What changed in Wan 3.0
Wan 2.1 and 2.2 were open-weight; Wan 2.5, 2.6 and 2.7 moved to closed commercial APIs. Wan 2.7 (April 2026) added a Thinking Mode, first and last frame control and instruction-based editing, with clips up to 15 seconds. Wan 3.0 doubles the ceiling to 30 seconds per generation, generates audio with the picture, and adds document and web page inputs so a deck, spreadsheet or product page can be turned into a video that follows its structure. Alibaba says instruction following, shot-to-shot consistency and audio quality all improved, though no independent benchmarks back that up yet.
How to get Wan 3.0
Alibaba Cloud Model Studio and Qwen Cloud APIs, billed per second of output by resolution. Consumer access through wan.video is rolling out to members, and third-party platforms such as Morphic also offer it. No self-hosting.
What Wan 3.0 means for short-form marketers
Thirty seconds changes what one generation can be. Most AI video models stop at 8 to 15 seconds, so a full hook, demo and call to action has meant stitching clips together. Wan 3.0 can produce the whole beat in one pass, which keeps a character, a product and a voice consistent across the entire ad.
Document and web page inputs are the genuinely new idea. Marketing teams already own the source material: a product page, a launch deck, a feature spreadsheet. Being able to hand that to the model and get a structured video back is a fast way to draft explainers and product walkthroughs, though Alibaba itself flags on-screen text accuracy as an area still improving, so check every caption.
At $0.20 per second, a 30 second 1080p clip costs about $6, and 720p halves that. Feeds compress hard, so test at 720p first, and remember that long single shots without a story turn tend to feel slow on TikTok: the hook still has to land in the first two seconds.
