Wan 2.6 T2V Spark

alibaba/wan2.6-t2v-spark — a video model by Alibaba. The full parameter schema is reproduced in full below.

Catalog slug	alibaba/wan2.6-t2v-spark
Kind	video
Vendor	Alibaba

How to call

Video generation is asynchronous (create → poll → result). Address it by its catalog slug through the standard video door — the gateway resolves the slug and routes it internally; the inference route is never exposed.

Through the Sociaro SDK / POST /v1/videos door

The SDK’s video.generate drives the whole cycle and blocks until a terminal state:

from sociaro_ai import Sociaro

client = Sociaro(api_key="gw_live_...")

result = client.video.generate(
    model="alibaba/wan2.6-t2v-spark",
    prompt="a timelapse of a blooming flower",
)
print(result.url)

Additional parameters (duration, size/resolution, etc.) are forwarded to the upstream as-is — use the exact field name and value format from the schema below. In particular the legacy size for this model is a width*height string such as 1280*720 (an asterisk, not a tier or x).

For a non-blocking handle use job = client.video.submit(model="alibaba/wan2.6-t2v-spark", …) then job.result(). Raw door: POST https://api.sociaro.com/v1/videos (create), then poll GET /v1/videos/{id} — see async jobs and media doors.

Parameters — full schema

The complete, one-to-one API schema for this model, transcribed from the official source. Nothing omitted. Examples in the schema may use a representative model of the family — for this model use the slug alibaba/wan2.6-t2v-spark as shown in How to call above.

COVERAGE MAP — which official page documents each requested gateway slug

gateway slug	documented on page	task family	endpoint	request body style
happyhorse-1.0-t2v	happyhorse-text-to-video-api-reference	text-to-video	video-generation/video-synthesis	input.prompt + parameters
happyhorse-1.0-i2v	happyhorse-image-to-video-api-reference	image-to-video (first frame)	video-generation/video-synthesis	input.prompt + input.media[] (first_frame)
happyhorse-1.0-r2v	happyhorse-reference-to-video-api-reference	reference-to-video (multi-image)	video-generation/video-synthesis	input.prompt + input.media[] (reference_image 1–9)
happyhorse-1.0-video-edit	happyhorse-video-edit-api-reference	video editing	video-generation/video-synthesis	input.prompt + input.media[] (video + reference_image 0–5)
wan2.2-animate-mix	wan-animate-mix-api	video character swap	image2video/video-synthesis	flat input.image_url + input.video_url; parameters.mode
wan2.2-animate-move	wan-animate-move-api	video motion transfer	image2video/video-synthesis	flat input.image_url + input.video_url; parameters.mode
wan2.2-kf2v-flash	legacy-image-to-video-by-first-and-last-frame-api-reference	keyframe (first+last frame)	image2video/video-synthesis	flat input.first_frame_url + input.last_frame_url
wan2.6-i2v	legacy-image-to-video-api-reference	image-to-video (first frame, LEGACY)	video-generation/video-synthesis	flat input.img_url (+ audio_url); parameters.shot_type
wan2.6-i2v-flash	legacy-image-to-video-api-reference	image-to-video (first frame, LEGACY)	video-generation/video-synthesis	flat input.img_url (+ audio_url); parameters.shot_type, parameters.audio
wan2.6-r2v	use-video-generation (capability matrix)	reference-to-video	video-generation/video-synthesis	media[]-style reference-to-video; request-parameter contract not published in a dedicated API reference
wan2.6-r2v-flash	use-video-generation (capability matrix)	reference-to-video	video-generation/video-synthesis	media[]-style reference-to-video; request-parameter contract not published in a dedicated API reference
wan2.6-t2v	text-to-video-api-reference (LEGACY protocol section)	text-to-video	video-generation/video-synthesis	input.prompt (+ audio_url); parameters.size, shot_type
wan2.7-i2v	image-to-video-general-api-reference	image-to-video (1st frame / 1st+last / continuation)	video-generation/video-synthesis	input.prompt + input.media[] (first_frame/last_frame/driving_audio/first_clip)
wan2.7-r2v	wan-video-to-video-api-reference	reference-to-video	video-generation/video-synthesis	input.prompt + input.media[] (reference_image/reference_video/first_frame + reference_voice)
wan2.7-t2v	text-to-video-api-reference (NEW protocol section)	text-to-video	video-generation/video-synthesis	input.prompt (+ audio_url); parameters.resolution/ratio
wan2.7-videoedit	wan-video-editing-api-reference	video editing	video-generation/video-synthesis	input.prompt + input.media[] (video + reference_image 0–4); audio_setting

Slug naming: the gateway slug is wan2.7-videoedit (no hyphen between video & edit), matching the DashScope model id wan2.7-videoedit. The HappyHorse edit slug is happyhorse-1.0-video-edit (hyphenated). These conventions differ and are not normalized.

The wan2.7-i2v model id appears in dated form wan2.7-i2v-2026-04-25 in examples; the un-dated wan2.7-i2v alias is also accepted.

---

CRITICAL PER-MODEL / PER-SUB-FAMILY DIFFERENCES (do not mix parameter sets)

1. TWO endpoints, NOT one. Most families post to POST .../api/v1/services/aigc/video-generation/video-synthesis, but the animate family (wan2.2-animate-move, wan2.2-animate-mix) AND the keyframe family (wan2.2-kf2v-flash, wan2.1-kf2v-plus) post to a DIFFERENT path: POST .../api/v1/services/aigc/image2video/video-synthesis. All poll via the SAME GET .../api/v1/tasks/{task_id}.

