Asynchronous jobs

Live
Submit, persist, poll, and reconcile media predictions.
Open in ChatGPT
(opens in new tab)
Last verified: 2026-09-08

Prediction lifecycle

Image, video, and audio requests share an internal Prediction record. The canonical states are queued, processing, succeeded, failed, cancelled, and expired. Store the returned generation or Prediction ID, model, region, submitted time, and your internal job ID before polling.

Backoff strategy

Poll GET /predictions/:id with exponential backoff and jitter. Existing video clients may keep using GET /videos/generations/:id. Stop on any terminal state and continue reconciliation in a worker if the user leaves.

Timeout and cancellation

Predictions expire seven days after their record lifecycle. DELETE /predictions/:id returns the unchanged terminal Prediction, or 409 cancellation_not_supported when the selected Provider operation cannot safely cancel an in-flight job.

Retry boundary

Retry status reads freely within rate limits. Do not blindly resubmit generation requests after an ambiguous timeout because that can create an additional billable job. Reconcile the original request first.

Spicy model API references

Open a model page for its request parameters and current prices. The four Wan 3 landing pages include a developer API section below the Playground. Read GET /v1/models for current schemas and key-specific access; GET /api/model-pricing exposes current customer prices.
Model and API referenceModel IDEndpoint
Wan 3 Spicymodelrush/wan3-spicyPOST /v1/videos/generations
Wan 3 Pro Spicymodelrush/wan3-pro-spicyPOST /v1/videos/generations
Wan 3 Prime Spicymodelrush/wan3-prime-spicyPOST /v1/videos/generations
Wan 3 Prime Pro Spicymodelrush/wan3-prime-pro-spicyPOST /v1/videos/generations
Wan 2.7 I2V Spicymodelrush/wan2.7-i2v-spicyPOST /v1/videos/generations
Wan 2.2 I2V Spicymodelrush/wan2.2-i2v-spicyPOST /v1/videos/generations
Wan 2.2 I2V LoRA Spicymodelrush/wan2.2-i2v-lora-spicyPOST /v1/videos/generations
Z-Image Spicymodelrush/z-image-spicyPOST /v1/images/generations
Z-Image Pro Spicymodelrush/z-image-spicy-proPOST /v1/images/generations
Qwen Image Edit Spicymodelrush/qwen-image-edit-spicyPOST /v1/images/edits
Face Swap Spicymodelrush/face-swapPOST /v1/images/edits
Head Swap Spicymodelrush/head-swapPOST /v1/images/edits
Wan Animate Spicymodelrush/wan-animatePOST /v1/videos/generations
FlashVSR Spicymodelrush/flashvsrPOST /v1/videos/generations
MuleRouter image calls can return HTTP 202 with id and poll_url. Poll that URL with the same API key; GET /v1/predictions/:id completes with status succeeded and outputs. Z-Image Pro Spicy accepts width and height from 256 to 2560 (default 1024), n = 1 and prompt_extend = false by default. Z-Image Spicy also produces one image per request. Do not copy another model's size or output-count limits.
Video polling via GET /v1/videos/generations/:id ends with completed or failed and returns video_url on success. The unified GET /v1/predictions/:id instead uses succeeded and outputs. Follow the contract of the poll_url you use. Wan 3 reference-video time plus generated time cannot exceed 30 seconds. Only Prime and Prime Pro currently accept prompt_extend; check the per-model schema. LoRA supports three optional weight arrays, 480P/720P and 5/8 seconds. Wan Animate and FlashVSR use the video generation endpoint with source media.
All Spicy models require adult-access confirmation on the account. API keys must permit the selected model and capability. Keep keys server-side and reuse Idempotency-Key for retries of the same submission.