Short reference notes on Alibaba's Wan 3.0 video model, written while building a front-end for it. Every number is checked against the vendor's own documentation and dated, including the numbers that are unflattering.
What each knob on a generative video request actually costs, and which limits are budgets rather than settings.
Alibaba's list price for Wan 3.0 is per second of output, and it moves with resolution:
Wan 3.0 takes reference material with one request. The published caps are:
Wan 3.0 will take a document or a web page as the starting point for a video. The published limit is precise, and the precision is the interesting part:
Short notes on what the individual knobs of a generative video request cost, and on the limits that read like preferences but behave like budgets.
The specification table these notes are drawn from lives at wan-3.run/wan-3-0-specs, with each row linked to its source.