Wan 2.7 is the latest-generation video creation model family from Alibaba Cloud Bailian, delivering cinematic 1080P quality, up to 15-second videos, multi-shot

Wan 2.7 Video Generation

Wan 2.7 is the latest-generation video creation model family from Alibaba Cloud Bailian, delivering cinematic 1080P quality, up to 15-second videos, multi-shot storytelling, and native audio synchronization. AutoRouter wraps the upstream DashScope async task API as a unified /v1/video/generations endpoint, covering four capabilities: text-to-video, image-to-video, reference-to-video, and video editing.

Wan 2.7 uses an async task-based API: submit a request, receive a task_id, poll for status, then download the video from data.result_url.

1. Supported Models

Model	Capability	Required Media Input	Description
`wan2.7-t2v`	Text-to-Video	—	Generate video from text prompt; supports multi-shot storytelling and audio sync
`wan2.7-i2v`	Image-to-Video	1 first-frame image	Generate video starting from a first-frame image
`wan2.7-r2v`	Reference-to-Video	1–9 reference images	Multi-image reference generation maintaining subject consistency; use `[Image N]` in prompt
`wan2.7-videoedit`	Video Editing	1 video + 0–3 reference images	Edit video via text instructions: style transfer, element replacement, local modification

All 4 models share the same endpoint /v1/video/generations, distinguished only by the model field. Response structures are identical across all models.

2. Common Endpoints

2.1 Submit Task

POST /v1/video/generations

Request Headers

Header	Required	Description
`Authorization`	✓	`Bearer sk-YourApiKey`
`Content-Type`	✓	`application/json`

Request Body Fields

Field	Type	Required	Description
`model`	string	✓	Wan 2.7 model name (see table above)
`prompt`	string	✓	Text prompt, max 5000 characters, Chinese and English supported
`resolution`	string	—	Resolution tier: `720P` / `1080P` (default `720P`). Recommended — aligns with the upstream DashScope API field name
`size`	string	—	Alias for `resolution`, kept for backward compatibility. If both are provided, `resolution` takes precedence. New integrations should use `resolution` instead
`duration`	int	—	Video duration in seconds. t2v/i2v: `2~15`, default `5`; r2v with reference video: `2~10`. Ignored by `videoedit` — duration follows input video (max 15s)
`input_reference`	string	i2v only	i2v only: publicly accessible first-frame image URL
`images`	array	videoedit only	videoedit only: media URL array — first element is the source video (MP4/MOV), remaining elements are reference images (optional, max 3)
`metadata`	object	—	Wan 2.7-specific parameters container, see Section 3

The images field is currently in a transition period: before the feature release, pass media via metadata.input.media with explicit type fields; after release, the system automatically infers media type from URL file extension and images can be used directly. Follow the actual API behavior in production.

Response

{
  "id": "task_AbCdEf1234567890XyZw",
  "task_id": "task_AbCdEf1234567890XyZw",
  "object": "video",
  "model": "wan2.7-t2v",
  "status": "queued",
  "progress": 0,
  "created_at": 1778300000
}

id / task_id is the AutoRouter public ID (with task_ prefix), used for subsequent queries. It is not the upstream DashScope task_id — always use the value returned here.

2.2 Query Task Status

GET /v1/video/generations/{task_id}

The response uses AutoRouter's unified {code, message, data} wrapper format. data contains full task details, billing info, channel info, and the raw upstream response.

Response (Queued / NOT_START)

{
  "code": "success",
  "message": "",
  "data": {
    "id": 42,
    "task_id": "task_AbCdEf1234567890XyZw",
    "platform": "17",
    "user_id": 3,
    "group": "default",
    "channel_id": 18,
    "quota": 250000,
    "action": "textGenerate",
    "status": "NOT_START",
    "fail_reason": "",
    "submit_time": 1778300000,
    "start_time": 0,
    "finish_time": 0,
    "created_at": 1778300000,
    "updated_at": 1778300000,
    "progress": "0%",
    "properties": {
      "input": "",
      "upstream_model_name": "wan2.7-t2v",
      "origin_model_name": "wan2.7-t2v"
    },
    "data": {
      "output": {
        "task_id": "xxxxxxxx-xxxx-xxxx-xxxx-xxxxxxxxxxxx",
        "task_status": "PENDING"
      },
      "request_id": "xxxxxxxx-xxxx-xxxx-xxxx-xxxxxxxxxxxx"
    }
  }
}

