Rate limits
Generation Rate Limit
To keep queue stability and predictable latency, tasks are scheduled with per-group concurrency limits.
| Task Group | Concurrency | Applies To |
|---|---|---|
| Multiview Image Generation | 1 | generate_multiview_image |
| Multiview Image Editing | 1 | edit_multiview_image |
| Refine | 5 | refine_model |
| P1 Model Generation | 5 | text_to_model, image_to_model, multiview_to_model with model_version = P1-20260311 |
| Standard Model Generation | 10 | text_to_model, image_to_model, multiview_to_model with other model versions |
| Animation | 10 | animate_model, animate_prerigcheck, animate_rig, animate_retarget |
| Other Tasks | 10 | All other task types |
Limits are enforced by task group, not by each individual task type. For example, text_to_model and image_to_model with non-P1-20260311 versions share the Standard Model Generation pool. If one task type occupies all 10 slots in that pool, other task types in the same pool must wait.
If you exceed the threshold, the API will return the error below.
{
"code": 2000,
"message": "You have exceeded the limit of generation",
"suggestion": "Try again later. You can also check `Retry-After` header."
}We recommend implementing an exponential backoff strategy in your code to handle these responses gracefully.
Face Count Limits
The maximum number of faces (face_limit) depends on the model version and geometry quality:
| Model version | Standard (geometry_quality=standard) | Ultra (geometry_quality=detailed) |
|---|---|---|
| v3.1 | 1,500,000 | 2,000,000 |
| v3.0 | 1,000,000 | 1,500,000 |
When quad=true, the quad face count should not exceed 150,000.
3D Model Upload Limits
The uploaded 3D model file size should not exceed 150MB.
Upload API
Image upload has a separate rate limit of 10 QPS.