Wan 2.5

Wan 2.5 is Alibaba's AI video model, first shown as Wan2.5-Preview in September 2025. It makes clips of up to 10 seconds at up to 1080p from a text prompt or a start image, and it was the first Wan model built to generate audio — voices, music and effects — in sync with the picture. There is also a Wan 2.5 image model for stills.

No. Wan 2.1 and Wan 2.2 have open weights under Apache 2.0, but Wan 2.5 and every later version (2.6, 2.7, 3.0) run only as hosted models through Alibaba Cloud and API platforms. You can't download Wan 2.5 for ComfyUI; you run it online, for example in Kenerate's video studio.

It's billed per second by resolution: 15 credits a second at 480p, 30 at 720p and 45 at 1080p. A 5-second clip costs 75, 150 or 225 credits, and a 10-second 1080p clip 450. Credits come in one-time packs that never expire, starting at $15 for 1,300 credits — no subscription.

Up to 10 seconds per clip. Text to video takes 5 or 10 seconds; image to video accepts any whole number from 3 to 10. To go longer, use video extend, which adds another 3–10 seconds to an existing clip, or chain clips by starting each from the last frame of the one before.

Alibaba built Wan 2.5 to make audio and video in one pass, including lip-synced speech, though one launch review found music and drumming drifting out of sync. On Kenerate you can attach your own wav or mp3 track for the clip to follow; there's no sound toggle, so for sound written from the prompt alone use Wan 3.0.

Yes. Upload a start image in /app/video, choose 3–10 seconds and describe the motion. Wan 2.5 keeps the image as the first frame, which is the most reliable way to keep a product, character or layout as designed. It doesn't take an end frame; Wan 2.7 and Kling 3.0 do.

Not on Kenerate's free plan: video needs a paid plan, which means buying any one-time credit pack. New accounts get starter credits for images on free-plan models such as Qwen Image 3.0. The Wan 2.5 image model costs 9 credits an image and is not a free-plan model.

It's Alibaba's text-to-image model from the same release, listed in Kenerate's image studio as Wan 2.5. It makes images up to 1440 px a side, supports a negative prompt and a seed, and can edit or blend up to 2 reference images. It costs 9 credits per image and 10.5 per edit. The four examples on this page are its real outputs.

Wan 2.5 is the cheapest way to draft Wan video (15 credits a second at 480p, which Wan 2.6 and 2.7 don't offer) and costs the same as them at 1080p (45 a second) when 10 seconds is enough. Wan 2.6 adds 15-second clips and reference images, Wan 2.7 adds editing and more aspect ratios, and Wan 3.0 makes up to 30 seconds with generated sound. Draft on Wan 2.5 at 480p, then move up if you need length or sound.

It's the API name model platforms such as WaveSpeed use for Wan 2.5, with separate text-to-video, image-to-video and video-extend endpoints. On Kenerate you don't need an API key or a provider account: the same model runs in the browser as Wan 2.5 and is billed in credits.