Wan 2.5 API

Wan 2.5 Text to video generates a 5- or 10-second clip with sound from a written prompt alone. prompt is the only required field, up to 1,500 characters; everything else has a default: 16:9, 720p, five seconds, prompt expansion off. Use it when no approved still exists yet and the setting, action and camera all have to come from the text. When you already have the opening frame, the image-led endpoint keeps that frame's subject and proportions instead.

Choose a Wan 2.5 mode by starting material

Picture and sound in one pass

The clip comes back with audio already in the file: ambience, effects and voices are generated alongside the picture, so a closing line such as "Audio: gulls, a splash, her breath" is worth writing. To make movement and mouth shapes follow a take you already have, pass audio_url, a WAV or MP3 of 3 to 30 seconds and up to 15 MB. A track longer than the clip is cut at the end; a shorter one leaves the rest of the clip silent.

Five shapes, three resolutions, two lengths

aspect_ratio accepts 16:9, 9:16, 1:1, 4:3 and 3:4, and resolution accepts 480p, 720p and 1080p. The two interact: at 480p only landscape, portrait and square frames render, so 4:3 and 3:4 need 720p or above. duration_seconds is exactly 5 or 10, and a value such as 7 is rejected rather than rounded. Resolution also sets the per-second rate and how long the task takes to return.

Exclusions and controlled comparisons

negative_prompt holds up to 500 characters and is where captions, logos, extra people or a recurring colour belong. seed (0 to 2147483647) keeps the random source steady while you change one field at a time, without promising an identical frame. enable_prompt_expansion is off by default; switched on, it rewrites a short brief into a fuller one before generation.

Vertical clips with the soundtrack inside

Set aspect_ratio to 9:16 and keep five seconds for a feed draft that needs no separate audio pass. The published record-shop example goes further: ten seconds at 1080p in 9:16, a slow tilt from a crate of vinyl up to the man's face, and a negative prompt of "text on screen, logos, extra people" so no captions or passers-by appear.

Checking sound against a visible action

The saxophone example asks for ten seconds at 1080p in 16:9, with the horn's sound matching the fingering. Tying the audio to something you can see (keys, a splash, a turntable crackle) gives you a concrete check on playback instead of judging ambience alone.

A spoken line on your own timing

Record the line, pass it as audio_url, and describe who is speaking and where. Movement and mouth shapes follow the file's timing, so the review can focus on the performance. When the speaker has to be a specific face or product, text alone will not hold that likeness between runs; start from a still on the image-to-video endpoint.

# 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-2.5/text-to-video",
  "input": {
    "aspect_ratio": "16:9",
    "duration_seconds": 5,
    "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.