Response (In Progress / IN_PROGRESS)

{
  "code": "success",
  "message": "",
  "data": {
    "id": 42,
    "task_id": "task_AbCdEf1234567890XyZw",
    "status": "IN_PROGRESS",
    "progress": "40%",
    "submit_time": 1778300000,
    "start_time": 1778300010,
    "finish_time": 0,
    "properties": {
      "upstream_model_name": "wan2.7-t2v",
      "origin_model_name": "wan2.7-t2v"
    },
    "data": {
      "output": { "task_status": "RUNNING" }
    }
  }
}

Response (Completed / SUCCESS)

{
  "code": "success",
  "message": "",
  "data": {
    "id": 42,
    "task_id": "task_AbCdEf1234567890XyZw",
    "platform": "17",
    "user_id": 3,
    "group": "default",
    "channel_id": 18,
    "quota": 250000,
    "action": "textGenerate",
    "status": "SUCCESS",
    "fail_reason": "",
    "result_url": "https://dashscope-result-bj.oss-cn-beijing.aliyuncs.com/.../output.mp4?Expires=...&Signature=...",
    "submit_time": 1778300000,
    "start_time": 1778300010,
    "finish_time": 1778300095,
    "created_at": 1778300000,
    "updated_at": 1778300095,
    "progress": "100%",
    "properties": {
      "input": "",
      "upstream_model_name": "wan2.7-t2v",
      "origin_model_name": "wan2.7-t2v"
    },
    "data": {
      "output": {
        "task_id": "xxxxxxxx-xxxx-xxxx-xxxx-xxxxxxxxxxxx",
        "task_status": "SUCCEEDED",
        "video_url": "https://dashscope-result-bj.oss-cn-beijing.aliyuncs.com/.../output.mp4"
      },
      "usage": {
        "video_duration": 5,
        "video_ratio": "16:9",
        "video_count": 1
      },
      "request_id": "xxxxxxxx-xxxx-xxxx-xxxx-xxxxxxxxxxxx"
    }
  }
}

Response (Failed / FAILURE)

{
  "code": "success",
  "message": "",
  "data": {
    "id": 42,
    "task_id": "task_AbCdEf1234567890XyZw",
    "status": "FAILURE",
    "fail_reason": "DataInspectionFailed: input data may contain inappropriate content",
    "progress": "100%",
    "data": {
      "output": { "task_status": "FAILED" }
    }
  }
}

Key Fields

Field	Type	Description
`code`	string	Response code, `"success"` on success
`message`	string	Response message
`data.id`	int	Internal task auto-increment ID
`data.task_id`	string	AutoRouter public task ID (prefixed with `task_`)
`data.status`	string	Task status: `NOT_START` / `QUEUED` / `IN_PROGRESS` / `SUCCESS` / `FAILURE`
`data.result_url`	string	Video result URL (non-empty only on `SUCCESS`; upstream OSS temporary signed link)
`data.fail_reason`	string	Failure reason (non-empty only on `FAILURE`)
`data.quota`	int	Raw quota deducted for this task
`data.progress`	string	Progress string, e.g. `"40%"` / `"100%"`
`data.submit_time` / `start_time` / `finish_time`	int64	Unix timestamps (seconds) for submit / start / finish
`data.created_at` / `updated_at`	int64	Record creation / last update time
`data.properties.upstream_model_name`	string	Actual model name sent to upstream DashScope
`data.properties.origin_model_name`	string	Model name from the client request
`data.data`	object	Raw upstream DashScope response (contains `output.video_url`, `usage`, `request_id`)

Status Enum

Status	DashScope Status	Description
`NOT_START`	`PENDING`	Created, not yet started
`QUEUED`	`PENDING`	Entered upstream queue
`IN_PROGRESS`	`RUNNING`	Generating
`SUCCESS`	`SUCCEEDED`	Completed; `result_url` contains the video URL
`FAILURE`	`FAILED` / `CANCELED` / `UNKNOWN`	Generation failed; `fail_reason` contains details

The video URL (data.result_url) is an upstream OSS signed link, valid for 24 hours. Download or transfer it to your own storage immediately after retrieval.