2. THREE request-body styles under input:

   • media[] array (NEW protocol): wan2.7-i2v, wan2.7-r2v, wan2.7-videoedit, all 4 HappyHorse models. Each element { "type": ..., "url": ... } (+ optional reference_voice on wan2.7-r2v).
   • flat scalar URLs (LEGACY i2v): wan2.6-i2v / wan2.6-i2v-flash use input.img_url (+ audio_url).
   • flat scalar URLs (animate / kf2v): input.image_url+input.video_url (animate) or input.first_frame_url+input.last_frame_url (kf2v).
   • prompt-only input (t2v): wan2.7-t2v, wan2.6-t2v, happyhorse-1.0-t2v (+ optional audio_url).

3. parameters.size (legacy) vs parameters.resolution+ratio (new).

   • LEGACY t2v (wan2.6/2.5/2.2/2.1) and LEGACY i2v use size = explicit width*height string (e.g. 1280*720).
   • NEW t2v (wan2.7), i2v, r2v, videoedit, and ALL HappyHorse use resolution (tier 720P/1080P/480P) + ratio (where applicable).
   • kf2v uses resolution (tier). animate uses neither (output follows input; only mode).

4. resolution allowed tiers & defaults differ:

   • wan2.7-t2v / wan2.7-i2v / wan2.7-r2v / wan2.7-videoedit: 720P, 1080P — default 1080P (i2v/r2v/videoedit) / default 1080P (t2v).
   • wan2.6-i2v / wan2.6-i2v-flash: 720P, 1080P — default 1080P.
   • wan2.6-t2v (size-based): all 720p & 1080p resolutions; default 1920*1080.
   • wan2.6-r2v / wan2.6-r2v-flash: 720P, 1080P.
   • wan2.2-kf2v-flash: 480P, 720P, 1080P — default 720P; wan2.1-kf2v-plus: 720P only.
   • happyhorse-1.0-t2v / -i2v / -r2v / -video-edit: 720P, 1080P — default 1080P.
   • animate-move / animate-mix: NO resolution param (mode wan-std/wan-pro controls quality tier).

5. ratio parameter presence & enum:

   • wan2.7-t2v / wan2.7-r2v / wan2.7-videoedit: 16:9(default), 9:16, 1:1, 4:3, 3:4 (5 values).
   • wan2.7-i2v: NO ratio (output aspect follows the input first_frame / first_clip).
   • happyhorse-1.0-t2v / -r2v: EXTENDED enum 16:9(default), 9:16, 1:1, 4:3, 3:4, 4:5, 5:4, 9:21, 21:9 (9 values).
   • happyhorse-1.0-i2v: NO ratio (follows input image).
   • r2v ratio is ignored if a first_frame is provided (wan2.7-r2v) — output then follows the first frame.

6. duration range / fixed-value differs:

   • wan2.7-t2v / wan2.7-i2v: integer 2–15, default 5.
   • wan2.7-r2v: default 5; 2–10 if reference material includes a video, 2–15 if not.
   • wan2.7-videoedit: integer 2–10, default 0 (= keep input duration; nonzero truncates from 0s).
   • wan2.6-t2v: integer 2–15, default 5. (wan2.5-t2v-preview 5/10; wan2.2/2.1-t2v fixed 5.)
   • wan2.6-i2v / wan2.6-i2v-flash: integer 2–15, default 5. (wan2.6-i2v-us: 5/10/15; wan2.5-i2v 5/10; wan2.2-i2v fixed 5; wan2.1-i2v-turbo 3/4/5.)
   • wan2.6-r2v / -flash: integer 2–10.
   • wan2.2-kf2v-flash / wan2.1-kf2v-plus: fixed 5 (no other value).
   • happyhorse-1.0-t2v / -i2v / -r2v: integer 3–15, default 5. happyhorse-1.0-video-edit: no duration param (output 3–15, follows input; >15s input truncated to first 15s).
   • animate-move / animate-mix: no duration param (output follows reference video duration).

7. AUDIO support (input audio_url / driving_audio AND audio output):

   • wan2.7-t2v: optional input.audio_url (drives lip-sync/timing); without it the model auto-generates BGM/SFX. Audio: WAV/MP3, 2–30s, ≤15 MB.
   • wan2.6-t2v / wan2.5-t2v: optional input.audio_url (wan2.6/2.5 only). wan2.2/2.1-t2v are SILENT by default (no audio).
   • wan2.7-i2v: optional driving_audio media element (2–30s, ≤15 MB).
   • wan2.6-i2v / wan2.5-i2v: optional input.audio_url (3–30s, ≤15 MB). wan2.6-i2v-flash adds parameters.audio (bool, default true) to force silent output.
   • wan2.7-r2v: per-entity reference_voice (1–10s, ≤15 MB) on each media element.
   • wan2.6-r2v: video with audio. wan2.6-r2v-flash: video with audio or silent video.
   • videoedit (wan2.7 & happyhorse): parameters.audio_setting = auto(default) | origin (keep input video audio).
   • HappyHorse t2v / i2v / r2v: NO audio input/output param documented (silent video unless future).
   • animate-move / animate-mix / kf2v: NO audio param.

