Wan 2.2 A14B

• 🎬 Alibaba's open-weight text-to-video model from the Wan family
• 🧠 Mixture-of-Experts diffusion transformer, roughly 14B active parameters
• 📏 Generates at both 480P and 720P resolutions natively
• 👁️ Photorealistic frames with strong subject coherence across motion
• 🔧 Family includes paired image-to-video and hybrid TI2V variants
• 📚 Open-weight release with public weights and inference code
• 🌐 Succeeded by Wan 2.5 Preview, 2.6 and 2.7

Alibaba

Alibaba Group is a Chinese multinational technology company founded in 1999 and headquartered in Hangzhou, Zhejiang. Originally built around e-commerce and cloud computing, Alibaba has become one of the most prolific contributors to open-weight AI research,…

Wan 2.2 A14B is a text-to-video model in Alibaba's open-weight Wan video family, built on a Mixture-of-Experts (MoE) diffusion transformer. The "A14B" designation reflects its activated parameter budget, with the MoE design routing through specialized experts so only part of the network runs per inference pass. Alibaba Cloud's Model Studio documents the Wan family generating at both 480P and 720P, with paired text-to-video and image-to-video pipelines.

Compared with the earlier Wan 2.1 generation, the 2.2 release moved the family toward the MoE diffusion architecture and added paired A14B text-to-video and image-to-video checkpoints, with prompt-extension tooling for richer scene descriptions.

Within the catalog, Wan 2.2 A14B is the oldest entry in its text-to-video lineage. It was followed by Wan 2.5 Preview, then Wan 2.6, and most recently Wan 2.7, which extends the family across text-to-video, image-to-video, reference-to-video and instruction-based editing variants.

For users wanting an openly available, reproducible pipeline rather than the newest one, Wan 2.2 A14B provides an openly available baseline, with public weights, documented inference code and dual-resolution output.