Wan 2.5 Preview

• 🎬 Alibaba's text-to-video model in the Wan family.
• 🎯 Photorealistic generation with strong subject coherence across frames.
• 🆕 Released as a preview build in September 2025.
• 🏢 Built and served by Alibaba.
• 👁️ Targets cinematic, short-form video generated from text prompts.
• 📚 Sits alongside an image-to-video preview variant.
• 🔧 Part of a broader Wan lineage spanning text, image, and reference inputs.

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 2.5 Preview is a text-to-video model from Alibaba's Wan family. Per the catalog description, it focuses on photorealistic generation with strong subject coherence across frames, turning written prompts into short video clips. It was released as a preview in September 2025.

Within the family timeline, Wan 2.5 Preview was released alongside Wan 2.2 A14B and was later followed by Wan 2.6 (December 2025) and Wan 2.7 (April 2026), which represent subsequent generations of the text-to-video line. A companion Wan 2.5 Preview variant handles image-to-video generation rather than starting purely from text.

As a preview build, Wan 2.5 marks an early point in the Wan video lineage, with the later 2.6 and 2.7 releases continuing the family's development. Beyond the catalog context, trustworthy primary or independent-evaluator details for this specific preview were limited, so this description stays to verifiable, factual positioning within the family.