8. prompt requiredness & length differ:

   • Required: wan2.7-t2v, wan2.6-t2v, wan2.7-r2v, wan2.7-videoedit, happyhorse-1.0-t2v, happyhorse-1.0-r2v, happyhorse-1.0-video-edit, kf2v.
   • Optional: wan2.7-i2v, wan2.6-i2v (legacy), happyhorse-1.0-i2v.
   • Length: wan2.7 t2v/i2v/r2v/videoedit = up to 5,000 chars. wan2.6/2.5 = 1,500 chars. wan2.2/2.1 = 800 chars. HappyHorse = 5,000 non-Chinese / 2,500 Chinese chars (any language).
   • No prompt at all: animate-move / animate-mix (no prompt field).

9. prompt_extend presence: present (default true) on wan2.7 t2v/i2v/r2v/videoedit, wan2.6 t2v/i2v, kf2v. ABSENT on HappyHorse (all 4), animate-move, animate-mix.

10. watermark default differs: false (default) on ALL Wan models (text “AI Generated”; animate text “Generated by Qwen AI” on move-API, “AI Generated” on mix-API). true (default) on ALL HappyHorse models (text “HappyHorse” / “Happy Horse”). watermark is in parameters for media-style families but in input for animate-move/animate-mix.

11. seed is universal across every page: integer [0, 2147483647], optional, random if omitted; same seed yields only similar (not identical) output. (Not present on animate-move/animate-mix.)

12. shot_type (single/multi, default single) is wan2.6-series ONLY (wan2.6-t2v, wan2.6-i2v). Effective only when prompt_extend=true; shot_type overrides prompt-stated shot count. Absent everywhere else (wan2.7 controls shots via natural-language prompt; its shot_type “has no effect”).

13. animate-only parameters: mode (REQUIRED; wan-std | wan-pro) + check_image (bool, default true). These are the ONLY two animate parameters. parameters object is REQUIRED for animate (because mode is required).

14. Response output shape differs between the two endpoints:

    • video-generation endpoint (t2v/i2v/r2v/videoedit/kf2v/happyhorse): output.video_url is a top-level field under output.
    • image2video endpoint (animate-move/animate-mix): output.results.video_url is nested under results.

15. usage object differs by family:

    • NEW media-style (wan2.7 + happyhorse t2v/i2v/r2v/videoedit): { duration, input_video_duration, output_video_duration, video_count, SR } (+ ratio for t2v/r2v). duration = total billed seconds = input+output. videoedit returns floats.
    • LEGACY t2v: wan2.6 → { duration, size, input_video_duration, output_video_duration, video_count, SR }; wan2.5-and-earlier → { video_duration, video_ratio, video_count }.
    • LEGACY i2v: { input_video_duration, output_video_duration, duration, SR, video_count }.
    • kf2v: { video_duration (fixed 5), video_count, SR (2.2 only) / video_ratio (2.1 only, ="standard") }.
    • animate-move / animate-mix: { video_duration (float), video_ratio } where video_ratio = service mode (standard for wan-std, pro for wan-pro).

16. Regions / endpoints (per-region API keys; cross-region calls FAIL):

    • Singapore (recommended, workspace domain): https://{WorkspaceId}.ap-southeast-1.maas.aliyuncs.com (legacy https://dashscope-intl.aliyuncs.com deprecating).
    • Beijing: https://dashscope.aliyuncs.com.
    • US (Virginia): https://dashscope-us.aliyuncs.com — legacy t2v/i2v (wan2.6-*-us), HappyHorse t2v/i2v/r2v/edit.
    • Germany (Frankfurt): https://{WorkspaceId}.eu-central-1.maas.aliyuncs.com — legacy t2v/i2v (async-only), HappyHorse.
    • Region availability is per-model — animate / kf2v / new-protocol pages list only Singapore + Beijing; HappyHorse & legacy t2v/i2v add Virginia + Frankfurt.

17. wan2.6-t2v-flash does NOT exist (no t2v-flash slug in catalog). -flash variants exist for i2v (wan2.6-i2v-flash) and r2v (wan2.6-r2v-flash) only in the wan2.6 series; kf2v has wan2.2-kf2v-flash.

---

ENDPOINTS

Operation	Method + path
Create task (video-generation families)	POST {base}/api/v1/services/aigc/video-generation/video-synthesis
Create task (animate + kf2v families)	POST {base}/api/v1/services/aigc/image2video/video-synthesis
Poll task result (ALL families)	GET {base}/api/v1/tasks/{task_id}

{base} = one of the region hosts in difference #16.

• Auth: header Authorization: Bearer $DASHSCOPE_API_KEY (region-specific key).
• Content-Type: application/json.
• Async-only over HTTP: header X-DashScope-Async: enable is REQUIRED on every Create call. Omitting it returns error "current user api does not support synchronous calls".
• Two-step async: Create returns only { output:{ task_status:"PENDING", task_id }, request_id }. The real result (video_url, usage, actual duration/ratio/SR) appears ONLY via poll/callback.
• task_id validity: 24 hours. video_url validity: 24 hours (then purged — save immediately). Video output: MP4, H.264.
• Poll interval recommendation: ~15 s. State machine: PENDING → RUNNING → SUCCEEDED | FAILED (also CANCELED; UNKNOWN = task not found / expired).
• DashScope SDK (Python VideoSynthesis, Java VideoSynthesis) wraps the async flow (call / async_call / fetch / wait). Python SDK min versions: 1.23.8 (kf2v), 1.25.8 (legacy t2v), 1.25.16 (i2v/r2v/videoedit). wan2.7 is NOT supported by the Python SDK for t2v (HTTP only) per the t2v page; i2v/r2v/videoedit SDK support wan2.7.

