Wan 3.0 API Is Live on VideoGenAPI: 30-Second Video With Audio

Why Wan 3.0 matters

Alibaba put Wan 3.0 into public beta on August 6, 2026 and released it formally on August 24. It doubles the 15-second ceiling of Wan 2.7 to 30 seconds per generation, which is long enough to carry a full product demo, a short scene or a social ad without stitching clips together. Alibaba highlights realistic faces with synchronized micro-expressions and voice output, and the model generates the soundtrack with the picture.

Unlike earlier Wan flagships, Wan 3.0 has no open weights: it is only offered as a hosted API. We covered the release and the open-source question in Alibaba Wan 3.0: 30-second video, and why it is not open source. Until now, using it meant region-matched Alibaba Cloud credentials; on VideoGenAPI it is one more value of the model field.

What you can do with it on VideoGenAPI

• Text-to-video from a prompt of up to 20,000 characters.
• Image-to-video from a start image (image_url), with an optional last frame (end_image_url) to control where the shot lands.
• 2 to 30 seconds per clip, at 480p, 720p or 1080p.
• Six aspect ratios: 16:9, 9:16, 1:1, 4:3, 3:4, or adaptive to let the model pick the framing.
• Native audio on by default; send "generate_audio": false for a silent clip.
• Reproducible results with a seed.

How to call it

The request is the same as for any other model. Only the model field changes:

curl -X POST "https://videogenapi.com/api/v1/generate" \
  -H "Authorization: Bearer $VIDEOGEN_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "wan-3",
    "prompt": "A street musician plays jazz on a saxophone under a neon sign in light rain, close-up, she smiles at the crowd, ambient city sounds",
    "duration": 10,
    "resolution": "720p",
    "aspect_ratio": "16:9"
  }'

The response returns a generation_id. Poll GET /api/v1/status/{id} until the status is completed, then read video_url. Longer clips take longer to render, so poll every few seconds rather than holding a connection open. For image-to-video, add "image_url" (and optionally "end_image_url"). Full parameter list in the generate reference and on the Wan 3.0 model page.

Pricing

Wan 3.0 is billed per second of output from your credit balance, on top of any plan:

• 480p: $0.10 per second (a 10-second clip costs $1.00)
• 720p: $0.20 per second (a 10-second clip costs $2.00)
• 1080p: $0.40 per second (a 10-second clip costs $4.00)

The exact amount is checked against your balance before the job starts, so you never get a half-paid render, and failed generations are refunded automatically. Draft at 480p, then render the version you keep at 1080p. For a broader view of per-second versus per-video billing, see our pricing comparison.

When to pick Wan 3.0

• Long single shots: up to 30 seconds, like Seedance 2.5, and three times the length of most premium models.
• Faces and voices: talking characters with expressions that follow the dialogue.
• Start-to-end control: pin both the first and the last frame for transitions and product reveals.

For human motion at a flat price, Kling 3 is included in every plan. For the closest prompt adherence on short clips with dialogue, FLUX 3 Max is the alternative billed per second. Switching between them only changes the model field.

Frequently asked questions

Sources