Wan 2.7 Video API Guide: Models, Parameters, and Your First Working Request

If you need Wan 2.7 in production, start with the hosted API and design for asynchronous jobs from the first request.

That is the real difference between a clean first integration and a frustrating one. Most early failures are not creative failures. They are setup failures.

Updated for April 27, 2026, this guide covers:

• which official Wan 2.7 API path maps to which job
• what the official docs confirm right now
• the region and polling rules that trip people up
• the minimum payload shape you need to think about
• when to skip code and just use wan27.org

What Is Officially Live Right Now

As of April 27, 2026, the official Alibaba-hosted documentation already confirms the key Wan 2.7 surfaces:

• the Wan 2.7 image generation and editing API reference was updated on April 1, 2026
• the Wan image-to-video API reference for wan2.7-i2v was updated on April 3, 2026
• the official text-to-video guide is live and documents the hosted workflow

If you want first-party references, start with:

• Wan image-to-video API reference
• Wan text-to-video guide
• Wan2.7 image generation and editing API reference
• Wan text-to-video and image-to-video prompt guide

That matters because search results often blur together three different things:

• official hosted APIs
• third-party hosted providers
• open-weight or self-hosted expectations

If your goal is to ship something soon, the hosted official path is the clean answer.

If your goal is self-hosting, that is a separate question and should not be mixed into day-one API setup.

Which Wan API Path Matches Which Job

Use this as the shortest decision table:

One important clarification:

Some third-party quickstarts make first/last frame sound like a separate model family. In the official Alibaba image-to-video docs, first-frame generation and first+last-frame generation both run through wan2.7-i2v. The difference is the media payload, not a separate public model name in that reference.

The Three Rules That Prevent Most Bad Starts

1. Region must match everything

The official docs are explicit here.

Your model availability, endpoint URL, and API key must belong to the same region. Cross-region requests fail.

That means you should decide region before you write helpers, SDK wrappers, or queue workers around the API.

2. Treat video generation as async by default

For Wan 2.7 image-to-video, the official docs say jobs typically take 1 to 5 minutes and the API uses asynchronous invocation:

1. create task
2. receive task_id
3. poll task status
4. fetch the result URL

Do not build around a synchronous mental model and then patch it later.

3. Results do not live forever

The official docs say the task_id and returned result URLs are valid for 24 hours.

That has two practical consequences:

• download outputs promptly
• persist your own metadata instead of assuming the upstream task will remain queryable

If you forget that rule, debugging gets painful fast.

The Minimum Payload You Should Understand

You do not need a huge abstraction layer for the first request. You do need to understand the payload shape.

For official Wan 2.7 image-to-video, the request is built around:

• model
• input
• parameters

A minimal first-frame example looks like this:

{
  "model": "wan2.7-i2v",
  "input": {
    "prompt": "A clean product shot slowly rotates under soft studio light.",
    "media": [
      { "type": "first_frame", "url": "https://your-cdn.com/frame.png" }
    ]
  },
  "parameters": {
    "resolution": "720P",
    "duration": 5,
    "prompt_extend": true,
    "watermark": false
  }
}

For first+last frame, the same model stays in place. You change the media array:

[
  { "type": "first_frame", "url": "https://your-cdn.com/start.png" },
  { "type": "last_frame", "url": "https://your-cdn.com/end.png" }
]

That is a cleaner mental model than inventing a new routing strategy for every mode.

The Parameters That Matter First

For the official Wan 2.7 image-to-video docs, the most useful early parameters are:

• resolution
• duration
• prompt_extend
• watermark
• seed
• negative_prompt

The official docs also state:

• wan2.7-i2v supports 720P and 1080P
• duration ranges from 2 to 15 seconds
• the main prompt can be up to 5,000 characters
• the negative prompt can be up to 500 characters

That does not mean you should use all that prompt space. It means the API will accept it.

For first integrations, shorter and clearer prompts are usually better.

The Prompt Rule That Saves Time

For text-to-video, describe the scene and the motion.

For image-to-video, the image already carries the look, so your prompt should focus on motion and camera behavior.

Alibaba’s official prompt guide reduces it nicely:

• text-to-video: entity + scene + motion
• image-to-video: motion + camera movement

That is why many beginner I2V prompts underperform. They waste tokens re-describing the uploaded image instead of describing what should happen next.

If you want a writing framework before you integrate, read Wan 2.7 Prompt Guide.

The Biggest Integration Mistakes

Mistake 1: Wrong region with the right API key

This is the most boring failure and one of the most common.

Check region first.

Mistake 2: Assuming first/last frame needs a separate public model

In the official image-to-video docs, it does not. The public path shown there is still wan2.7-i2v with different media types.

Mistake 3: Polling too aggressively

If the task typically takes minutes, polling every second does not make the model faster.

It only creates noisy infrastructure and worse failure behavior.

Mistake 4: Forgetting result expiry

If your worker does not store outputs or follow up within the 24-hour window, the task becomes much less useful operationally.

Mistake 5: Starting with the most expensive settings

Do not make your first successful request a long 1080P job unless there is a good reason.

Prove the payload at shorter duration and lower cost first.

A Better First-Request Strategy

For most teams, the safest order is:

1. start with one short official mode
2. make the request work end to end
3. verify polling and result storage
4. test prompt behavior
5. only then scale resolution, duration, or concurrency

That is how you avoid mixing product questions with infrastructure questions.

When You Should Not Use the API Yet

If you are still learning:

• which Wan 2.7 mode you actually need
• what a good prompt looks like
• whether first/last frame or 9-grid fits the use case
• how much iteration the team will do

then the browser workflow is often the better first step.

That is exactly where wan27.org is useful. You can test the practical workflow, see which mode fits, and only move to code when the job shape is clear.

Bottom Line

Wan 2.7 API work gets easier once you simplify the first question.

Do not ask “How do I integrate everything?” first.

Ask:

• which mode do I actually need
• which official doc governs that mode
• what region am I using
• how will I poll and store results

If you only need a browser-first workflow right now, use wan27.org. If you need the broader product overview before you wire anything up, start with What Is Wan 2.7? Complete Guide to Features, API Access, Pricing, and Open-Source Options.