Unified image-to-video
One I2V endpoint handles a first frame, first and last frames, video continuation, or continuation with a controlled ending.
Wan 2.7 model guide
Wan 2.7 is Alibaba Cloud's production-focused hosted family for audiovisual generation, unifying first-frame, first/last-frame and continuation tasks while adding subject-level voice references and instruction-based video editing.
2–15sOutput duration
1080PMaximum resolution
30 fpsFixed frame rate
A/VSynchronized audio
InputsText · image · audio · video reference
Model familyWan 2.7
Output720P / 1080P · 30 fps · MP4
What this model family is
Wan 2.7 is delivered through task-specific Model Studio endpoints rather than one public download. Its April and June 2026 snapshots cover T2V, a unified multimodal I2V workflow, multi-subject reference generation and video editing; a separate image family adds 2K/4K generation, editing and sequential outputs.
Production capabilities
The useful distinction is not a benchmark score, but which input assets and production controls each endpoint accepts.
Subject-level voice reference
R2V can combine multiple images and videos and attach a reference voice to each subject in the media array.
Instruction video editing
The video-edit model performs local or global changes and can transfer appearance, motion, effects or camera movement.
Companion image family
Wan 2.7 Image supports generation, editing and consistent sequences; Pro reaches 4K for text-to-image.
Models and endpoints
Rolling aliases can be updated by the provider. Dated IDs lock a documented snapshot and are safer when a production workflow must remain reproducible.
| Model ID | Task | Inputs | Output | Use it when |
|---|---|---|---|---|
wan2.7-t2vwan2.7-t2v-2026-06-12wan2.7-t2v-2026-04-25 | Text to video | Text + optional audio | 720P / 1080P · 2–15s · 30 fps · MP4 | Multi-shot audiovisual generation |
wan2.7-i2vwan2.7-i2v-2026-04-25 | Unified image to video | Text + image + optional audio/video | 720P / 1080P · 2–15s · 30 fps · MP4 | First frame, first/last frame and continuation |
wan2.7-r2vwan2.7-r2v-2026-06-12 | Reference to video | Text + images + videos + reference voices | 720P / 1080P · 2–10s · 30 fps · MP4 | Multiple subjects with per-subject voice |
wan2.7-videoedit | Instruction video editing | Text + video + optional reference image | 720P / 1080P · 2–10s · 30 fps · MP4 | Local/global edits and motion or style transfer |
wan2.7-imagewan2.7-image-pro | Image generation and editing | Text + up to four reference images | PNG · ≤ 2048² / 4096² (Pro T2I) | Fast 2K model or 4K Pro text-to-image |
How the API works
Video jobs are asynchronous: submit once, poll the task ID, then save the returned asset before its temporary URL expires.
Create a task
Send the model, prompt, media and generation parameters to the regional synthesis endpoint.
Poll the result
Read the task instead of submitting duplicate jobs while it moves from pending to running.
Store the video
Download the MP4 after success and copy it to durable storage for production use.
POST /api/v1/services/aigc/
video-generation/video-synthesis
X-DashScope-Async: enable
GET /api/v1/tasks/{task_id}The API key, workspace endpoint and model must belong to the same region. Task IDs and result URLs remain available for 24 hours.
Estimate API cost
The calculator uses Singapore international list prices. T2V and I2V bill output only; R2V bills up to five seconds of input video, and VideoEdit bills both input and output duration.
Estimated list-price cost
Output seconds × resolution rate
USD · Singapore international deploymentList-price snapshot for August 28, 2026. Promotions can differ. Eligible Singapore models include a 50-second quota valid for 90 days under Model Studio's quota rules.
What changed from Wan 2.6
Wan 2.7 keeps the 1080P, 30 fps and 15-second ceiling for general generation, but changes how assets, aspect ratios and shot instructions are represented in the API.
| Capability | Wan 2.6 | Wan 2.7 |
|---|---|---|
| I2V workflow | First-frame endpoint | First frame, first/last frame and continuation in one endpoint |
| Media schema | reference_urls and character labels | Typed media array with Image N / Video N references |
| Resolution control | size such as 1280*720 | resolution tier plus ratio |
| Shot control | shot_type=multi | Natural-language shot plan or timestamps in the prompt |
| Editing and voice | No dedicated editor; voice follows reference | VideoEdit plus reference_voice per subject |
Continue exploring
Move between the model generations and the practical tools built around the Wan workflow.
Frequently asked questions
Is Wan 2.7 open source?
The model IDs documented here are hosted Alibaba Cloud Model Studio services and require a regional API key. They are not the same delivery as an open-weight Wan release.
What is the difference between an alias and a dated model ID?
A rolling alias such as wan2.7-t2v can move to a newer provider version. A dated ID such as wan2.7-t2v-2026-06-12 pins a specific snapshot for reproducible production.
Can Wan 2.7 generate synchronized audio?
Yes. It can automatically dub a video or use supported custom audio. R2V can also associate a reference voice with an individual subject.
Does Wan 2.7 support first and last frames?
Yes. The unified I2V endpoint supports first-frame, first/last-frame, video continuation, and continuation with a final frame.
Is the Wan 2.7 API free?
It is not an anonymous free API. Eligible Model Studio accounts can receive a 50-second quota for supported Singapore endpoints, subject to the 90-day quota rules.