Wan 3.0 AI Video Generator
Wan 3.0

Wan 3.0 is the newest generation of Alibaba's Wan video model family, released into public beta. Its two headline changes are a maximum single-clip length of about 30 seconds, roughly double the previous generation, and multimodal input: alongside text and images it can take video, audio, web pages, PDFs and slide decks as prompt material, so a deck or a product page can seed a video instead of a hand-written brief.

Not yet, and we would rather say so plainly than sell you a page for a model you cannot run. Wan 3.0 is in public beta on Alibaba's own platform and a general API has not opened to third-party services. VO3 currently runs Wan 2.5 and Wan 2.7, and we add new models within days of their APIs going live, as we did with Sora 2, Seedance 2.5 and Kling. If Wan 3.0 is what brought you here, Wan 2.7 is the closest thing you can generate today.

Wan 2.5 is the cheaper, audio-first option: native synchronised sound, 480p through 1080p, text-to-video and image-to-video, available from the Trial tier. Wan 2.7 is the newer model with stronger motion, clips up to about 15 seconds at 1080p, and first-frame plus last-frame control for pinning the start and end of a shot. Wan 2.5 is the better default for volume; Wan 2.7 is worth the extra credits when a shot has to match something else.

On paper, the Wan 3.0 generation's advantages are clip length and the breadth of input types it accepts. Veo 3.1 still tends to lead on physical realism and dialogue lip-sync, and Sora 2 on stylised and surreal scenes. There is no universal winner, which is exactly why VO3 puts them in one editor: run the same prompt across all three, compare the actual output on your own subject matter, and pay for one balance instead of three subscriptions.

Yes. Both Wan 2.5 and Wan 2.7 generate audio natively and in sync with the picture, covering spoken dialogue, ambient sound and simple music beds. This is the main practical reason people pick a Wan model over a silent generator: one prompt gives you a clip you can post, instead of a video file that still needs a voiceover session and an editing pass.

VO3 uses a single credit balance across every model, and each model card shows its exact credit cost per resolution before you generate. Wan 2.5 is the cheapest tier in the family and is available on the free Trial level; Wan 2.7 costs more and scales with duration and resolution. New accounts get free credits, so you can test the Wan AI video generator on a real brief before paying anything. Full plan pricing is below.

Yes. Use the image-to-video tab, upload the asset, and describe the motion you want. This is the most reliable way to get brand-accurate output, because the model animates the file you gave it rather than generating its own interpretation of your product. Wan 2.7's first-frame control is particularly useful here: your uploaded image becomes the opening frame exactly as supplied.

Prompts and on-screen dialogue work in Spanish, Portuguese, French, Arabic, Japanese, Chinese, English and others. Our public gallery has plenty of non-English briefs that rendered cleanly on the first attempt, and the VO3 interface itself ships in eight languages.

Most Wan clips come back within a few minutes, depending on resolution and current queue load. Generation runs in the background, so you can queue several prompt variations at once and come back to a library of finished takes rather than waiting on each one in turn.