Introduction
Wan 3.0 Spicy Image-to-Video is the image-to-video endpoint of the Wan 3.0 Spicy family, tuned to render what the standard version tones down. An opening frame in image_url and an optional prompt become 2 to 30 seconds of video at 480p, 720p or 1080p, with generated audio. The still supplies the person, outfit, set and light; the prompt supplies motion, camera and sound. Add last_image_url to fix the ending.
There is no content flag in the request body and no per-request declaration. In our six-tier content test on 2026-09-30 every tier came back as written.
Key features & highlights
Six tiers tested, six rendered
We ran our standard content test at 480p and five seconds: six prompts across five intensity levels, from Mild to Maximum (the test prompts are not published). Nothing was refused, covered or swapped for a different person, and none of the prompts needed softening.
It goes further than the prompt on its own
This is the part to plan for. In our example shoots we sent six opening frames with prompts that said the outfit stays as it is; four came back with the outfit changed, including every frame with more than one person in it. Treat the prompt as a request, not a guarantee, and review the whole clip.
Up to 30 seconds in one request
duration_seconds takes any whole number from 2 to 30, default 5. A five-second request returns about 5.04 seconds at 30 frames per second. resolution is 480p, 720p (default) or 1080p, and both length and resolution affect price and turnaround.
Shape, sound and the last frame
aspect_ratio takes 16:9, 9:16, 1:1, 4:3 or 3:4; leave it out and the clip follows the still. generate_audio is on by default. last_image_url pins the final frame — in our test the clip ended on the supplied image, cropped to the requested ratio.
Request parameters
| Parameter | Required | Description |
|---|---|---|
prompt | No | Optional but worth writing: one action, one camera move, one note on sound. Asking for the outfit to stay unchanged does not reliably keep it unchanged on this model. at least 1 characters |
image_url | Yes | Required. The opening frame: the person, outfit, room and light all come from it. |
last_image_url | No | Optional final frame. The clip ends on this image, cropped to the requested aspect ratio. Requires |
aspect_ratio | No | 16:9, 9:16, 1:1, 4:3 or 3:4. Leave it out and the clip follows the opening frame's shape; a different ratio crops the frame. |
resolution | No | 480p, 720p or 1080p, default 720p. Higher resolution costs more per second and takes longer. |
duration_seconds | No | Any whole number from 2 to 30, default 5. Billed per requested second. |
generate_audio | No | On by default, at no extra charge: the clip returns with sound aligned to the picture. Set it to false for a silent clip. Default: |
enable_prompt_expansion | No | Off by default, at no extra charge. When on, a short prompt is rewritten into a longer one before generation; leave it off when your wording is already precise. Default: |
seed | No | 0 to 2147483647. Hold it while you compare resolutions or durations; it keeps runs comparable, not frame-identical. |
Pricing
| Specification | Price |
|---|---|
| $0.057 / second |
| $0.114 / second |
| $0.228 / second |
Use cases
Scenes other APIs refuse
Scenes other APIs refuse or tone down: send the still, describe the movement, get the clip. Only consenting adults and properly licensed material may be used, and you are responsible for rights, age and local law for everyone depicted.
Companion and roleplay characters
Keep one character still and reuse it as the opening frame for every clip, changing only the action. The still carries the face and outfit, so the character stays the same person from clip to clip.
Long takes with a planned ending
Ask for up to 30 seconds and add last_image_url so the take lands on the pose your next shot starts from.
Prompt guide & pro tips
Let the frame do the styling
Put wardrobe, room and light into the still, and keep the prompt for what moves. The window example ran at duration_seconds 5 and resolution 720p:
She turns from the rainy window toward the camera; the loose strap slides a little further down her shoulder and she lets it rest there, a slow half smile. Pink and cyan neon ripples over the wet glass and her collarbone. She stays in the velvet dress the whole time. Slow push-in. Rain on the glass, distant traffic hum.
The exact request behind the clip for this endpoint, generated from the opening frame shown next to it.
Draft at 480p, then render
A five-second 480p clip took about one and a half minutes in our tests. Settle the idea there, hold seed, then re-render at 720p or 1080p. Change one field per run so each comparison tells you something.
When the outfit must stay on
No prompt wording reliably stops this model from changing a subject's outfit. For a clip where the wardrobe has to survive, use the standard Wan 3.0 image-to-video and keep Spicy for scenes where that lean is the point.
API quick start
# Generate the idempotency key once; reuse it when retrying after a lost response.
export SPICY_IDEMPOTENCY_KEY="${SPICY_IDEMPOTENCY_KEY:-$(uuidgen)}"
curl -X POST https://api.spicyapi.ai/api/v1/jobs/createTask \
-H "Authorization: Bearer $SPICY_API_KEY" \
-H "Content-Type: application/json" \
-H "Idempotency-Key: $SPICY_IDEMPOTENCY_KEY" \
-d '{
"model": "alibaba/wan-3.0-spicy/image-to-video",
"input": {
"duration_seconds": 5,
"image_url": "https://example.com/reference.png",
"prompt": "A slow aerial shot drifting over a misty pine forest at sunrise",
"resolution": "720p"
}
}'
# Copy data.taskId from the response into TASK_ID, then poll the status.
TASK_ID="TASK_ID_FROM_RESPONSE"
curl "https://api.spicyapi.ai/api/v1/jobs/recordInfo?taskId=$TASK_ID" \
-H "Authorization: Bearer $SPICY_API_KEY"
See the API reference for authentication, the complete schema, polling, and error handling.