---

COMMON CREATE HEADERS (all families)

Header	Type	Req	Value
Content-Type	string	yes	application/json
Authorization	string	yes	Bearer sk-xxxx
X-DashScope-Async	string	yes	enable (Create only; HTTP is async-only)

COMMON POLL (GET …/tasks/{task_id})

• Header: Authorization: Bearer sk-xxxx.
• Path param: task_id (string, required).
• No body.

---

SUB-FAMILY 1 — TEXT-TO-VIDEO (t2v)

Slugs: wan2.7-t2v (NEW protocol), wan2.6-t2v (LEGACY protocol), happyhorse-1.0-t2v. Endpoint: video-generation/video-synthesis.

1a. wan2.7-t2v — request body

{
  "model": "wan2.7-t2v",                 // REQUIRED
  "input": {                              // REQUIRED
    "prompt": "...",                      // REQUIRED, ≤5000 chars (zh/en); controls shots via natural language
    "negative_prompt": "...",            // optional, ≤500 chars
    "audio_url": "https://.../x.mp3"      // optional; WAV/MP3, 2–30s, ≤15MB; absent ⇒ auto BGM/SFX
  },
  "parameters": {                         // optional
    "resolution": "1080P",               // optional; 720P | 1080P; default 1080P
    "ratio": "16:9",                     // optional; 16:9(def) | 9:16 | 1:1 | 4:3 | 3:4
    "duration": 5,                       // optional; integer 2–15; default 5
    "prompt_extend": true,               // optional; default true
    "watermark": false,                  // optional; default false ("AI Generated")
    "seed": 12345                        // optional; [0, 2147483647]
  }
}

Output-resolution table (wan2.7-t2v), width×height by tier×ratio:

Tier	16:9	9:16	1:1	4:3	3:4
720P	1280×720	720×1280	960×960	1104×832	832×1104
1080P	1920×1080	1080×1920	1440×1440	1648×1248	1248×1648

1b. wan2.6-t2v — request body (LEGACY)

{
  "model": "wan2.6-t2v",                 // REQUIRED
  "input": {                              // REQUIRED
    "prompt": "...",                      // REQUIRED, ≤1500 chars (wan2.6/2.5); ≤800 (wan2.2/2.1)
    "negative_prompt": "...",            // optional, ≤500 chars
    "audio_url": "https://.../x.mp3"      // optional (wan2.6 & wan2.5 only); 3–30s, ≤15MB
  },
  "parameters": {                         // optional
    "size": "1280*720",                  // optional; "width*height" string (NOT a tier/ratio). wan2.6-t2v default 1920*1080; valid = all 720p & 1080p
    "duration": 5,                       // optional; wan2.6-t2v integer 2–15 (def 5). wan2.6-t2v-us 5/10/15. wan2.5 5/10. wan2.2/2.1 fixed 5.
    "prompt_extend": true,               // optional; default true
    "shot_type": "single",               // optional (wan2.6 only); single(def) | multi; needs prompt_extend=true
    "watermark": false,                  // optional; default false
    "seed": 11                           // optional
  }
}

LEGACY size valid resolutions (per tier):

• 480p: 832*480(16:9), 480*832(9:16), 624*624(1:1).
• 720p: 1280*720(16:9), 720*1280(9:16), 960*960(1:1), 1088*832(4:3), 832*1088(3:4).
• 1080p: 1920*1080(16:9), 1080*1920(9:16), 1440*1440(1:1), 1632*1248(4:3), 1248*1632(3:4).

1c. happyhorse-1.0-t2v — request body

{
  "model": "happyhorse-1.0-t2v",         // REQUIRED
  "input": { "prompt": "..." },          // REQUIRED; prompt REQUIRED, ≤5000 non-zh / ≤2500 zh, any language
  "parameters": {                         // optional
    "resolution": "1080P",               // optional; 720P | 1080P; default 1080P
    "ratio": "16:9",                     // optional; 16:9(def),9:16,1:1,4:3,3:4,4:5,5:4,9:21,21:9 (9 values)
    "duration": 5,                       // optional; integer 3–15; default 5
    "watermark": true,                   // optional; DEFAULT TRUE ("HappyHorse")
    "seed": 12345                        // optional
  }
}

happyhorse-1.0-t2v has NO negative_prompt, NO prompt_extend, NO audio_url.

---

SUB-FAMILY 2 — IMAGE-TO-VIDEO (i2v)

Slugs: wan2.7-i2v (NEW, media[]), wan2.6-i2v & wan2.6-i2v-flash (LEGACY, flat img_url), happyhorse-1.0-i2v (NEW, media[] first_frame only). Endpoint: video-generation/video-synthesis.

