Wan 3.0Alibaba’s 30-second AI video model, online

Wan 3.0 (also written Wan3.0 or Wan 3; Tongyi Wanxiang 3.0 in Chinese) is the newest model in Alibaba’s Wan AI video family. One model covers text-to-video, image-to-video from a first frame or from a first and last frame, and reference-guided generation, with clips of up to 30 seconds at up to 1080p and sound generated together with the picture.

Alibaba opened a public beta on 6 August 2026, listed the faster Wan 3.0 Prime on 20 August and released Wan 3.0 officially on 24 August 2026.

No. As of October 2026, Alibaba offers Wan 3.0 only through its cloud API and has not released the model weights; the most recent flagship Wan video model with open weights is Wan 2.2, from July 2025. On HelloGen there is nothing to install — open the video generator and Wan 3.0 is in the model list.

Video takes credits. Credit packs start at $4.90 for 500 credits that never expire — enough for seven 5-second Wan 3.0 clips at 480p — and Pro includes 3,600 credits a month, about 25 five-second clips at 720p. If what you want is volume, Pro and Max also include MiniMax H3 Unlimited: 0 credits per clip, with guaranteed priority clips every day (at least 2 a day on Pro, 5 on Max).

14 credits per second at 480p, 28 at 720p and 56 at 1080p: a 5-second 480p clip is 70 credits, a 10-second 720p clip is 280 and a 30-second 1080p clip is 1,680. Sound on or off costs the same. With a reference video attached, its seconds are billed at the same rate as the output seconds, which is how Alibaba itself prices it. The exact price shows before you generate, and a failed render is never charged.

They are the same model, with the same capabilities, resolutions, lengths and inputs. Prime renders faster; Alibaba’s own summary is “Prime buys time, not quality”. On HelloGen, Prime costs about 1.6 times the credits of Wan 3.0, so it is worth it when you are waiting on the result — for drafts and most finished clips, Wan 3.0 is enough.

Wan 2.7 is the previous generation, released in April 2026: 720p or 1080p, 2 to 15 seconds, sound always on, and up to 10 seconds in reference mode. Wan 3.0 runs to 30 seconds, adds a cheaper 480p tier, lets you switch sound off and takes more reference material. If you only need 1080p for up to 15 seconds, Wan 2.7 costs 42 credits per second against Wan 3.0’s 56.

Both generate sound together with the picture. Wan 3.0 reaches 1080p and costs about half as many credits per second on HelloGen, or less; Seedance 2.5 offers 21:9 widescreen and is strongest at keeping a scene consistent across several shots. A credit-saving way to work: draft on Wan 3.0, then switch to Seedance 2.5 when the shots need to hold together.

Alibaba recommends this order: the overall scene, then references, shots, dialogue, sound and style, skipping whatever you do not need. Write shots as “Shot 1 [0-3s]: …”, 2 to 5 seconds each with no gaps between them. Write dialogue as she says: "…", naming the speaker every time, and write “No dialogue.” when nobody should speak. On-screen text is not reliable yet, so avoid asking for captions or signs.

Type @ in the generator and pick an uploaded file: it appears in the prompt as @Image1, @Video1 or @Audio1 and is rewritten into Wan’s own form, “Image 1”, when it is sent. You can attach up to nine reference images, one reference video of 1 to 14 seconds and one audio track of 1 to 15 seconds; with a reference video attached, the clip itself can be up to 15 seconds.

480p, 720p and 1080p; any length from 2 to 30 seconds per clip, with 5, 10, 15 and 30 as presets; 16:9, 9:16, 1:1, 4:3 and 3:4, plus Auto — which follows the shape of your picture — once you upload one.

Yes — clips you generate on HelloGen are yours to use, including commercially, to the extent the law allows. You are responsible for making sure your prompt and any image, video or audio you upload do not infringe someone else’s rights; the Acceptable Use Policy covers the details.

Alibaba screens for content that looks too close to well-known characters or brand designs. A refused video costs no credits: change the picture or reword the prompt, and generate again.