Wan 3.0 AI Video Generator

Wan 3.0 is Alibaba's Wan video generation model, version 3.0. It makes 2–15 second clips with sound from text, a first frame (optionally with a last frame) or omni-reference inputs, and it can replace the person in a video. blubiu runs the model directly and bills it in blubiu credits; this is not Alibaba's own app or website.

Only speed and price. Standard takes about 6 minutes and Prime about 1 minute; Prime costs more per second. Inputs, length, resolutions and quality options are the same, so pick Prime only when you're waiting on the result.

Per second of output, with Prime costing more per second than Standard. When you pass a reference video, its seconds are billed too. Current rates are in the table above, the Generate button shows the total before you confirm, and failed generations are refunded.

Yes, with the Free plan's monthly credits. Standard at 480P is the cheapest way to test a shot or a swap, the Generate button shows the exact cost before you confirm, and failed runs are refunded automatically. Plans and credit packs are on the pricing page.

2 to 15 seconds per generation. With reference videos, those can total up to 15 seconds, and reference plus output must stay within 30 seconds. For a longer film, generate several clips and join them.

Not in one run. The model works either from frames (a first frame, optionally a last one) or from references. When you add a reference video, an audio clip or three or more images, blubiu sends the run as references, and your first image becomes one of them.

Yes. In our tests it replaced a dancing woman with a robot character, and in another clip swapped in a fox pilot and a skateboarding panda, keeping the room, the camera and the rhythm. Characters need a clear head, torso and limbs; try it on recast video with a character from the mascot generator or your own sheet.

Both generate sound with the picture and go up to 15 seconds at 1080P. Choose Wan 3.0 for clips shorter than 4 seconds, swapping the person in a video, or a fast Prime run. Choose Seedance 2.0 when you want to draft on its Mini version first and rerun the keeper on the full model.

Not reliably. The output's audio track is generated and may not match the original word for word. If the voice matters, mute the result and put the original track back in your editor.

Not from this model, which tops out at 1080P, and swaps default to 720P. Upscale the finished clip with FlashVSR on the AI video upscaler: it reaches 2K or 4K and keeps the motion, length and frame rate.