2a. wan2.7-i2v — request body (NEW, media[])

wan2.7-i2v performs THREE tasks via media[] combinations:

• first-frame-to-video: first_frame (± driving_audio)
• first-and-last-frame: first_frame + last_frame (± driving_audio)
• video continuation: first_clip (± last_frame)

{
  "model": "wan2.7-i2v",                 // REQUIRED (dated: wan2.7-i2v-2026-04-25)
  "input": {                              // REQUIRED
    "prompt": "...",                      // OPTIONAL, ≤5000 chars
    "negative_prompt": "...",            // optional, ≤500 chars
    "media": [                            // REQUIRED; each {type,url}; each type at most once
      { "type": "first_frame",   "url": "https://.../a.png" },
      { "type": "driving_audio", "url": "https://.../a.mp3" }
    ]
  },
  "parameters": {                         // optional
    "resolution": "1080P",               // optional; 720P | 1080P; default 1080P
    "duration": 5,                       // optional; integer 2–15; default 5
    "prompt_extend": true,               // optional; default true
    "watermark": false,                  // optional; default false
    "seed": 12345                        // optional
  }
}

• NO ratio — output aspect ratio follows the input first_frame / first_clip (may drift to nearest multiple-of-16 size).
• media type enum: first_frame, last_frame, driving_audio, first_clip.
• Valid combinations ONLY: first_frame | first_frame+driving_audio | first_frame+last_frame | first_frame+last_frame+driving_audio | first_clip | first_clip+last_frame.
• Image (first_frame/last_frame): JPEG/JPG/PNG(no alpha)/BMP/WEBP, side 240–8000 px, aspect 1:8–8:1, ≤20 MB. URL or Base64 data:{MIME};base64,{data}.
• Audio (driving_audio): WAV/MP3, 2–30s, ≤15 MB.
• Video (first_clip): MP4/MOV, 2–10s, side 240–4096 px, aspect 1:8–8:1, ≤100 MB. With first_clip, duration = max TOTAL output (e.g. duration=15 + 3s input ⇒ 12s continuation, billed 15s).

2b. wan2.6-i2v / wan2.6-i2v-flash — request body (LEGACY, flat)

{
  "model": "wan2.6-i2v-flash",           // REQUIRED (or wan2.6-i2v)
  "input": {                              // REQUIRED
    "prompt": "...",                      // OPTIONAL, ≤1500 chars (wan2.6/2.5); ≤800 (wan2.2/2.1)
    "negative_prompt": "...",            // optional, ≤500 chars
    "img_url": "https://.../a.png",       // REQUIRED; first/init frame. JPEG/JPG/PNG(no alpha)/BMP/WEBP, side 240–8000px, ≤20MB (wan2.6/2.5) / ≤10MB (wan2.2/2.1). URL or Base64.
    "audio_url": "https://.../a.mp3"      // optional (wan2.6 & wan2.5 only); 3–30s, ≤15MB
  },
  "parameters": {                         // optional
    "resolution": "1080P",               // optional; wan2.6-i2v & -flash: 720P|1080P, default 1080P
    "duration": 5,                       // optional; wan2.6-i2v & -flash: integer 2–15, default 5
    "prompt_extend": true,               // optional; default true
    "shot_type": "single",               // optional (wan2.6 series only); single(def)|multi; needs prompt_extend=true
    "audio": true,                       // optional (wan2.6-i2v-flash ONLY); true(def)=with sound, false=silent; priority audio > audio_url; affects price
    "watermark": false,                  // optional; default false
    "seed": 12345                        // optional
  }
}

• Output aspect ratio is kept close to the input img_url aspect (no ratio/size for the new-tier resolution field on these models — they use the tier resolution, not legacy size).
• Other legacy i2v models on the same page: wan2.6-i2v-us (5/10/15), wan2.5-i2v-preview (480P/720P/1080P, 5/10), wan2.2-i2v-flash (480P/720P, fixed 5), wan2.2-i2v-plus (480P/1080P, fixed 5), wan2.1-i2v-turbo (480P/720P, 3/4/5), wan2.1-i2v-plus (720P, fixed 5).

2c. happyhorse-1.0-i2v — request body (NEW, media[] first_frame only)

{
  "model": "happyhorse-1.0-i2v",         // REQUIRED
  "input": {                              // REQUIRED
    "prompt": "...",                      // OPTIONAL, ≤5000 non-zh / ≤2500 zh, any language
    "media": [ { "type": "first_frame", "url": "https://.../a.png" } ]  // REQUIRED; exactly ONE first_frame
  },
  "parameters": {                         // optional
    "resolution": "1080P",               // optional; 720P | 1080P; default 1080P
    "duration": 5,                       // optional; integer 3–15; default 5
    "watermark": true,                   // optional; DEFAULT TRUE ("Happy Horse")
    "seed": 12345                        // optional
  }
}

• media type enum: first_frame ONLY (exactly one required).
• Image: JPEG/JPG/PNG/WEBP, width & height ≥300 px, aspect 1:2.5–2.5:1, ≤20 MB. URL or Base64.
• NO ratio (follows input image), NO negative_prompt, NO prompt_extend, NO audio.

---

SUB-FAMILY 3 — REFERENCE-TO-VIDEO (r2v)

Slugs: wan2.7-r2v (NEW, rich media[]), wan2.6-r2v & wan2.6-r2v-flash, happyhorse-1.0-r2v (NEW, reference_image only). Endpoint: video-generation/video-synthesis.

