What's this about?
Alibaba opened public beta for Wan3.0, its text-to-video model. Headline features: it can generate a full 30-second clip natively, aims for "reality-grade" rendering, and adds an "Omni-Reference" mode that conditions on more than just a text prompt.
What exactly is "text-to-video" / "Omni-Reference"?
Text-to-video turns a written description into a moving clip, the way image generators turn text into a picture. Most models only manage a few seconds and lose consistency; 30 seconds native is notable. "Omni-Reference" means you can steer the output with reference material (images, subjects, styles) instead of relying on words alone — useful for keeping a character or look consistent.
Why it matters
Long, consistent generation is the hard part of AI video, and open Chinese labs pushing to 30-second reality-grade clips raises the competitive floor against closed video models.