WAN3 models and billing
Choose the input family, resolution and speed tier.
Model families
| Family | Input | Billable seconds |
|---|---|---|
wan3.0-image-* | Image, optional audio reference | Output |
wan3.0-image-prime-* | Faster image family, optional audio | Output |
wan3.0-video-* | Text, image, video and audio references | Output + reference video |
wan3.0-video-prime-* | Faster standard family | Output + reference video |
All families generate video. Actual waiting time depends on service load.
Public model IDs
| Family | 480p | 720p | 1080p |
|---|---|---|---|
| Image | wan3.0-image-480p | wan3.0-image-720p | wan3.0-image-1080p |
| Image prime | wan3.0-image-prime-480p | wan3.0-image-prime-720p | wan3.0-image-prime-1080p |
| Video | wan3.0-video-480p | wan3.0-video-720p | wan3.0-video-1080p |
| Video prime | wan3.0-video-prime-480p | wan3.0-video-prime-720p | wan3.0-video-prime-1080p |
Use GET /v1/models for your key's current permissions. The suffix determines the resolution tier; changing size alone does not change that tier.
Cost
Read the current rate P for your account/group, and its currency/unit, from video pricing.
- Image:
P × output seconds. - Video:
P × (output + unique reference video seconds). - Images and audio do not add billable seconds.
- Identical reference video URLs are counted once and must declare consistent durations. Avoid duplicates.
Ten output seconds plus a five-second reference costs fifteen seconds. Billing is based on video durations, not time spent waiting or a fixed per-generation fee. Do not infer the whole charge from quota_type alone.
Charges may be reserved during creation. Use task and usage/refund records to check the final result.
Duration limits
Output: 2–30 whole seconds. Up to 5 reference videos, 1–15 whole seconds each, at most 15 reference seconds total. Output plus references must not exceed 30 seconds. Thirty-second output therefore cannot include a reference video.
Declare the actual duration. Do not invent a rounding policy for fractional durations; use suitable whole-second media or confirm the handling first.