates videos from text (text-to-video) or animates an image (image-to-video). Alibaba released four models in 14B and 1.3B parameter versions, available on ModelScope and Hugging Face under the Apache 2.0 license, and presents it as the first video model able to render text in Chinese and English.

What is the latest version of Wan?

As of October 2026, the latest version is Wan 3.0, generally available since August 24, 2026 on Alibaba Cloud Model Studio. It generates clips of up to 30 seconds in up to 1080p, with API pricing from $0.05 to $0.20 per generated second depending on resolution.

What is the difference between Wan and Qwen?

Wan is Alibaba's video (and image) generation family. Qwen, formerly Tongyi Qianwen, is its family of large language models for text, code and reasoning. Both are offered through Alibaba Cloud, and some versions of each have been published with open weights.

Can a brand use Wan videos commercially?

The open Wan 2.1 and 2.2 models are under Apache 2.0, which allows commercial use, and Alibaba claims no rights over generated content. Closed versions such as Wan 3.0 follow Alibaba Cloud's terms. Brands must also check copyright limits, the consent of people shown and EU labelling rules for deepfakes.