# Connect with AI

https://video-doc.1route.dev/en/docs/ai



## Copy these instructions to your AI [#copy-these-instructions-to-your-ai]

Add your tool name, operating system or workflow. The AI should ask only for missing details.

Help me integrate 1Route WAN3 video generation with my tool or project. I am a beginner.
Read https://video-doc.1route.dev/en/llms-full.txt first.
Ask only for missing details: my tool, whether it supports asynchronous video tasks, text/image/video input, and spending limit. A chat-only client may need an integration.
API base: https://video.1route.dev. Create a video-group API key at https://video.1route.dev/keys/; do not assume an image-service key works.
Load WAN3_API_KEY from a private local environment file or protected input. Never print it or put a shared key in a public frontend. Preserve existing configuration.
Verify GET /v1/models before choosing an available WAN3 model. For the first paid test propose 2 seconds at 480p and obtain confirmation of the price and content.
POST /v1/videos accepts JSON only; send seconds as a string. Use reference_images:[{url:"HTTPS_URL"}] for images, reference_videos:[{url:"HTTPS_URL",duration:5}] for video. Validate accessible URLs; do not upload private material to arbitrary hosts.
Output is 2–30 whole seconds. Video references: up to 5, 1–15 whole seconds each, total <=15; output + reference seconds <=30. Image models reject video references. Do not invent rounding for fractional reference durations.
Image price = per-second rate × output seconds; video price = rate × (output + unique reference video seconds). Use https://video.1route.dev/pricing/ and account/group pricing. Quote units accurately.
Submit once; immediately save the full returned id and X-Oneapi-Request-Id. Poll GET /v1/videos/{id} every 15 seconds. queued/in_progress continue waiting; progress alone is not completion.
Read completed.metadata.url and immediately save the file locally. The link lasts up to 5 minutes and the cached file lasts 10 minutes after becoming ready. Use the returned URL verbatim; do not forward the API key to arbitrary download hosts.
If creation times out or returns an ambiguous error, do not automatically POST again. Preserve the request information and inspect existing task/usage records. Retry transient GET errors with backoff. Resume an existing task after restarting the client; never regenerate just because polling stopped. No cancellation or idempotent creation is promised.
Provide a complete create/save-id/poll/download flow and beginner instructions. Do not execute a paid test before I confirm.

## What you need [#what-you-need]

1. Your tool or environment.
2. A description of the video and any reference materials.
3. A spending limit for the first test.

Enter the key in your trusted tool configuration or a private local file. Keep it out of public code, group chats and screenshots.

## Can the AI execute the request? [#can-the-ai-execute-the-request]

A chat-only AI can explain and write code. Executing requires code execution or external-tool support. A client that only supports chat completion is not automatically a video client.

A complete integration lists available models, creates once, saves the task ID, polls that same ID, saves the MP4 and checks usage.

For an existing task, tell the AI to query it without creating another. This site documents the video API and does not provide a video MCP endpoint.