Wan 3 Spicy Reference-to-Video API: Request, Price, and Polling

Send approved image, video, or audio references through the Wan 3 Spicy API, estimate a five-second run, and retrieve the asynchronous result.

Key takeawayUse mode=reference-to-video with at least one reference array. Start with one image and a fixed five-second duration, then check your account quote before submitting.

When to use reference-to-video

Use references when a new shot should borrow a product's appearance, a movement pattern, or a sound without fixing its first frame. Wan 3 Spicy accepts up to 10 reference_images, 5 reference_videos, and 5 reference_audios on the same POST /v1/videos/generations endpoint. At least one of those arrays is required in reference-to-video mode. Do not combine them with first_frame or last_frame; those fields belong to image-to-video mode.

Start with an image that you own or have permission to use. For a simple product shot, one approved bottle image is easier to evaluate than a large, conflicting reference pack. Give each additional reference a clear role in the prompt.

Send a five-second request

Complete the Spicy model's adult-access confirmation in your account and set MODELRUSH_API_KEY in your server environment. The API key needs video permission and access to Wan 3 Spicy. Replace the example image URL with an accessible HTTPS image that you are authorized to process. Create an idempotency key once and reuse it if the same submission needs a network retry.

bash
REQUEST_KEY="$(uuidgen)"
curl -sS https://api.modelrush.ai/v1/videos/generations \
  -H "Authorization: Bearer $MODELRUSH_API_KEY" \
  -H "Content-Type: application/json" \
  -H "Idempotency-Key: $REQUEST_KEY" \
  -d '{
    "model": "modelrush/wan3-spicy",
    "mode": "reference-to-video",
    "prompt": "The blue ceramic bottle from the reference image sits on a stone plinth. One continuous shot with a slow clockwise camera orbit and warm side light.",
    "reference_images": ["https://example.com/approved-bottle.jpg"],
    "duration": 5,
    "resolution": "480P",
    "ratio": "16:9",
    "audio": true
  }'

The API returns HTTP 202 with a generation ID, queued status, and poll_url. Save that response. If a client times out after submission, query the saved job or retry with the same idempotency key instead of creating another job.

Poll for the result

Copy the relative poll_url from the 202 response into POLL_URL, then call it on the API host with the same key. Repeat with a bounded delay until the video endpoint reports completed or failed. A completed response carries video_url; download the result you need because that URL can expire after 24 hours.

bash
curl -sS "https://api.modelrush.ai$POLL_URL" \
  -H "Authorization: Bearer $MODELRUSH_API_KEY"

Duration and price to check

Wan 3 Spicy accepts a fixed whole-number output duration from 2 to 30 seconds; omitting it defaults to 5 seconds, and -1 is unavailable. Reference videos can total at most 15 seconds, and their combined time plus the requested output duration cannot exceed 30 seconds. Reference audio can total at most 15 seconds. The gateway validates media before submitting the generation.

On September 28, 2026, the public Registry rate for Wan 3 Spicy at 480P was $0.045 per output second. The five-second, image-reference request above therefore has a $0.225 public-rate output estimate (5 × $0.045). This example contains no reference video. Account-specific rates may differ; the quote shown to your account before generation is the charge to use. Check the current Wan 3 Spicy model price or pricing page before increasing resolution or volume.

For the complete current schema, see the video API documentation. Review Wan 3 Spicy model details, then try reference-to-video in the Playground.