WAN 3.0 AI動画生成

WAN 3.0 is the 2026 generation of Alibaba's WAN video model line and the longest single take you can generate on FlyAIgh: any whole-second duration from 2 to 30 seconds in one continuous shot, with no stitching. Where most models top out at 10–15 seconds and force you to cut, WAN 3.0 covers slow reveals, monologues, walkthroughs and other shots that only work uncut.

Audio is generated natively alongside the picture — dialogue, ambience and effects land in the same pass, and the audio toggle costs nothing extra. Output comes in 480p, 720p and 1080p at 16:9, 9:16, 1:1, 4:3 or 3:4, or adaptive to your source image.

Four workflows are supported from one account: text-to-video, image-to-video, first-and-last-frame interpolation, and multimodal reference-to-video. The reference mode is unusually deep: a single generation can carry up to 10 reference images, 5 reference videos (15 seconds combined) and 5 reference audio clips (15 seconds combined), and the prompt can address each one by position — "image 2" for a costume, "video 1" for a camera move, "audio 1" for a voice. That makes it practical to direct a shot from a folder of materials rather than a single starting frame.

Compared to WAN 2.7, the 3.0 generation doubles the maximum take length from 15 to 30 seconds, adds native audio, and expands references from images only to images, video and audio together.

FlyAIgh currently prices WAN 3.0 at 7.9 credits per second at 1080p — a limited-time 30% supplier promotion off the 11.3 list rate — so a 5-second 1080p clip costs 40 credits and a 10-second clip costs 79. The 720p tier is half the 1080p rate and 480p is a quarter, which makes long-take drafts cheap: a 30-second 480p test run costs about 60 credits. The exact total is always shown before generation, and no separate Alibaba Cloud account is required.