Wan - AI Video Generator

Explore API

Model

Your active AI engine — tap to switch

Wan 3.0

Alibaba

Version

Capability

Text to Video

What is Wan?

Wan is a family of video generation models by Alibaba, available in the Pixazo playground as three specialised workflows: character videos driven by reference footage (Wan 2.6), image animation with synced sound (WAN 2.5), and pure visual-input character animation (Wan 2.2 Animate). Different inputs, one idea — turn stills and references into moving characters.

Which Wan model does what?

Wan 2.6ref → video

Character video from reference footage

Accepts two reference videos (one required, one optional) and synthesises a new video from the visual patterns in both — a dual-input approach for nuanced character motion. Optional text prompt up to 1,000 characters. Portrait or Landscape only.

Video ref 1Video ref 2Prompt ≤1000Portrait / Landscape

WAN 2.5image → video

Animate an image with synced sound

Takes one reference image and animates it, with audio-synchronised motion where applicable. Prompt up to 3,000 characters; choose 5 or 10 second output.

Reference imagePrompt ≤30005s / 10s

Wan 2.2animate

Character animation from visuals only

No prompt field at all — upload a character image (required) and optionally a motion-reference video. All generation parameters come from image analysis, which keeps character consistency high.

Character imageMotion ref videoNo prompt

Input guidelines that actually matter

Reference videos (2.6)

Clear, well-lit clips, single subject, simple background. No fast cuts or scene changes. Keep refs under ~10 seconds for the best motion extraction.

Motion that transfers

Gross body motion — walking, gesturing, turning — transfers well. Subtle expressions and micro-gestures often get lost.

Proportions

The reference person’s proportions don’t need to match your character exactly, but extreme differences produce unnatural results.

Character images (2.2)

Clean, front-facing character images with clear silhouettes animate most consistently — the model reads everything from the pixels.

Honest limitations

No square output on 2.6

Wan 2.6 renders Portrait or Landscape only — plan your crop accordingly.

Micro-expression loss

Fine facial acting rarely survives motion transfer; expect body-level performance, not close-up nuance.

Short clips

Outputs are short-form (5–10s on WAN 2.5) — stitch takes in an editor for longer pieces.

Reference quality in, quality out

Blurry, busy or multi-person references produce unstable motion — the guidelines above aren’t optional.

Frequently asked questions

Is Wan free to try on Pixazo?

Yes — free starter credits on signup; each generation costs credits.

Which Wan model should I pick?

Have reference footage of the motion you want? Wan 2.6. Have one image you want animated with sound? WAN 2.5. Want consistent character animation without writing prompts? Wan 2.2 Animate.

How long can my clips be?

WAN 2.5 offers 5 or 10 second outputs. Keep reference videos under ~10 seconds for the best extraction on 2.6.

Can I use a text prompt?

On 2.6 (up to 1,000 chars) and 2.5 (up to 3,000 chars), yes. Wan 2.2 Animate is deliberately prompt-free — visuals only.

Who develops Wan?

Alibaba. Pixazo provides access through its playground layer.