3a. wan2.7-r2v — request body (NEW)

{
  "model": "wan2.7-r2v",                 // REQUIRED
  "input": {                              // REQUIRED
    "prompt": "Image 1 ... Video 1 ...", // REQUIRED, ≤5000 chars; use "Image 1/2", "Video 1/2" identifiers (space + capital) matching media order; single ref ⇒ "the reference image/video"
    "negative_prompt": "...",            // optional, ≤500 chars
    "media": [                            // REQUIRED; each {type,url}(+reference_voice)
      { "type": "reference_image", "url": "https://.../g.jpg", "reference_voice": "https://.../v.mp3" },
      { "type": "reference_video", "url": "https://.../r.mp4", "reference_voice": "https://.../b.mp3" }
    ]
  },
  "parameters": {                         // optional
    "resolution": "1080P",               // optional; 720P | 1080P; default 1080P
    "ratio": "16:9",                     // optional; 16:9(def),9:16,1:1,4:3,3:4 — IGNORED if a first_frame is provided (then follows first frame)
    "duration": 5,                       // optional; default 5; 2–10 if media includes a video, else 2–15
    "prompt_extend": true,               // optional; default true
    "watermark": false,                  // optional; default false
    "seed": 12345                        // optional
  }
}

• media type enum: reference_image, reference_video, first_frame.
  • reference_image: subject/scene; single character if a subject. JPEG/JPG/PNG(no alpha)/BMP/WEBP, 240–8000px, 1:8–8:1, ≤20MB. URL or Base64.
  • reference_video: subject + optional voice; avoid empty scenes; single character. MP4/MOV, 1–30s, 240–4096px, 1:8–8:1, ≤100MB.
  • first_frame: at most 1; video starts from it; combine with subject refs.
• Asset limits: max 1 first_frame; ≥1 reference_image or reference_video; reference_images + reference_videos ≤ 5.
• reference_voice (optional, per element): WAV/MP3, 1–10s, ≤15MB. Voice reference only (not spoken content). Priority: explicit reference_voice > original reference_video audio.
• usage adds ratio; duration = input_video_duration + output_video_duration (billed total).

3b. wan2.6-r2v / wan2.6-r2v-flash

• Capability: wan2.6-r2v generates video with audio, with single-role / multi-role generation, multi-shot narrative, and audio-video synchronization. wan2.6-r2v-flash additionally supports silent video and is faster and more cost-effective.
• Input modality: text, image, video. Resolution: 720P, 1080P. Duration: integer [2s, 10s]. 30 fps. Output: MP4 (H.264).
• Endpoint: video-generation/video-synthesis (same async create/poll flow as the other reference-to-video models).
• Alibaba Cloud does not publish a dedicated request-parameter reference for these two models; the per-field request contract is therefore not enumerated here. Use the Video generation overview for the capability matrix above. (The wan-video-to-video-api-reference page documents the request parameters for wan2.7-r2v only.)

3c. happyhorse-1.0-r2v — request body (NEW)

{
  "model": "happyhorse-1.0-r2v",         // REQUIRED
  "input": {                              // REQUIRED
    "prompt": "... [Image 1] ... [Image 2] ...", // REQUIRED, ≤5000 non-zh/≤2500 zh; use "[Image 1]","[Image 2]" identifiers matching media order
    "media": [                            // REQUIRED; reference_image only, 1–9 elements
      { "type": "reference_image", "url": "https://.../g.jpg" }
    ]
  },
  "parameters": {                         // optional
    "resolution": "1080P",               // optional; 720P | 1080P; default 1080P
    "ratio": "16:9",                     // optional; 16:9(def),9:16,3:4,4:3,4:5,5:4,1:1,9:21,21:9 (9 values)
    "duration": 5,                       // optional; integer 3–15; default 5
    "watermark": true,                   // optional; DEFAULT TRUE ("Happy Horse")
    "seed": 12345                        // optional
  }
}

• media type: reference_image ONLY; 1–9 images. JPEG/JPG/PNG/WEBP, shortest side ≥400px (720P+ recommended), ≤20MB. URL or Base64.
• Prompt syntax uses bracketed [Image N] (HappyHorse) vs un-bracketed Image N/Video N (wan2.7-r2v).
• NO reference_video, NO first_frame, NO reference_voice, NO negative_prompt, NO prompt_extend, NO audio.

---

SUB-FAMILY 4 — VIDEO EDITING (video-edit)

Slugs: wan2.7-videoedit, happyhorse-1.0-video-edit. Endpoint: video-generation/video-synthesis.

4a. wan2.7-videoedit — request body

{
  "model": "wan2.7-videoedit",           // REQUIRED
  "input": {                              // REQUIRED
    "prompt": "...",                      // REQUIRED, ≤5000 chars; instruction (edit / style transfer)
    "negative_prompt": "...",            // optional, ≤500 chars
    "media": [                            // REQUIRED; exactly 1 video + 0–4 reference_image
      { "type": "video",           "url": "https://.../v.mp4" },
      { "type": "reference_image", "url": "https://.../r.png" }
    ]
  },
  "parameters": {                         // optional
    "resolution": "1080P",               // optional; 720P | 1080P; default 1080P
    "ratio": "16:9",                     // optional; 16:9,9:16,1:1,4:3,3:4 — omit ⇒ follows input video aspect
    "duration": 0,                       // optional; integer 2–10; DEFAULT 0 = keep input length; nonzero truncates from 0s
    "audio_setting": "auto",             // optional; auto(def)=model decides (may regen audio) | origin=keep input audio
    "prompt_extend": true,               // optional; default true
    "watermark": false,                  // optional; default false
    "seed": 12345                        // optional
  }
}

• media type enum: video (exactly 1, REQUIRED) + reference_image (0–4, optional).
• Video: MP4/MOV, 2–10s, side 240–4096px, aspect 1:8–8:1, ≤100MB.
• reference_image: JPEG/JPG/PNG(no alpha)/BMP/WEBP, side 240–8000px, aspect 1:8–8:1, ≤20MB. URL or Base64.
• usage returns floats; duration = input_video_duration + output_video_duration.

