Introduction
Wan 2.6 Flash Image-to-Video is the faster, lower-cost tier of Wan 2.6 image-to-video, published as its own API. It takes an opening frame in image_url and returns exactly 5, 10 or 15 seconds at 720p or 1080p, as one continuous take or a multi-shot edit. Some fine detail is traded for quicker turnaround. Clips are silent by default; switch generate_audio on when a take needs its own soundtrack.
Key features & highlights
The same request as Wan 2.6, with sound as a switch
Flash reads the same fields as Wan 2.6 Image to video, audio_url included: opening frame, prompt, negative prompt, length, resolution, shot type, prompt expansion, seed and a guiding audio track. On top of that, generate_audio decides whether the clip comes back with a generated soundtrack: off, the default, returns picture only at the lower rate; on returns voices, effects and ambience at twice the per-second rate. With it off, audio_url only steers mouth shapes and pacing. A prompt that works here can be sent to Wan 2.6 unchanged when one take has to carry finer texture, so drafting on Flash never means writing the request twice.
Where the speed shows
Flash favours speed and cost over fine detail. Dense texture and fast, complex motion can hold up less well than on Wan 2.6, so review a take before treating it as final. Resolution still matters here: it sets the per-second rate and how long the task takes, with 720p as the default and 1080p available for the takes you keep.
Shape from the frame, cuts from shot_type
There is no aspect ratio field. The clip keeps the proportions of the frame, so crop to 16:9, 9:16 or square before the call. shot_type chooses between single, one continuous take, and multi, which cuts between angles inside the clip and switches prompt expansion on for that request.
Request parameters
| Parameter | Required | Description |
|---|---|---|
image_url | Yes | The opening frame. Subject, wardrobe, set, light direction and the clip's proportions all come from it, so crop it first. |
prompt | No | Optional. One action for a single take, or numbered shots for a multi-shot edit. at least 1 characters · at most 1500 characters |
negative_prompt | No | What must not appear: a second person, captions, a prop the frame keeps bringing back. Up to 500 characters. at most 500 characters |
audio_url | No | A voice line or music cue for the clip to follow: mouth shapes and movement pace themselves to this file. WAV or MP3, 3 to 30 seconds, up to 15 MB. |
generate_audio | No | Off by default, so the clip comes back as picture only. Default: |
resolution | No | 720p or 1080p, default 720p. It sets the per-second rate and how long the task takes. |
duration_seconds | No | Exactly 5, 10 or 15. Multi-shot edits need the longer lengths to breathe. |
shot_type | No | single keeps one continuous take; multi lets the model cut between shots, which suits short stories but breaks continuity. |
enable_prompt_expansion | No | Off by default. Turning it on helps a one-line idea grow into a shot list for multi; leave it off when your prompt is already precise. Default: |
seed | No | 0 to 2147483647. Fix it when comparing single against multi from the same frame. |
Pricing
| Resolution | Generate audio: No | Generate audio: Yes |
|---|---|---|
| $0.0225 / second | $0.045 / second |
| $0.03375 / second | $0.0675 / second |
Use cases
Prompt and shot-list drafts
Run several phrasings of the same move from one frame at 720p and compare them side by side. Quick, low-cost renders make it practical to explore widely, testing single against multi or one camera move against another, before you commit to a final take on either tier.
Variants for feeds and ads
Short vertical loops for product listings, social posts and ad tests, where many takes matter more than the last degree of texture. Crop the frame to 9:16 first, because the clip keeps its proportions, then vary the action line while the frame stays fixed.
Picture for your own sound mix
With generate_audio off, Flash returns picture only, which suits edits where music, voice-over and effects are laid in afterwards anyway. The Flash version of the evening apartment example uses the same three-shot prompt as the Wan 2.6 version without its closing sound line, and still runs 15 seconds at 720p with shot_type set to multi. Turn generate_audio on when a draft should already carry room tone or a spoken line.
Prompt guide & pro tips
Write sound into the prompt only when sound is on
With generate_audio off, a sentence about room tone or a chime has nothing to shape, so spend the 1,500 characters on the action, the camera and the light. The Flash serum example is the Wan 2.6 prompt with its sound note removed:
A single drop falls from the dropper tip into the bottle and sends slow rings across the liquid surface; the light band on the glass brightens a little. Locked macro framing.
With generate_audio on, describe the sounds you want in plain words, after the action.
One action for single, numbered shots for multi
A single take wants one action and one camera move. A multi-shot edit wants numbered shots, each with its own framing, and the longer lengths of 10 or 15 seconds so the cuts have room. Leave enable_prompt_expansion off when your prompt is already precise; it switches on by itself whenever you choose multi.
Hold the variables when you compare
seed accepts 0 to 2147483647. Fix it, together with the frame, when comparing single against multi, so the difference you see comes from the setting you changed. When a take moves to Wan 2.6, send the same frame and prompt so the comparison shows what the finer tier adds.
Check the frame and the output
Send JPEG, PNG without transparency, BMP or WEBP, 240 to 8,000 pixels on each side and no larger than 20 MB. Use negative_prompt, up to 500 characters, for a second person, captions or a prop the frame keeps bringing back. Watch the full clip before delivery, paying most attention to textures and fast movement.
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: $SPICY_IDEMPOTENCY_KEY" \
-d '{
"model": "alibaba/wan-2.6-flash/image-to-video",
"input": {
"duration_seconds": 5,
"generate_audio": false,
"image_url": "https://example.com/reference.png",
"resolution": "720p",
"shot_type": "single"
}
}'
# Copy data.taskId from the response into TASK_ID, then poll the status.
TASK_ID="TASK_ID_FROM_RESPONSE"
curl "https://api.spicyapi.ai/api/v1/jobs/recordInfo?taskId=$TASK_ID" \
-H "Authorization: Bearer $SPICY_API_KEY"
See the API reference for authentication, the complete schema, polling, and error handling.
