Wan 3.0 Prime Reference

• 🎯 Reference-to-video variant of Alibaba's third-generation Wan video family.
• 👁️ Reference images anchor subject appearance so identity stays consistent across frames.
• 📏 Alibaba reports Wan3.0 generating up to 30 seconds in one pass.
• 🔧 Prompt acts as the control layer over varied generation inputs.
• 🆕 Prime tier dated 2026-08-24, three weeks after base Wan 3.0.
• 🏢 Built by Alibaba; documented on Alibaba Cloud.
• 🌐 Same-day sibling entries cover text-to-video and image-to-video.
• 🔒 No independent evaluator scores published; description limited to documented capabilities.

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 3.0 Prime Reference is the reference-conditioned entry in Alibaba's third-generation Wan video line, listed alongside the same-day text-to-video and image-to-video Wan 3.0 Prime and Wan 3.0 Prime variants. You supply one or more reference images together with a prompt, and the model produces footage that carries the referenced subject through the shot rather than letting its appearance drift between frames.

Alibaba's own write-up describes Wan3.0 as a single model able to generate up to 30 seconds of video in one pass, accepting a broad range of inputs with the prompt serving as the orchestrating control layer. In this catalog the reference mode is exposed as its own endpoint, so subject conditioning is explicit rather than something the prompt has to imply.

Within its own lineage, this is the second reference-mode listing of the 3.0 generation: the base Wan 3.0 Reference entry is dated 1 August 2026, with Prime following on 24 August 2026 as a separate tier of the same family. Both succeed Wan 2.7 Reference from April 2026, and the catalog does not document the specific differences between the base and Prime tiers.

Alibaba has not published benchmark results for Wan3.0's reference mode, and no independent top-tier evaluator figures were available, so this page stays with the capabilities the provider documents and the metadata in the catalog rather than any quality ranking.