4b. happyhorse-1.0-video-edit — request body

{
  "model": "happyhorse-1.0-video-edit",  // REQUIRED
  "input": {                              // REQUIRED
    "prompt": "...",                      // REQUIRED, ≤5000 non-zh/≤2500 zh; style transfer / local replacement instruction
    "media": [                            // REQUIRED; exactly 1 video + 0–5 reference_image
      { "type": "video",           "url": "https://.../v.mp4" },
      { "type": "reference_image", "url": "https://.../r.webp" }
    ]
  },
  "parameters": {                         // optional
    "resolution": "1080P",               // optional; 720P | 1080P; default 1080P
    "audio_setting": "auto",             // optional; auto(def) | origin (keep input audio)
    "watermark": true,                   // optional; DEFAULT TRUE ("Happy Horse")
    "seed": 12345                        // optional
  }
}

• media type enum: video (exactly 1, REQUIRED) + reference_image (0–5, optional).
• Video: MP4/MOV (H.264 recommended), 3–60s input, side ≤4096 / ≥360px, aspect 1:2.5–2.5:1, ≤100MB, fps >8. Output 3–15s: ≤15s input → same length; >15s input → first 15s only.
• reference_image: JPEG/JPG/PNG/WEBP, width & height ≥300px, aspect 1:2.5–2.5:1, ≤20MB. URL or Base64.
• NO ratio, NO duration, NO prompt_extend, NO negative_prompt (vs wan2.7-videoedit which has all four).

---

SUB-FAMILY 5 — ANIMATE (motion transfer / character swap)

Slugs: wan2.2-animate-move (motion transfer), wan2.2-animate-mix (character swap). Endpoint: image2video/video-synthesis (DIFFERENT from all above). Request body shape is IDENTICAL for move and mix; only the model name and semantics differ.

{
  "model": "wan2.2-animate-move",        // REQUIRED (or wan2.2-animate-mix)
  "input": {                              // REQUIRED
    "image_url": "https://.../char.jpeg", // REQUIRED; character image. JPG/JPEG/PNG/BMP/WEBP, side 200–4096px, aspect 1:3–3:1, ≤5MB. URL must be ASCII (encode non-ASCII).
    "video_url": "https://.../ref.mp4",   // REQUIRED; reference video. MP4/AVI/MOV, side 200–2048px, aspect 1:3–3:1, 2–30s, ≤200MB.
    "watermark": true                     // OPTIONAL — note: in `input`, NOT in `parameters`. default false (move: "Generated by Qwen AI"; mix: "AI Generated")
  },
  "parameters": {                         // REQUIRED (because `mode` is required)
    "mode": "wan-std",                   // REQUIRED; wan-std (fast/cheap) | wan-pro (smoother/higher quality)
    "check_image": true                  // optional; default true; validate input image before processing
  }
}

• NO prompt, NO negative_prompt, NO resolution, NO ratio, NO duration, NO seed, NO prompt_extend.
• Difference move vs mix:
  • animate-move = “video motion transfer” (animate the character in the image using motion from the reference video; keeps photo background static).
  • animate-mix = “video character swap” (replace the person in the video with the person from the image; keeps original background).
• Content moderation on all inputs/outputs (IPInfringementSuspect / DataInspectionFailed on violation).

Animate response (poll) — note nested results

{
  "request_id": "...",
  "output": {
    "task_id": "...",
    "task_status": "SUCCEEDED",
    "submit_time": "2025-09-18 15:32:00.105",
    "scheduled_time": "2025-09-18 15:32:15.066",
    "end_time": "2025-09-18 15:34:41.898",
    "results": { "video_url": "http://...mp4?Expires=..." }   // ← nested under results
  },
  "usage": { "video_duration": 5.2, "video_ratio": "standard" } // video_ratio = mode: standard|pro
}

• Billing: output only, per output-video second; failed calls free.

---

SUB-FAMILY 6 — KEYFRAME (first-and-last-frame, kf2v)

Slugs: wan2.2-kf2v-flash, wan2.1-kf2v-plus. Endpoint: image2video/video-synthesis (same as animate).

{
  "model": "wan2.2-kf2v-flash",          // REQUIRED (or wan2.1-kf2v-plus)
  "input": {                              // REQUIRED
    "prompt": "...",                      // REQUIRED; describe transition (camera/subject motion). zh/en
    "negative_prompt": "...",            // optional, ≤500 chars
    "first_frame_url": "https://.../f.png", // REQUIRED; output aspect follows this frame. JPEG/JPG/PNG(no alpha)/BMP/WEBP, side 240–8000px, ≤10MB
    "last_frame_url":  "https://.../l.png"  // REQUIRED; need not match first-frame resolution/aspect. same format limits, ≤10MB
  },
  "parameters": {                         // optional
    "resolution": "720P",                // optional; wan2.2-kf2v-flash: 480P|720P|1080P (default 720P); wan2.1-kf2v-plus: 720P only
    "duration": 5,                       // optional; FIXED 5 (only valid value)
    "prompt_extend": true,               // optional; default true
    "watermark": false,                  // optional; default false ("AI-generated")
    "seed": 12345                        // optional
  }
}

• Inputs are flat first_frame_url/last_frame_url (NOT a media[] array). Also accepts local file:// paths via SDK.
• usage: { video_duration: 5, video_count: 1, SR (2.2 only) | video_ratio="standard" (2.1 only) }.
• Response returns output.video_url top-level (video-generation-style), plus actual_prompt when prompt_extend=true.