3. Wan 2.7-Specific Parameters (metadata)

These fields are passed via the metadata object in the request body and forwarded transparently to upstream DashScope.

3.1 Common Control Parameters

Field	Type	Applicable Models	Description	Default
`ratio`	string	t2v / r2v	Aspect ratio: `16:9` / `9:16` / `1:1` / `4:3` / `3:4`	`16:9`
`seed`	int	All	Random seed, range `0 ~ 2147483647`; auto-generated if not specified	—
`negative_prompt`	string	All	Negative prompt describing content to exclude; max 500 characters	—

i2v does not support ratio: the output aspect ratio automatically follows the input first-frame image.

3.2 Multimedia Input (metadata.input.media)

For r2v and videoedit, pass media assets via the metadata.input.media array:

{
  "metadata": {
    "input": {
      "media": [
        { "type": "reference_image", "url": "https://.../img1.jpg" },
        { "type": "reference_image", "url": "https://.../img2.jpg" },
        { "type": "video", "url": "https://.../video.mp4" }
      ]
    }
  }
}

`type`	Purpose	Count Limit
`reference_image`	Reference image	r2v: 1–9; videoedit: 0–3
`video`	Source video to edit (videoedit only)	Exactly 1 required
`first_frame`	First-frame image (i2v only; can also be passed via top-level `input_reference`)	Exactly 1

r2v prompt convention: reference images in metadata.input.media by [Image 1], [Image 2], etc., in array order. Example: "The woman in [Image 1] wearing a red cheongsam gracefully unfolds the fan from [Image 2]."

3.3 Media Asset Constraints

Type	Formats	Resolution	File Size	Other
Image (i2v first frame)	JPEG / JPG / PNG / WEBP	240–8000 px per side	≤ 10 MB	Aspect ratio 1:8 ~ 8:1
Image (r2v reference)	JPEG / JPG / PNG / WEBP	Short side ≥ 300 px	≤ 10 MB	Prefer sharp, uncompressed images
Image (videoedit reference)	JPEG / JPG / PNG / BMP / WEBP	240–8000 px per side	≤ 20 MB	Aspect ratio 1:8 ~ 8:1
Video (videoedit source)	MP4 / MOV (H.264 recommended)	240–4096 px per side	≤ 100 MB	Duration 2–10s, aspect ratio 1:8 ~ 8:1

4. Full Request Examples

# Step 1: Submit task
curl -X POST https://autorouter.top/v1/video/generations \
  -H "Authorization: Bearer sk-YourApiKey" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "wan2.7-t2v",
    "prompt": "A small cat running under moonlight, fur gently flowing in the wind, cinematic slow motion",
    "resolution": "720P",
    "duration": 5,
    "metadata": {
      "ratio": "16:9",
      "seed": 42
    }
  }'

# Response:
# {
#   "id": "task_AbCdEf1234567890XyZw",
#   "task_id": "task_AbCdEf1234567890XyZw",
#   "status": "queued", ...
# }

# Step 2: Poll (recommended interval: 10~15 seconds)
curl https://autorouter.top/v1/video/generations/task_AbCdEf1234567890XyZw \
  -H "Authorization: Bearer sk-YourApiKey"

# Step 3: When data.status == "SUCCESS", download from data.result_url
curl -o output.mp4 "https://dashscope-result-bj.oss-cn-beijing.aliyuncs.com/.../output.mp4?Expires=..."

# i2v: pass first-frame image via input_reference
curl -X POST https://autorouter.top/v1/video/generations \
  -H "Authorization: Bearer sk-YourApiKey" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "wan2.7-i2v",
    "prompt": "Realistic style, a small black cat curiously looks up at the sky, camera slowly zooms in",
    "resolution": "720P",
    "duration": 5,
    "input_reference": "https://wanx.alicdn.com/material/20250318/first_frame.png"
  }'

The output aspect ratio for i2v automatically follows the input_reference image. metadata.ratio is not supported for i2v.