---

CREATE RESPONSE (all video-generation & image2video families)

{
  "output": { "task_status": "PENDING", "task_id": "0385dc79-..." },
  "request_id": "4909100c-..."
}

On creation failure (top-level, no output.task_id):

{ "code": "InvalidApiKey", "message": "No API-key provided.", "request_id": "..." }

POLL RESPONSE — SUCCEEDED (video-generation families: t2v / i2v / r2v / videoedit / kf2v / happyhorse)

{
  "request_id": "...",
  "output": {
    "task_id": "...",
    "task_status": "SUCCEEDED",
    "submit_time": "2025-09-29 14:18:52.331",   // UTC+8, YYYY-MM-DD HH:mm:ss.SSS
    "scheduled_time": "2025-09-29 14:18:59.290",
    "end_time": "2025-09-29 14:23:39.407",
    "orig_prompt": "...",                        // = request prompt
    "actual_prompt": "...",                      // only when prompt_extend=true (NOT on wan2.6; present on wan2.7/kf2v/legacy)
    "video_url": "https://...mp4?Expires=..."    // top-level; 24h; MP4 H.264; SUCCEEDED only
  },
  "usage": { /* family-specific — see difference #15 */ }
}

Example usage shapes:

• wan2.7-t2v: { "duration":10, "input_video_duration":0, "output_video_duration":10, "video_count":1, "ratio":"16:9", "SR":720 }
• wan2.6-t2v (legacy): { "duration":10, "size":"1280*720", "input_video_duration":0, "output_video_duration":10, "video_count":1, "SR":720 }
• wan2.7-i2v: { "video_count":1, "duration":10, "input_video_duration":0, "output_video_duration":10, "SR":720 }
• wan2.7-r2v: { "duration":15, "input_video_duration":5, "output_video_duration":10, "video_count":1, "SR":720, "ratio":"16:9" }
• wan2.7-videoedit: { "duration":10.04, "input_video_duration":5.02, "output_video_duration":5.02, "video_count":1, "SR":720 }
• kf2v: { "video_duration":5, "video_count":1, "SR":480 } (2.1: video_ratio":"standard" instead of SR)
• happyhorse-1.0-t2v/r2v: { "duration":5, "input_video_duration":0, "output_video_duration":5, "video_count":1, "SR":720, "ratio":"16:9" }

POLL RESPONSE — FAILED (all families)

{
  "request_id": "...",
  "output": { "task_id": "...", "task_status": "FAILED", "code": "InvalidParameter", "message": "The size does not match ..." }
}

POLL RESPONSE — expired (task_id older than 24h)

{ "request_id": "...", "output": { "task_id": "...", "task_status": "UNKNOWN" } }

---

OUTPUT (poll) field dictionary

field	type	notes
output.task_id	string	valid 24h
output.task_status	string	PENDING / RUNNING / SUCCEEDED / FAILED / CANCELED / UNKNOWN
output.submit_time / scheduled_time / end_time	string	UTC+8, YYYY-MM-DD HH:mm:ss.SSS
output.video_url	string	SUCCEEDED only; 24h; MP4 H.264. Animate: under output.results.video_url
output.results	object	animate-move / animate-mix only; { video_url }
output.orig_prompt	string	echoes request prompt (absent on animate, which has no prompt)
output.actual_prompt	string	rewritten prompt when prompt_extend=true; NOT returned by wan2.6; not on HappyHorse (no prompt_extend)
output.code / output.message	string	FAILED only
usage.duration	float/int	total billed seconds = input_video_duration + output_video_duration (video-generation families)
usage.input_video_duration	int/float	input video seconds (0 for t2v/i2v-no-clip)
usage.output_video_duration	int/float	output seconds = requested duration (where fixed)
usage.video_duration	float/int	LEGACY t2v ≤2.5 / kf2v / animate billed seconds
usage.SR	int	resolution tier number (480/720/1080) — wan2.6+ / wan2.7 / kf2v(2.2) / happyhorse
usage.size	string	legacy wan2.6-t2v only, e.g. 1920×1080
usage.ratio	string	wan2.7-t2v / wan2.7-r2v / happyhorse-t2v/r2v, e.g. 16:9
usage.video_ratio	string	legacy t2v ≤2.5 (resolution) / kf2v-2.1 (“standard”) / animate (service mode: standard|pro)
usage.video_count	int	always 1
request_id	string	trace id

BILLING

• All video families: billed on output video duration in seconds (per-second pricing), resolution-dependent unit price (1080P > 720P > 480P).
• r2v / video continuation / video-edit: duration (= input + output) is the billed amount.
• animate: mode (wan-pro > wan-std price). i2v-flash audio=false ⇒ silent-video price.
• Failed tasks / processing errors are FREE and do not consume free quota.
• Output video_url and task_id expire after 24 hours — download immediately.