# r2v: pass reference images via metadata.input.media
# Reference them in the prompt with [Image 1], [Image 2], etc.
curl -X POST https://autorouter.top/v1/video/generations \
  -H "Authorization: Bearer sk-YourApiKey" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "wan2.7-r2v",
    "prompt": "3D cartoon adventure movie style, chibi character with refined textures, [Image 1] character running joyfully",
    "resolution": "720P",
    "duration": 5,
    "metadata": {
      "ratio": "16:9",
      "input": {
        "media": [
          {
            "type": "reference_image",
            "url": "https://help-static-aliyun-doc.aliyuncs.com/file-manage-files/zh-CN/20260403/wgjaxy/banana_storyboard_00000020.png"
          }
        ]
      }
    }
  }'

# videoedit: source video + reference image via images array
# images[0] = source video (MP4/MOV), images[1+] = reference images
curl -X POST https://autorouter.top/v1/video/generations \
  -H "Authorization: Bearer sk-YourApiKey" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "wan2.7-videoedit",
    "prompt": "Replace the outfit of the girl in the video with the clothing shown in the reference image",
    "resolution": "720P",
    "images": [
      "https://help-static-aliyun-doc.aliyuncs.com/file-manage-files/zh-CN/20260403/nlspwm/T2VA_22.mp4",
      "https://help-static-aliyun-doc.aliyuncs.com/file-manage-files/zh-CN/20260402/fwjpqf/wan2.7-videoedit-change-clothes.png"
    ]
  }'

No duration needed for videoedit — output duration follows the source video (max 15s). The system infers media type from file extension: MP4/MOV → source video, PNG/JPG/WEBP → reference image.

5. OpenAI-Compatible Endpoints (Optional)

In addition to /v1/video/generations, Wan 2.7 also supports OpenAI-style /v1/videos endpoints. Both are fully equivalent — choose based on your client integration preference.

5.1 Submit (OpenAI-Compatible)

POST /v1/videos

Request body fields are identical to /v1/video/generations.

5.2 Query (OpenAI-Compatible)

GET /v1/videos/{task_id}

Returns OpenAI standard flat format:

{
  "id": "task_AbCdEf1234567890XyZw",
  "object": "video",
  "model": "wan2.7-t2v",
  "status": "completed",
  "progress": 100,
  "created_at": 1778300000,
  "completed_at": 1778300095,
  "metadata": {
    "url": "https://dashscope-result-bj.oss-cn-beijing.aliyuncs.com/.../output.mp4?Expires=...&Signature=..."
  }
}

OpenAI Status	Unified Format Status
`queued`	`NOT_START` / `QUEUED`
`in_progress`	`IN_PROGRESS`
`completed`	`SUCCESS`
`failed`	`FAILURE`

6. Billing

Wan 2.7 is billed by video seconds × resolution multiplier:

Resolution	Unit Price	Example (5s)
`720P`	$0.10 / second	5 × $0.10 = $0.50
`1080P`	$0.15 / second	5 × $0.15 = $0.75

Quota calculation formula:

quota = modelPrice × QuotaPerUnit × groupRatio × seconds × resolutionRatio

QuotaPerUnit    = 500,000
modelPrice      = 0.10  (all wan2.7 models)
groupRatio      = 1.0   (production default group)
resolutionRatio = 720P → 1.0, 1080P → 1.5

Examples:
  720P  × 5s  = 0.10 × 500,000 × 1 × 5 × 1.0 = 250,000 quota
  1080P × 10s = 0.10 × 500,000 × 1 × 10 × 1.5 = 750,000 quota

On task FAILURE, AutoRouter automatically refunds quota to your account (logged as type=6). No manual action required.

7. Polling Guide

Item	Recommendation
Poll interval	Every 10–15 seconds; do not poll more often than every 5 seconds
Terminal state	`data.status == "SUCCESS"` or `data.status == "FAILURE"`
Total timeout	t2v / i2v / r2v: 5 minutes; videoedit: 8–10 minutes
Typical duration	t2v / i2v / r2v: 30–100s; videoedit: 3–5 minutes
Video URL validity	24 hours; download or transfer immediately

Simple Polling Script (Bash)

#!/bin/bash
TID="task_AbCdEf1234567890XyZw"
KEY="sk-YourApiKey"
while true; do
  RESP=$(curl -sS "https://autorouter.top/v1/video/generations/$TID" \
    -H "Authorization: Bearer $KEY")
  echo "$RESP"
  STATUS=$(echo "$RESP" | python3 -c \
    "import sys,json;print(json.load(sys.stdin)['data'].get('status',''))")
  case "$STATUS" in
    SUCCESS|FAILURE) break ;;
  esac
  sleep 15
done

8. Error Handling

8.1 HTTP 400 — Invalid Parameters (rejected before billing)

Scenario	Response
Missing `prompt`	`{"code":"invalid_request","message":"prompt is required","data":null}`
Unknown model name	`{"code":"invalid_request","message":"unknown model: ..."}`
Empty `images` array	`{"code":"invalid_request","message":"images is required for videoedit"}`

8.2 HTTP 401 / 403 — Auth Error

401 Unauthorized: API Key invalid or expired
403 Forbidden: API Key has no access to this model

8.3 HTTP 402 — Insufficient Balance

Returns insufficient user quota. Recharge at the AutoRouter console.

8.4 Task FAILURE Status

Common Reason	Recommendation
`DataInspectionFailed`	Upstream content moderation triggered. Adjust prompt or media. Note: some official Alibaba example assets may also trigger this — it is upstream behavior, not an AutoRouter restriction.
i2v missing `input_reference`	First-frame image is required for image-to-video
r2v with zero reference images	Provide at least 1 `reference_image`
videoedit missing source video	`images[0]` must be an MP4/MOV video link
Inaccessible media URL	Ensure URLs are publicly accessible and not expired
Media asset out of spec	See Section 3.3 for constraints

9. FAQ

What is the difference between Wan 2.7 and HappyHorse?

Both are Alibaba Cloud video generation models accessible via AutoRouter's unified API:

Wan 2.7: General-purpose cinematic video generation. Excels at detailed scenes, multi-shot storytelling, and up to 15-second clips. Best for creative content production.
HappyHorse: Focused on character motion and fluid movement. Better suited for character-driven short video scenarios.

Choose based on your content needs — both share the same API interface.

Do all 4 models use the same endpoint?

Yes. All models use POST /v1/video/generations, differentiated only by the model field. Response structures are identical — no per-model client-side handling required.

What is the difference between /v1/video/generations and /v1/videos?

AutoRouter provides both endpoints simultaneously:

/v1/video/generations (recommended): Query returns {code, message, data} wrapper with full billing, channel, and raw upstream data.
/v1/videos (OpenAI-compatible): Query returns OpenAI standard flat structure for direct OpenAI SDK integration.

Both point to the same backend task — the same task_id works across both endpoints.

How does videoedit distinguish video from images in the images array?

The system infers media type from the URL file extension:

.mp4 / .mov → source video to edit
.jpg / .jpeg / .png / .webp / .bmp → reference image

images[0] must be the video link; subsequent elements are reference images (max 3). Before feature release, use metadata.input.media with explicit type fields instead.

What duration values are supported?

wan2.7-t2v / wan2.7-i2v: integers 2–15, default 5
wan2.7-r2v: 2–10 when reference video is included; 2–15 for image-only reference, default 5
wan2.7-videoedit: field is ignored; output duration follows source video (max 15s)

How do [Image N] references work in r2v prompts?

[Image N] maps strictly to the Nth reference_image entry in the metadata.input.media array. The first reference_image = [Image 1], the second = [Image 2], and so on.

When does the video result URL expire?

The upstream OSS signed URL typically expires 24 hours after generation. Best practice: download or transfer to your own storage immediately after the task succeeds.

How do I handle DataInspectionFailed errors?

This is Alibaba's upstream content moderation system at work. Note that some official Alibaba demo assets can also trigger this error — it is an upstream issue, not an AutoRouter restriction. Swap out the media or adjust the prompt and retry. AutoRouter automatically refunds quota on task failure.

Does the same seed reproduce identical results?

A fixed seed improves reproducibility but does not guarantee identical output every time due to the model's inherent stochasticity.

How do I choose the right model?

Choose by input type:

Text description only → wan2.7-t2v
One starting image, want the action to flow from it → wan2.7-i2v
Multiple reference images (character, props, scene), want them fused into one video → wan2.7-r2v
Existing video to edit (style transfer, element replacement, local modification) → wan2.7-videoedit

Wan 2.7 Video Generation

On this page